1 /* |
|
2 * Copyright 2003-2008 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 #include "incls/_precompiled.incl" |
|
26 #include "incls/_assembler_x86_64.cpp.incl" |
|
27 |
|
28 // Implementation of AddressLiteral |
|
29 |
|
30 AddressLiteral::AddressLiteral(address target, relocInfo::relocType rtype) { |
|
31 _is_lval = false; |
|
32 _target = target; |
|
33 switch (rtype) { |
|
34 case relocInfo::oop_type: |
|
35 // Oops are a special case. Normally they would be their own section |
|
36 // but in cases like icBuffer they are literals in the code stream that |
|
37 // we don't have a section for. We use none so that we get a literal address |
|
38 // which is always patchable. |
|
39 break; |
|
40 case relocInfo::external_word_type: |
|
41 _rspec = external_word_Relocation::spec(target); |
|
42 break; |
|
43 case relocInfo::internal_word_type: |
|
44 _rspec = internal_word_Relocation::spec(target); |
|
45 break; |
|
46 case relocInfo::opt_virtual_call_type: |
|
47 _rspec = opt_virtual_call_Relocation::spec(); |
|
48 break; |
|
49 case relocInfo::static_call_type: |
|
50 _rspec = static_call_Relocation::spec(); |
|
51 break; |
|
52 case relocInfo::runtime_call_type: |
|
53 _rspec = runtime_call_Relocation::spec(); |
|
54 break; |
|
55 case relocInfo::none: |
|
56 break; |
|
57 default: |
|
58 ShouldNotReachHere(); |
|
59 break; |
|
60 } |
|
61 } |
|
62 |
|
63 // Implementation of Address |
|
64 |
|
65 Address Address::make_array(ArrayAddress adr) { |
|
66 #ifdef _LP64 |
|
67 // Not implementable on 64bit machines |
|
68 // Should have been handled higher up the call chain. |
|
69 ShouldNotReachHere(); |
|
70 return Address(); |
|
71 #else |
|
72 AddressLiteral base = adr.base(); |
|
73 Address index = adr.index(); |
|
74 assert(index._disp == 0, "must not have disp"); // maybe it can? |
|
75 Address array(index._base, index._index, index._scale, (intptr_t) base.target()); |
|
76 array._rspec = base._rspec; |
|
77 return array; |
|
78 #endif // _LP64 |
|
79 } |
|
80 |
|
81 // exceedingly dangerous constructor |
|
82 Address::Address(int disp, address loc, relocInfo::relocType rtype) { |
|
83 _base = noreg; |
|
84 _index = noreg; |
|
85 _scale = no_scale; |
|
86 _disp = disp; |
|
87 switch (rtype) { |
|
88 case relocInfo::external_word_type: |
|
89 _rspec = external_word_Relocation::spec(loc); |
|
90 break; |
|
91 case relocInfo::internal_word_type: |
|
92 _rspec = internal_word_Relocation::spec(loc); |
|
93 break; |
|
94 case relocInfo::runtime_call_type: |
|
95 // HMM |
|
96 _rspec = runtime_call_Relocation::spec(); |
|
97 break; |
|
98 case relocInfo::none: |
|
99 break; |
|
100 default: |
|
101 ShouldNotReachHere(); |
|
102 } |
|
103 } |
|
104 |
|
105 // Convert the raw encoding form into the form expected by the constructor for |
|
106 // Address. An index of 4 (rsp) corresponds to having no index, so convert |
|
107 // that to noreg for the Address constructor. |
|
108 Address Address::make_raw(int base, int index, int scale, int disp) { |
|
109 bool valid_index = index != rsp->encoding(); |
|
110 if (valid_index) { |
|
111 Address madr(as_Register(base), as_Register(index), (Address::ScaleFactor)scale, in_ByteSize(disp)); |
|
112 return madr; |
|
113 } else { |
|
114 Address madr(as_Register(base), noreg, Address::no_scale, in_ByteSize(disp)); |
|
115 return madr; |
|
116 } |
|
117 } |
|
118 |
|
119 |
|
120 // Implementation of Assembler |
|
121 int AbstractAssembler::code_fill_byte() { |
|
122 return (u_char)'\xF4'; // hlt |
|
123 } |
|
124 |
|
125 // This should only be used by 64bit instructions that can use rip-relative |
|
126 // it cannot be used by instructions that want an immediate value. |
|
127 |
|
128 bool Assembler::reachable(AddressLiteral adr) { |
|
129 int64_t disp; |
|
130 |
|
131 // None will force a 64bit literal to the code stream. Likely a placeholder |
|
132 // for something that will be patched later and we need to certain it will |
|
133 // always be reachable. |
|
134 if (adr.reloc() == relocInfo::none) { |
|
135 return false; |
|
136 } |
|
137 if (adr.reloc() == relocInfo::internal_word_type) { |
|
138 // This should be rip relative and easily reachable. |
|
139 return true; |
|
140 } |
|
141 if (adr.reloc() != relocInfo::external_word_type && |
|
142 adr.reloc() != relocInfo::runtime_call_type ) { |
|
143 return false; |
|
144 } |
|
145 |
|
146 // Stress the correction code |
|
147 if (ForceUnreachable) { |
|
148 // Must be runtimecall reloc, see if it is in the codecache |
|
149 // Flipping stuff in the codecache to be unreachable causes issues |
|
150 // with things like inline caches where the additional instructions |
|
151 // are not handled. |
|
152 if (CodeCache::find_blob(adr._target) == NULL) { |
|
153 return false; |
|
154 } |
|
155 } |
|
156 // For external_word_type/runtime_call_type if it is reachable from where we |
|
157 // are now (possibly a temp buffer) and where we might end up |
|
158 // anywhere in the codeCache then we are always reachable. |
|
159 // This would have to change if we ever save/restore shared code |
|
160 // to be more pessimistic. |
|
161 |
|
162 disp = (int64_t)adr._target - ((int64_t)CodeCache::low_bound() + sizeof(int)); |
|
163 if (!is_simm32(disp)) return false; |
|
164 disp = (int64_t)adr._target - ((int64_t)CodeCache::high_bound() + sizeof(int)); |
|
165 if (!is_simm32(disp)) return false; |
|
166 |
|
167 disp = (int64_t)adr._target - ((int64_t)_code_pos + sizeof(int)); |
|
168 |
|
169 // Because rip relative is a disp + address_of_next_instruction and we |
|
170 // don't know the value of address_of_next_instruction we apply a fudge factor |
|
171 // to make sure we will be ok no matter the size of the instruction we get placed into. |
|
172 // We don't have to fudge the checks above here because they are already worst case. |
|
173 |
|
174 // 12 == override/rex byte, opcode byte, rm byte, sib byte, a 4-byte disp , 4-byte literal |
|
175 // + 4 because better safe than sorry. |
|
176 const int fudge = 12 + 4; |
|
177 if (disp < 0) { |
|
178 disp -= fudge; |
|
179 } else { |
|
180 disp += fudge; |
|
181 } |
|
182 return is_simm32(disp); |
|
183 } |
|
184 |
|
185 |
|
186 // make this go away eventually |
|
187 void Assembler::emit_data(jint data, |
|
188 relocInfo::relocType rtype, |
|
189 int format) { |
|
190 if (rtype == relocInfo::none) { |
|
191 emit_long(data); |
|
192 } else { |
|
193 emit_data(data, Relocation::spec_simple(rtype), format); |
|
194 } |
|
195 } |
|
196 |
|
197 void Assembler::emit_data(jint data, |
|
198 RelocationHolder const& rspec, |
|
199 int format) { |
|
200 assert(imm64_operand == 0, "default format must be imm64 in this file"); |
|
201 assert(imm64_operand != format, "must not be imm64"); |
|
202 assert(inst_mark() != NULL, "must be inside InstructionMark"); |
|
203 if (rspec.type() != relocInfo::none) { |
|
204 #ifdef ASSERT |
|
205 check_relocation(rspec, format); |
|
206 #endif |
|
207 // Do not use AbstractAssembler::relocate, which is not intended for |
|
208 // embedded words. Instead, relocate to the enclosing instruction. |
|
209 |
|
210 // hack. call32 is too wide for mask so use disp32 |
|
211 if (format == call32_operand) |
|
212 code_section()->relocate(inst_mark(), rspec, disp32_operand); |
|
213 else |
|
214 code_section()->relocate(inst_mark(), rspec, format); |
|
215 } |
|
216 emit_long(data); |
|
217 } |
|
218 |
|
219 void Assembler::emit_data64(jlong data, |
|
220 relocInfo::relocType rtype, |
|
221 int format) { |
|
222 if (rtype == relocInfo::none) { |
|
223 emit_long64(data); |
|
224 } else { |
|
225 emit_data64(data, Relocation::spec_simple(rtype), format); |
|
226 } |
|
227 } |
|
228 |
|
229 void Assembler::emit_data64(jlong data, |
|
230 RelocationHolder const& rspec, |
|
231 int format) { |
|
232 assert(imm64_operand == 0, "default format must be imm64 in this file"); |
|
233 assert(imm64_operand == format, "must be imm64"); |
|
234 assert(inst_mark() != NULL, "must be inside InstructionMark"); |
|
235 // Do not use AbstractAssembler::relocate, which is not intended for |
|
236 // embedded words. Instead, relocate to the enclosing instruction. |
|
237 code_section()->relocate(inst_mark(), rspec, format); |
|
238 #ifdef ASSERT |
|
239 check_relocation(rspec, format); |
|
240 #endif |
|
241 emit_long64(data); |
|
242 } |
|
243 |
|
244 void Assembler::emit_arith_b(int op1, int op2, Register dst, int imm8) { |
|
245 assert(isByte(op1) && isByte(op2), "wrong opcode"); |
|
246 assert(isByte(imm8), "not a byte"); |
|
247 assert((op1 & 0x01) == 0, "should be 8bit operation"); |
|
248 int dstenc = dst->encoding(); |
|
249 if (dstenc >= 8) { |
|
250 dstenc -= 8; |
|
251 } |
|
252 emit_byte(op1); |
|
253 emit_byte(op2 | dstenc); |
|
254 emit_byte(imm8); |
|
255 } |
|
256 |
|
257 void Assembler::emit_arith(int op1, int op2, Register dst, int imm32) { |
|
258 assert(isByte(op1) && isByte(op2), "wrong opcode"); |
|
259 assert((op1 & 0x01) == 1, "should be 32bit operation"); |
|
260 assert((op1 & 0x02) == 0, "sign-extension bit should not be set"); |
|
261 int dstenc = dst->encoding(); |
|
262 if (dstenc >= 8) { |
|
263 dstenc -= 8; |
|
264 } |
|
265 if (is8bit(imm32)) { |
|
266 emit_byte(op1 | 0x02); // set sign bit |
|
267 emit_byte(op2 | dstenc); |
|
268 emit_byte(imm32 & 0xFF); |
|
269 } else { |
|
270 emit_byte(op1); |
|
271 emit_byte(op2 | dstenc); |
|
272 emit_long(imm32); |
|
273 } |
|
274 } |
|
275 |
|
276 // immediate-to-memory forms |
|
277 void Assembler::emit_arith_operand(int op1, |
|
278 Register rm, Address adr, |
|
279 int imm32) { |
|
280 assert((op1 & 0x01) == 1, "should be 32bit operation"); |
|
281 assert((op1 & 0x02) == 0, "sign-extension bit should not be set"); |
|
282 if (is8bit(imm32)) { |
|
283 emit_byte(op1 | 0x02); // set sign bit |
|
284 emit_operand(rm, adr, 1); |
|
285 emit_byte(imm32 & 0xFF); |
|
286 } else { |
|
287 emit_byte(op1); |
|
288 emit_operand(rm, adr, 4); |
|
289 emit_long(imm32); |
|
290 } |
|
291 } |
|
292 |
|
293 |
|
294 void Assembler::emit_arith(int op1, int op2, Register dst, Register src) { |
|
295 assert(isByte(op1) && isByte(op2), "wrong opcode"); |
|
296 int dstenc = dst->encoding(); |
|
297 int srcenc = src->encoding(); |
|
298 if (dstenc >= 8) { |
|
299 dstenc -= 8; |
|
300 } |
|
301 if (srcenc >= 8) { |
|
302 srcenc -= 8; |
|
303 } |
|
304 emit_byte(op1); |
|
305 emit_byte(op2 | dstenc << 3 | srcenc); |
|
306 } |
|
307 |
|
308 void Assembler::emit_operand(Register reg, Register base, Register index, |
|
309 Address::ScaleFactor scale, int disp, |
|
310 RelocationHolder const& rspec, |
|
311 int rip_relative_correction) { |
|
312 relocInfo::relocType rtype = (relocInfo::relocType) rspec.type(); |
|
313 int regenc = reg->encoding(); |
|
314 if (regenc >= 8) { |
|
315 regenc -= 8; |
|
316 } |
|
317 if (base->is_valid()) { |
|
318 if (index->is_valid()) { |
|
319 assert(scale != Address::no_scale, "inconsistent address"); |
|
320 int indexenc = index->encoding(); |
|
321 if (indexenc >= 8) { |
|
322 indexenc -= 8; |
|
323 } |
|
324 int baseenc = base->encoding(); |
|
325 if (baseenc >= 8) { |
|
326 baseenc -= 8; |
|
327 } |
|
328 // [base + index*scale + disp] |
|
329 if (disp == 0 && rtype == relocInfo::none && |
|
330 base != rbp && base != r13) { |
|
331 // [base + index*scale] |
|
332 // [00 reg 100][ss index base] |
|
333 assert(index != rsp, "illegal addressing mode"); |
|
334 emit_byte(0x04 | regenc << 3); |
|
335 emit_byte(scale << 6 | indexenc << 3 | baseenc); |
|
336 } else if (is8bit(disp) && rtype == relocInfo::none) { |
|
337 // [base + index*scale + imm8] |
|
338 // [01 reg 100][ss index base] imm8 |
|
339 assert(index != rsp, "illegal addressing mode"); |
|
340 emit_byte(0x44 | regenc << 3); |
|
341 emit_byte(scale << 6 | indexenc << 3 | baseenc); |
|
342 emit_byte(disp & 0xFF); |
|
343 } else { |
|
344 // [base + index*scale + disp32] |
|
345 // [10 reg 100][ss index base] disp32 |
|
346 assert(index != rsp, "illegal addressing mode"); |
|
347 emit_byte(0x84 | regenc << 3); |
|
348 emit_byte(scale << 6 | indexenc << 3 | baseenc); |
|
349 emit_data(disp, rspec, disp32_operand); |
|
350 } |
|
351 } else if (base == rsp || base == r12) { |
|
352 // [rsp + disp] |
|
353 if (disp == 0 && rtype == relocInfo::none) { |
|
354 // [rsp] |
|
355 // [00 reg 100][00 100 100] |
|
356 emit_byte(0x04 | regenc << 3); |
|
357 emit_byte(0x24); |
|
358 } else if (is8bit(disp) && rtype == relocInfo::none) { |
|
359 // [rsp + imm8] |
|
360 // [01 reg 100][00 100 100] disp8 |
|
361 emit_byte(0x44 | regenc << 3); |
|
362 emit_byte(0x24); |
|
363 emit_byte(disp & 0xFF); |
|
364 } else { |
|
365 // [rsp + imm32] |
|
366 // [10 reg 100][00 100 100] disp32 |
|
367 emit_byte(0x84 | regenc << 3); |
|
368 emit_byte(0x24); |
|
369 emit_data(disp, rspec, disp32_operand); |
|
370 } |
|
371 } else { |
|
372 // [base + disp] |
|
373 assert(base != rsp && base != r12, "illegal addressing mode"); |
|
374 int baseenc = base->encoding(); |
|
375 if (baseenc >= 8) { |
|
376 baseenc -= 8; |
|
377 } |
|
378 if (disp == 0 && rtype == relocInfo::none && |
|
379 base != rbp && base != r13) { |
|
380 // [base] |
|
381 // [00 reg base] |
|
382 emit_byte(0x00 | regenc << 3 | baseenc); |
|
383 } else if (is8bit(disp) && rtype == relocInfo::none) { |
|
384 // [base + disp8] |
|
385 // [01 reg base] disp8 |
|
386 emit_byte(0x40 | regenc << 3 | baseenc); |
|
387 emit_byte(disp & 0xFF); |
|
388 } else { |
|
389 // [base + disp32] |
|
390 // [10 reg base] disp32 |
|
391 emit_byte(0x80 | regenc << 3 | baseenc); |
|
392 emit_data(disp, rspec, disp32_operand); |
|
393 } |
|
394 } |
|
395 } else { |
|
396 if (index->is_valid()) { |
|
397 assert(scale != Address::no_scale, "inconsistent address"); |
|
398 int indexenc = index->encoding(); |
|
399 if (indexenc >= 8) { |
|
400 indexenc -= 8; |
|
401 } |
|
402 // [index*scale + disp] |
|
403 // [00 reg 100][ss index 101] disp32 |
|
404 assert(index != rsp, "illegal addressing mode"); |
|
405 emit_byte(0x04 | regenc << 3); |
|
406 emit_byte(scale << 6 | indexenc << 3 | 0x05); |
|
407 emit_data(disp, rspec, disp32_operand); |
|
408 #ifdef _LP64 |
|
409 } else if (rtype != relocInfo::none ) { |
|
410 // [disp] RIP-RELATIVE |
|
411 // [00 000 101] disp32 |
|
412 |
|
413 emit_byte(0x05 | regenc << 3); |
|
414 // Note that the RIP-rel. correction applies to the generated |
|
415 // disp field, but _not_ to the target address in the rspec. |
|
416 |
|
417 // disp was created by converting the target address minus the pc |
|
418 // at the start of the instruction. That needs more correction here. |
|
419 // intptr_t disp = target - next_ip; |
|
420 assert(inst_mark() != NULL, "must be inside InstructionMark"); |
|
421 address next_ip = pc() + sizeof(int32_t) + rip_relative_correction; |
|
422 int64_t adjusted = (int64_t) disp - (next_ip - inst_mark()); |
|
423 assert(is_simm32(adjusted), |
|
424 "must be 32bit offset (RIP relative address)"); |
|
425 emit_data((int) adjusted, rspec, disp32_operand); |
|
426 |
|
427 #endif // _LP64 |
|
428 } else { |
|
429 // [disp] ABSOLUTE |
|
430 // [00 reg 100][00 100 101] disp32 |
|
431 emit_byte(0x04 | regenc << 3); |
|
432 emit_byte(0x25); |
|
433 emit_data(disp, rspec, disp32_operand); |
|
434 } |
|
435 } |
|
436 } |
|
437 |
|
438 void Assembler::emit_operand(XMMRegister reg, Register base, Register index, |
|
439 Address::ScaleFactor scale, int disp, |
|
440 RelocationHolder const& rspec, |
|
441 int rip_relative_correction) { |
|
442 relocInfo::relocType rtype = (relocInfo::relocType) rspec.type(); |
|
443 int regenc = reg->encoding(); |
|
444 if (regenc >= 8) { |
|
445 regenc -= 8; |
|
446 } |
|
447 if (base->is_valid()) { |
|
448 if (index->is_valid()) { |
|
449 assert(scale != Address::no_scale, "inconsistent address"); |
|
450 int indexenc = index->encoding(); |
|
451 if (indexenc >= 8) { |
|
452 indexenc -= 8; |
|
453 } |
|
454 int baseenc = base->encoding(); |
|
455 if (baseenc >= 8) { |
|
456 baseenc -= 8; |
|
457 } |
|
458 // [base + index*scale + disp] |
|
459 if (disp == 0 && rtype == relocInfo::none && |
|
460 base != rbp && base != r13) { |
|
461 // [base + index*scale] |
|
462 // [00 reg 100][ss index base] |
|
463 assert(index != rsp, "illegal addressing mode"); |
|
464 emit_byte(0x04 | regenc << 3); |
|
465 emit_byte(scale << 6 | indexenc << 3 | baseenc); |
|
466 } else if (is8bit(disp) && rtype == relocInfo::none) { |
|
467 // [base + index*scale + disp8] |
|
468 // [01 reg 100][ss index base] disp8 |
|
469 assert(index != rsp, "illegal addressing mode"); |
|
470 emit_byte(0x44 | regenc << 3); |
|
471 emit_byte(scale << 6 | indexenc << 3 | baseenc); |
|
472 emit_byte(disp & 0xFF); |
|
473 } else { |
|
474 // [base + index*scale + disp32] |
|
475 // [10 reg 100][ss index base] disp32 |
|
476 assert(index != rsp, "illegal addressing mode"); |
|
477 emit_byte(0x84 | regenc << 3); |
|
478 emit_byte(scale << 6 | indexenc << 3 | baseenc); |
|
479 emit_data(disp, rspec, disp32_operand); |
|
480 } |
|
481 } else if (base == rsp || base == r12) { |
|
482 // [rsp + disp] |
|
483 if (disp == 0 && rtype == relocInfo::none) { |
|
484 // [rsp] |
|
485 // [00 reg 100][00 100 100] |
|
486 emit_byte(0x04 | regenc << 3); |
|
487 emit_byte(0x24); |
|
488 } else if (is8bit(disp) && rtype == relocInfo::none) { |
|
489 // [rsp + imm8] |
|
490 // [01 reg 100][00 100 100] disp8 |
|
491 emit_byte(0x44 | regenc << 3); |
|
492 emit_byte(0x24); |
|
493 emit_byte(disp & 0xFF); |
|
494 } else { |
|
495 // [rsp + imm32] |
|
496 // [10 reg 100][00 100 100] disp32 |
|
497 emit_byte(0x84 | regenc << 3); |
|
498 emit_byte(0x24); |
|
499 emit_data(disp, rspec, disp32_operand); |
|
500 } |
|
501 } else { |
|
502 // [base + disp] |
|
503 assert(base != rsp && base != r12, "illegal addressing mode"); |
|
504 int baseenc = base->encoding(); |
|
505 if (baseenc >= 8) { |
|
506 baseenc -= 8; |
|
507 } |
|
508 if (disp == 0 && rtype == relocInfo::none && |
|
509 base != rbp && base != r13) { |
|
510 // [base] |
|
511 // [00 reg base] |
|
512 emit_byte(0x00 | regenc << 3 | baseenc); |
|
513 } else if (is8bit(disp) && rtype == relocInfo::none) { |
|
514 // [base + imm8] |
|
515 // [01 reg base] disp8 |
|
516 emit_byte(0x40 | regenc << 3 | baseenc); |
|
517 emit_byte(disp & 0xFF); |
|
518 } else { |
|
519 // [base + imm32] |
|
520 // [10 reg base] disp32 |
|
521 emit_byte(0x80 | regenc << 3 | baseenc); |
|
522 emit_data(disp, rspec, disp32_operand); |
|
523 } |
|
524 } |
|
525 } else { |
|
526 if (index->is_valid()) { |
|
527 assert(scale != Address::no_scale, "inconsistent address"); |
|
528 int indexenc = index->encoding(); |
|
529 if (indexenc >= 8) { |
|
530 indexenc -= 8; |
|
531 } |
|
532 // [index*scale + disp] |
|
533 // [00 reg 100][ss index 101] disp32 |
|
534 assert(index != rsp, "illegal addressing mode"); |
|
535 emit_byte(0x04 | regenc << 3); |
|
536 emit_byte(scale << 6 | indexenc << 3 | 0x05); |
|
537 emit_data(disp, rspec, disp32_operand); |
|
538 #ifdef _LP64 |
|
539 } else if ( rtype != relocInfo::none ) { |
|
540 // [disp] RIP-RELATIVE |
|
541 // [00 reg 101] disp32 |
|
542 emit_byte(0x05 | regenc << 3); |
|
543 // Note that the RIP-rel. correction applies to the generated |
|
544 // disp field, but _not_ to the target address in the rspec. |
|
545 |
|
546 // disp was created by converting the target address minus the pc |
|
547 // at the start of the instruction. That needs more correction here. |
|
548 // intptr_t disp = target - next_ip; |
|
549 |
|
550 assert(inst_mark() != NULL, "must be inside InstructionMark"); |
|
551 address next_ip = pc() + sizeof(int32_t) + rip_relative_correction; |
|
552 |
|
553 int64_t adjusted = (int64_t) disp - (next_ip - inst_mark()); |
|
554 assert(is_simm32(adjusted), |
|
555 "must be 32bit offset (RIP relative address)"); |
|
556 emit_data((int) adjusted, rspec, disp32_operand); |
|
557 #endif // _LP64 |
|
558 } else { |
|
559 // [disp] ABSOLUTE |
|
560 // [00 reg 100][00 100 101] disp32 |
|
561 emit_byte(0x04 | regenc << 3); |
|
562 emit_byte(0x25); |
|
563 emit_data(disp, rspec, disp32_operand); |
|
564 } |
|
565 } |
|
566 } |
|
567 |
|
568 // Secret local extension to Assembler::WhichOperand: |
|
569 #define end_pc_operand (_WhichOperand_limit) |
|
570 |
|
571 address Assembler::locate_operand(address inst, WhichOperand which) { |
|
572 // Decode the given instruction, and return the address of |
|
573 // an embedded 32-bit operand word. |
|
574 |
|
575 // If "which" is disp32_operand, selects the displacement portion |
|
576 // of an effective address specifier. |
|
577 // If "which" is imm64_operand, selects the trailing immediate constant. |
|
578 // If "which" is call32_operand, selects the displacement of a call or jump. |
|
579 // Caller is responsible for ensuring that there is such an operand, |
|
580 // and that it is 32/64 bits wide. |
|
581 |
|
582 // If "which" is end_pc_operand, find the end of the instruction. |
|
583 |
|
584 address ip = inst; |
|
585 bool is_64bit = false; |
|
586 |
|
587 debug_only(bool has_disp32 = false); |
|
588 int tail_size = 0; // other random bytes (#32, #16, etc.) at end of insn |
|
589 |
|
590 again_after_prefix: |
|
591 switch (0xFF & *ip++) { |
|
592 |
|
593 // These convenience macros generate groups of "case" labels for the switch. |
|
594 #define REP4(x) (x)+0: case (x)+1: case (x)+2: case (x)+3 |
|
595 #define REP8(x) (x)+0: case (x)+1: case (x)+2: case (x)+3: \ |
|
596 case (x)+4: case (x)+5: case (x)+6: case (x)+7 |
|
597 #define REP16(x) REP8((x)+0): \ |
|
598 case REP8((x)+8) |
|
599 |
|
600 case CS_segment: |
|
601 case SS_segment: |
|
602 case DS_segment: |
|
603 case ES_segment: |
|
604 case FS_segment: |
|
605 case GS_segment: |
|
606 assert(0, "shouldn't have that prefix"); |
|
607 assert(ip == inst + 1 || ip == inst + 2, "only two prefixes allowed"); |
|
608 goto again_after_prefix; |
|
609 |
|
610 case 0x67: |
|
611 case REX: |
|
612 case REX_B: |
|
613 case REX_X: |
|
614 case REX_XB: |
|
615 case REX_R: |
|
616 case REX_RB: |
|
617 case REX_RX: |
|
618 case REX_RXB: |
|
619 // assert(ip == inst + 1, "only one prefix allowed"); |
|
620 goto again_after_prefix; |
|
621 |
|
622 case REX_W: |
|
623 case REX_WB: |
|
624 case REX_WX: |
|
625 case REX_WXB: |
|
626 case REX_WR: |
|
627 case REX_WRB: |
|
628 case REX_WRX: |
|
629 case REX_WRXB: |
|
630 is_64bit = true; |
|
631 // assert(ip == inst + 1, "only one prefix allowed"); |
|
632 goto again_after_prefix; |
|
633 |
|
634 case 0xFF: // pushq a; decl a; incl a; call a; jmp a |
|
635 case 0x88: // movb a, r |
|
636 case 0x89: // movl a, r |
|
637 case 0x8A: // movb r, a |
|
638 case 0x8B: // movl r, a |
|
639 case 0x8F: // popl a |
|
640 debug_only(has_disp32 = true;) |
|
641 break; |
|
642 |
|
643 case 0x68: // pushq #32 |
|
644 if (which == end_pc_operand) { |
|
645 return ip + 4; |
|
646 } |
|
647 assert(0, "pushq has no disp32 or imm64"); |
|
648 ShouldNotReachHere(); |
|
649 |
|
650 case 0x66: // movw ... (size prefix) |
|
651 again_after_size_prefix2: |
|
652 switch (0xFF & *ip++) { |
|
653 case REX: |
|
654 case REX_B: |
|
655 case REX_X: |
|
656 case REX_XB: |
|
657 case REX_R: |
|
658 case REX_RB: |
|
659 case REX_RX: |
|
660 case REX_RXB: |
|
661 case REX_W: |
|
662 case REX_WB: |
|
663 case REX_WX: |
|
664 case REX_WXB: |
|
665 case REX_WR: |
|
666 case REX_WRB: |
|
667 case REX_WRX: |
|
668 case REX_WRXB: |
|
669 goto again_after_size_prefix2; |
|
670 case 0x8B: // movw r, a |
|
671 case 0x89: // movw a, r |
|
672 break; |
|
673 case 0xC7: // movw a, #16 |
|
674 tail_size = 2; // the imm16 |
|
675 break; |
|
676 case 0x0F: // several SSE/SSE2 variants |
|
677 ip--; // reparse the 0x0F |
|
678 goto again_after_prefix; |
|
679 default: |
|
680 ShouldNotReachHere(); |
|
681 } |
|
682 break; |
|
683 |
|
684 case REP8(0xB8): // movl/q r, #32/#64(oop?) |
|
685 if (which == end_pc_operand) return ip + (is_64bit ? 8 : 4); |
|
686 assert((which == call32_operand || which == imm64_operand) && is_64bit || |
|
687 which == narrow_oop_operand && !is_64bit, ""); |
|
688 return ip; |
|
689 |
|
690 case 0x69: // imul r, a, #32 |
|
691 case 0xC7: // movl a, #32(oop?) |
|
692 tail_size = 4; |
|
693 debug_only(has_disp32 = true); // has both kinds of operands! |
|
694 break; |
|
695 |
|
696 case 0x0F: // movx..., etc. |
|
697 switch (0xFF & *ip++) { |
|
698 case 0x12: // movlps |
|
699 case 0x28: // movaps |
|
700 case 0x2E: // ucomiss |
|
701 case 0x2F: // comiss |
|
702 case 0x54: // andps |
|
703 case 0x57: // xorps |
|
704 case 0x6E: // movd |
|
705 case 0x7E: // movd |
|
706 case 0xAE: // ldmxcsr a |
|
707 debug_only(has_disp32 = true); // has both kinds of operands! |
|
708 break; |
|
709 case 0xAD: // shrd r, a, %cl |
|
710 case 0xAF: // imul r, a |
|
711 case 0xBE: // movsbl r, a |
|
712 case 0xBF: // movswl r, a |
|
713 case 0xB6: // movzbl r, a |
|
714 case 0xB7: // movzwl r, a |
|
715 case REP16(0x40): // cmovl cc, r, a |
|
716 case 0xB0: // cmpxchgb |
|
717 case 0xB1: // cmpxchg |
|
718 case 0xC1: // xaddl |
|
719 case 0xC7: // cmpxchg8 |
|
720 case REP16(0x90): // setcc a |
|
721 debug_only(has_disp32 = true); |
|
722 // fall out of the switch to decode the address |
|
723 break; |
|
724 case 0xAC: // shrd r, a, #8 |
|
725 debug_only(has_disp32 = true); |
|
726 tail_size = 1; // the imm8 |
|
727 break; |
|
728 case REP16(0x80): // jcc rdisp32 |
|
729 if (which == end_pc_operand) return ip + 4; |
|
730 assert(which == call32_operand, "jcc has no disp32 or imm64"); |
|
731 return ip; |
|
732 default: |
|
733 ShouldNotReachHere(); |
|
734 } |
|
735 break; |
|
736 |
|
737 case 0x81: // addl a, #32; addl r, #32 |
|
738 // also: orl, adcl, sbbl, andl, subl, xorl, cmpl |
|
739 tail_size = 4; |
|
740 debug_only(has_disp32 = true); // has both kinds of operands! |
|
741 break; |
|
742 |
|
743 case 0x83: // addl a, #8; addl r, #8 |
|
744 // also: orl, adcl, sbbl, andl, subl, xorl, cmpl |
|
745 debug_only(has_disp32 = true); // has both kinds of operands! |
|
746 tail_size = 1; |
|
747 break; |
|
748 |
|
749 case 0x9B: |
|
750 switch (0xFF & *ip++) { |
|
751 case 0xD9: // fnstcw a |
|
752 debug_only(has_disp32 = true); |
|
753 break; |
|
754 default: |
|
755 ShouldNotReachHere(); |
|
756 } |
|
757 break; |
|
758 |
|
759 case REP4(0x00): // addb a, r; addl a, r; addb r, a; addl r, a |
|
760 case REP4(0x10): // adc... |
|
761 case REP4(0x20): // and... |
|
762 case REP4(0x30): // xor... |
|
763 case REP4(0x08): // or... |
|
764 case REP4(0x18): // sbb... |
|
765 case REP4(0x28): // sub... |
|
766 case 0xF7: // mull a |
|
767 case 0x87: // xchg r, a |
|
768 debug_only(has_disp32 = true); |
|
769 break; |
|
770 case REP4(0x38): // cmp... |
|
771 case 0x8D: // lea r, a |
|
772 case 0x85: // test r, a |
|
773 debug_only(has_disp32 = true); // has both kinds of operands! |
|
774 break; |
|
775 |
|
776 case 0xC1: // sal a, #8; sar a, #8; shl a, #8; shr a, #8 |
|
777 case 0xC6: // movb a, #8 |
|
778 case 0x80: // cmpb a, #8 |
|
779 case 0x6B: // imul r, a, #8 |
|
780 debug_only(has_disp32 = true); // has both kinds of operands! |
|
781 tail_size = 1; // the imm8 |
|
782 break; |
|
783 |
|
784 case 0xE8: // call rdisp32 |
|
785 case 0xE9: // jmp rdisp32 |
|
786 if (which == end_pc_operand) return ip + 4; |
|
787 assert(which == call32_operand, "call has no disp32 or imm32"); |
|
788 return ip; |
|
789 |
|
790 case 0xD1: // sal a, 1; sar a, 1; shl a, 1; shr a, 1 |
|
791 case 0xD3: // sal a, %cl; sar a, %cl; shl a, %cl; shr a, %cl |
|
792 case 0xD9: // fld_s a; fst_s a; fstp_s a; fldcw a |
|
793 case 0xDD: // fld_d a; fst_d a; fstp_d a |
|
794 case 0xDB: // fild_s a; fistp_s a; fld_x a; fstp_x a |
|
795 case 0xDF: // fild_d a; fistp_d a |
|
796 case 0xD8: // fadd_s a; fsubr_s a; fmul_s a; fdivr_s a; fcomp_s a |
|
797 case 0xDC: // fadd_d a; fsubr_d a; fmul_d a; fdivr_d a; fcomp_d a |
|
798 case 0xDE: // faddp_d a; fsubrp_d a; fmulp_d a; fdivrp_d a; fcompp_d a |
|
799 debug_only(has_disp32 = true); |
|
800 break; |
|
801 |
|
802 case 0xF3: // For SSE |
|
803 case 0xF2: // For SSE2 |
|
804 switch (0xFF & *ip++) { |
|
805 case REX: |
|
806 case REX_B: |
|
807 case REX_X: |
|
808 case REX_XB: |
|
809 case REX_R: |
|
810 case REX_RB: |
|
811 case REX_RX: |
|
812 case REX_RXB: |
|
813 case REX_W: |
|
814 case REX_WB: |
|
815 case REX_WX: |
|
816 case REX_WXB: |
|
817 case REX_WR: |
|
818 case REX_WRB: |
|
819 case REX_WRX: |
|
820 case REX_WRXB: |
|
821 ip++; |
|
822 default: |
|
823 ip++; |
|
824 } |
|
825 debug_only(has_disp32 = true); // has both kinds of operands! |
|
826 break; |
|
827 |
|
828 default: |
|
829 ShouldNotReachHere(); |
|
830 |
|
831 #undef REP8 |
|
832 #undef REP16 |
|
833 } |
|
834 |
|
835 assert(which != call32_operand, "instruction is not a call, jmp, or jcc"); |
|
836 assert(which != imm64_operand, "instruction is not a movq reg, imm64"); |
|
837 assert(which != disp32_operand || has_disp32, "instruction has no disp32 field"); |
|
838 |
|
839 // parse the output of emit_operand |
|
840 int op2 = 0xFF & *ip++; |
|
841 int base = op2 & 0x07; |
|
842 int op3 = -1; |
|
843 const int b100 = 4; |
|
844 const int b101 = 5; |
|
845 if (base == b100 && (op2 >> 6) != 3) { |
|
846 op3 = 0xFF & *ip++; |
|
847 base = op3 & 0x07; // refetch the base |
|
848 } |
|
849 // now ip points at the disp (if any) |
|
850 |
|
851 switch (op2 >> 6) { |
|
852 case 0: |
|
853 // [00 reg 100][ss index base] |
|
854 // [00 reg 100][00 100 esp] |
|
855 // [00 reg base] |
|
856 // [00 reg 100][ss index 101][disp32] |
|
857 // [00 reg 101] [disp32] |
|
858 |
|
859 if (base == b101) { |
|
860 if (which == disp32_operand) |
|
861 return ip; // caller wants the disp32 |
|
862 ip += 4; // skip the disp32 |
|
863 } |
|
864 break; |
|
865 |
|
866 case 1: |
|
867 // [01 reg 100][ss index base][disp8] |
|
868 // [01 reg 100][00 100 esp][disp8] |
|
869 // [01 reg base] [disp8] |
|
870 ip += 1; // skip the disp8 |
|
871 break; |
|
872 |
|
873 case 2: |
|
874 // [10 reg 100][ss index base][disp32] |
|
875 // [10 reg 100][00 100 esp][disp32] |
|
876 // [10 reg base] [disp32] |
|
877 if (which == disp32_operand) |
|
878 return ip; // caller wants the disp32 |
|
879 ip += 4; // skip the disp32 |
|
880 break; |
|
881 |
|
882 case 3: |
|
883 // [11 reg base] (not a memory addressing mode) |
|
884 break; |
|
885 } |
|
886 |
|
887 if (which == end_pc_operand) { |
|
888 return ip + tail_size; |
|
889 } |
|
890 |
|
891 assert(0, "fix locate_operand"); |
|
892 return ip; |
|
893 } |
|
894 |
|
895 address Assembler::locate_next_instruction(address inst) { |
|
896 // Secretly share code with locate_operand: |
|
897 return locate_operand(inst, end_pc_operand); |
|
898 } |
|
899 |
|
900 #ifdef ASSERT |
|
901 void Assembler::check_relocation(RelocationHolder const& rspec, int format) { |
|
902 address inst = inst_mark(); |
|
903 assert(inst != NULL && inst < pc(), |
|
904 "must point to beginning of instruction"); |
|
905 address opnd; |
|
906 |
|
907 Relocation* r = rspec.reloc(); |
|
908 if (r->type() == relocInfo::none) { |
|
909 return; |
|
910 } else if (r->is_call() || format == call32_operand) { |
|
911 opnd = locate_operand(inst, call32_operand); |
|
912 } else if (r->is_data()) { |
|
913 assert(format == imm64_operand || format == disp32_operand || |
|
914 format == narrow_oop_operand, "format ok"); |
|
915 opnd = locate_operand(inst, (WhichOperand) format); |
|
916 } else { |
|
917 assert(format == 0, "cannot specify a format"); |
|
918 return; |
|
919 } |
|
920 assert(opnd == pc(), "must put operand where relocs can find it"); |
|
921 } |
|
922 #endif |
|
923 |
|
924 int Assembler::prefix_and_encode(int reg_enc, bool byteinst) { |
|
925 if (reg_enc >= 8) { |
|
926 prefix(REX_B); |
|
927 reg_enc -= 8; |
|
928 } else if (byteinst && reg_enc >= 4) { |
|
929 prefix(REX); |
|
930 } |
|
931 return reg_enc; |
|
932 } |
|
933 |
|
934 int Assembler::prefixq_and_encode(int reg_enc) { |
|
935 if (reg_enc < 8) { |
|
936 prefix(REX_W); |
|
937 } else { |
|
938 prefix(REX_WB); |
|
939 reg_enc -= 8; |
|
940 } |
|
941 return reg_enc; |
|
942 } |
|
943 |
|
944 int Assembler::prefix_and_encode(int dst_enc, int src_enc, bool byteinst) { |
|
945 if (dst_enc < 8) { |
|
946 if (src_enc >= 8) { |
|
947 prefix(REX_B); |
|
948 src_enc -= 8; |
|
949 } else if (byteinst && src_enc >= 4) { |
|
950 prefix(REX); |
|
951 } |
|
952 } else { |
|
953 if (src_enc < 8) { |
|
954 prefix(REX_R); |
|
955 } else { |
|
956 prefix(REX_RB); |
|
957 src_enc -= 8; |
|
958 } |
|
959 dst_enc -= 8; |
|
960 } |
|
961 return dst_enc << 3 | src_enc; |
|
962 } |
|
963 |
|
964 int Assembler::prefixq_and_encode(int dst_enc, int src_enc) { |
|
965 if (dst_enc < 8) { |
|
966 if (src_enc < 8) { |
|
967 prefix(REX_W); |
|
968 } else { |
|
969 prefix(REX_WB); |
|
970 src_enc -= 8; |
|
971 } |
|
972 } else { |
|
973 if (src_enc < 8) { |
|
974 prefix(REX_WR); |
|
975 } else { |
|
976 prefix(REX_WRB); |
|
977 src_enc -= 8; |
|
978 } |
|
979 dst_enc -= 8; |
|
980 } |
|
981 return dst_enc << 3 | src_enc; |
|
982 } |
|
983 |
|
984 void Assembler::prefix(Register reg) { |
|
985 if (reg->encoding() >= 8) { |
|
986 prefix(REX_B); |
|
987 } |
|
988 } |
|
989 |
|
990 void Assembler::prefix(Address adr) { |
|
991 if (adr.base_needs_rex()) { |
|
992 if (adr.index_needs_rex()) { |
|
993 prefix(REX_XB); |
|
994 } else { |
|
995 prefix(REX_B); |
|
996 } |
|
997 } else { |
|
998 if (adr.index_needs_rex()) { |
|
999 prefix(REX_X); |
|
1000 } |
|
1001 } |
|
1002 } |
|
1003 |
|
1004 void Assembler::prefixq(Address adr) { |
|
1005 if (adr.base_needs_rex()) { |
|
1006 if (adr.index_needs_rex()) { |
|
1007 prefix(REX_WXB); |
|
1008 } else { |
|
1009 prefix(REX_WB); |
|
1010 } |
|
1011 } else { |
|
1012 if (adr.index_needs_rex()) { |
|
1013 prefix(REX_WX); |
|
1014 } else { |
|
1015 prefix(REX_W); |
|
1016 } |
|
1017 } |
|
1018 } |
|
1019 |
|
1020 |
|
1021 void Assembler::prefix(Address adr, Register reg, bool byteinst) { |
|
1022 if (reg->encoding() < 8) { |
|
1023 if (adr.base_needs_rex()) { |
|
1024 if (adr.index_needs_rex()) { |
|
1025 prefix(REX_XB); |
|
1026 } else { |
|
1027 prefix(REX_B); |
|
1028 } |
|
1029 } else { |
|
1030 if (adr.index_needs_rex()) { |
|
1031 prefix(REX_X); |
|
1032 } else if (reg->encoding() >= 4 ) { |
|
1033 prefix(REX); |
|
1034 } |
|
1035 } |
|
1036 } else { |
|
1037 if (adr.base_needs_rex()) { |
|
1038 if (adr.index_needs_rex()) { |
|
1039 prefix(REX_RXB); |
|
1040 } else { |
|
1041 prefix(REX_RB); |
|
1042 } |
|
1043 } else { |
|
1044 if (adr.index_needs_rex()) { |
|
1045 prefix(REX_RX); |
|
1046 } else { |
|
1047 prefix(REX_R); |
|
1048 } |
|
1049 } |
|
1050 } |
|
1051 } |
|
1052 |
|
1053 void Assembler::prefixq(Address adr, Register src) { |
|
1054 if (src->encoding() < 8) { |
|
1055 if (adr.base_needs_rex()) { |
|
1056 if (adr.index_needs_rex()) { |
|
1057 prefix(REX_WXB); |
|
1058 } else { |
|
1059 prefix(REX_WB); |
|
1060 } |
|
1061 } else { |
|
1062 if (adr.index_needs_rex()) { |
|
1063 prefix(REX_WX); |
|
1064 } else { |
|
1065 prefix(REX_W); |
|
1066 } |
|
1067 } |
|
1068 } else { |
|
1069 if (adr.base_needs_rex()) { |
|
1070 if (adr.index_needs_rex()) { |
|
1071 prefix(REX_WRXB); |
|
1072 } else { |
|
1073 prefix(REX_WRB); |
|
1074 } |
|
1075 } else { |
|
1076 if (adr.index_needs_rex()) { |
|
1077 prefix(REX_WRX); |
|
1078 } else { |
|
1079 prefix(REX_WR); |
|
1080 } |
|
1081 } |
|
1082 } |
|
1083 } |
|
1084 |
|
1085 void Assembler::prefix(Address adr, XMMRegister reg) { |
|
1086 if (reg->encoding() < 8) { |
|
1087 if (adr.base_needs_rex()) { |
|
1088 if (adr.index_needs_rex()) { |
|
1089 prefix(REX_XB); |
|
1090 } else { |
|
1091 prefix(REX_B); |
|
1092 } |
|
1093 } else { |
|
1094 if (adr.index_needs_rex()) { |
|
1095 prefix(REX_X); |
|
1096 } |
|
1097 } |
|
1098 } else { |
|
1099 if (adr.base_needs_rex()) { |
|
1100 if (adr.index_needs_rex()) { |
|
1101 prefix(REX_RXB); |
|
1102 } else { |
|
1103 prefix(REX_RB); |
|
1104 } |
|
1105 } else { |
|
1106 if (adr.index_needs_rex()) { |
|
1107 prefix(REX_RX); |
|
1108 } else { |
|
1109 prefix(REX_R); |
|
1110 } |
|
1111 } |
|
1112 } |
|
1113 } |
|
1114 |
|
1115 void Assembler::emit_operand(Register reg, Address adr, |
|
1116 int rip_relative_correction) { |
|
1117 emit_operand(reg, adr._base, adr._index, adr._scale, adr._disp, |
|
1118 adr._rspec, |
|
1119 rip_relative_correction); |
|
1120 } |
|
1121 |
|
1122 void Assembler::emit_operand(XMMRegister reg, Address adr, |
|
1123 int rip_relative_correction) { |
|
1124 emit_operand(reg, adr._base, adr._index, adr._scale, adr._disp, |
|
1125 adr._rspec, |
|
1126 rip_relative_correction); |
|
1127 } |
|
1128 |
|
1129 void Assembler::emit_farith(int b1, int b2, int i) { |
|
1130 assert(isByte(b1) && isByte(b2), "wrong opcode"); |
|
1131 assert(0 <= i && i < 8, "illegal stack offset"); |
|
1132 emit_byte(b1); |
|
1133 emit_byte(b2 + i); |
|
1134 } |
|
1135 |
|
1136 // pushad is invalid, use this instead. |
|
1137 // NOTE: Kills flags!! |
|
1138 void Assembler::pushaq() { |
|
1139 // we have to store original rsp. ABI says that 128 bytes |
|
1140 // below rsp are local scratch. |
|
1141 movq(Address(rsp, -5 * wordSize), rsp); |
|
1142 |
|
1143 subq(rsp, 16 * wordSize); |
|
1144 |
|
1145 movq(Address(rsp, 15 * wordSize), rax); |
|
1146 movq(Address(rsp, 14 * wordSize), rcx); |
|
1147 movq(Address(rsp, 13 * wordSize), rdx); |
|
1148 movq(Address(rsp, 12 * wordSize), rbx); |
|
1149 // skip rsp |
|
1150 movq(Address(rsp, 10 * wordSize), rbp); |
|
1151 movq(Address(rsp, 9 * wordSize), rsi); |
|
1152 movq(Address(rsp, 8 * wordSize), rdi); |
|
1153 movq(Address(rsp, 7 * wordSize), r8); |
|
1154 movq(Address(rsp, 6 * wordSize), r9); |
|
1155 movq(Address(rsp, 5 * wordSize), r10); |
|
1156 movq(Address(rsp, 4 * wordSize), r11); |
|
1157 movq(Address(rsp, 3 * wordSize), r12); |
|
1158 movq(Address(rsp, 2 * wordSize), r13); |
|
1159 movq(Address(rsp, wordSize), r14); |
|
1160 movq(Address(rsp, 0), r15); |
|
1161 } |
|
1162 |
|
1163 // popad is invalid, use this instead |
|
1164 // NOTE: Kills flags!! |
|
1165 void Assembler::popaq() { |
|
1166 movq(r15, Address(rsp, 0)); |
|
1167 movq(r14, Address(rsp, wordSize)); |
|
1168 movq(r13, Address(rsp, 2 * wordSize)); |
|
1169 movq(r12, Address(rsp, 3 * wordSize)); |
|
1170 movq(r11, Address(rsp, 4 * wordSize)); |
|
1171 movq(r10, Address(rsp, 5 * wordSize)); |
|
1172 movq(r9, Address(rsp, 6 * wordSize)); |
|
1173 movq(r8, Address(rsp, 7 * wordSize)); |
|
1174 movq(rdi, Address(rsp, 8 * wordSize)); |
|
1175 movq(rsi, Address(rsp, 9 * wordSize)); |
|
1176 movq(rbp, Address(rsp, 10 * wordSize)); |
|
1177 // skip rsp |
|
1178 movq(rbx, Address(rsp, 12 * wordSize)); |
|
1179 movq(rdx, Address(rsp, 13 * wordSize)); |
|
1180 movq(rcx, Address(rsp, 14 * wordSize)); |
|
1181 movq(rax, Address(rsp, 15 * wordSize)); |
|
1182 |
|
1183 addq(rsp, 16 * wordSize); |
|
1184 } |
|
1185 |
|
1186 void Assembler::pushfq() { |
|
1187 emit_byte(0x9C); |
|
1188 } |
|
1189 |
|
1190 void Assembler::popfq() { |
|
1191 emit_byte(0x9D); |
|
1192 } |
|
1193 |
|
1194 void Assembler::pushq(int imm32) { |
|
1195 emit_byte(0x68); |
|
1196 emit_long(imm32); |
|
1197 } |
|
1198 |
|
1199 void Assembler::pushq(Register src) { |
|
1200 int encode = prefix_and_encode(src->encoding()); |
|
1201 |
|
1202 emit_byte(0x50 | encode); |
|
1203 } |
|
1204 |
|
1205 void Assembler::pushq(Address src) { |
|
1206 InstructionMark im(this); |
|
1207 prefix(src); |
|
1208 emit_byte(0xFF); |
|
1209 emit_operand(rsi, src); |
|
1210 } |
|
1211 |
|
1212 void Assembler::popq(Register dst) { |
|
1213 int encode = prefix_and_encode(dst->encoding()); |
|
1214 emit_byte(0x58 | encode); |
|
1215 } |
|
1216 |
|
1217 void Assembler::popq(Address dst) { |
|
1218 InstructionMark im(this); |
|
1219 prefix(dst); |
|
1220 emit_byte(0x8F); |
|
1221 emit_operand(rax, dst); |
|
1222 } |
|
1223 |
|
1224 void Assembler::prefix(Prefix p) { |
|
1225 a_byte(p); |
|
1226 } |
|
1227 |
|
1228 void Assembler::movb(Register dst, Address src) { |
|
1229 InstructionMark im(this); |
|
1230 prefix(src, dst, true); |
|
1231 emit_byte(0x8A); |
|
1232 emit_operand(dst, src); |
|
1233 } |
|
1234 |
|
1235 void Assembler::movb(Address dst, int imm8) { |
|
1236 InstructionMark im(this); |
|
1237 prefix(dst); |
|
1238 emit_byte(0xC6); |
|
1239 emit_operand(rax, dst, 1); |
|
1240 emit_byte(imm8); |
|
1241 } |
|
1242 |
|
1243 void Assembler::movb(Address dst, Register src) { |
|
1244 InstructionMark im(this); |
|
1245 prefix(dst, src, true); |
|
1246 emit_byte(0x88); |
|
1247 emit_operand(src, dst); |
|
1248 } |
|
1249 |
|
1250 void Assembler::movw(Address dst, int imm16) { |
|
1251 InstructionMark im(this); |
|
1252 emit_byte(0x66); // switch to 16-bit mode |
|
1253 prefix(dst); |
|
1254 emit_byte(0xC7); |
|
1255 emit_operand(rax, dst, 2); |
|
1256 emit_word(imm16); |
|
1257 } |
|
1258 |
|
1259 void Assembler::movw(Register dst, Address src) { |
|
1260 InstructionMark im(this); |
|
1261 emit_byte(0x66); |
|
1262 prefix(src, dst); |
|
1263 emit_byte(0x8B); |
|
1264 emit_operand(dst, src); |
|
1265 } |
|
1266 |
|
1267 void Assembler::movw(Address dst, Register src) { |
|
1268 InstructionMark im(this); |
|
1269 emit_byte(0x66); |
|
1270 prefix(dst, src); |
|
1271 emit_byte(0x89); |
|
1272 emit_operand(src, dst); |
|
1273 } |
|
1274 |
|
1275 // Uses zero extension. |
|
1276 void Assembler::movl(Register dst, int imm32) { |
|
1277 int encode = prefix_and_encode(dst->encoding()); |
|
1278 emit_byte(0xB8 | encode); |
|
1279 emit_long(imm32); |
|
1280 } |
|
1281 |
|
1282 void Assembler::movl(Register dst, Register src) { |
|
1283 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
1284 emit_byte(0x8B); |
|
1285 emit_byte(0xC0 | encode); |
|
1286 } |
|
1287 |
|
1288 void Assembler::movl(Register dst, Address src) { |
|
1289 InstructionMark im(this); |
|
1290 prefix(src, dst); |
|
1291 emit_byte(0x8B); |
|
1292 emit_operand(dst, src); |
|
1293 } |
|
1294 |
|
1295 void Assembler::movl(Address dst, int imm32) { |
|
1296 InstructionMark im(this); |
|
1297 prefix(dst); |
|
1298 emit_byte(0xC7); |
|
1299 emit_operand(rax, dst, 4); |
|
1300 emit_long(imm32); |
|
1301 } |
|
1302 |
|
1303 void Assembler::movl(Address dst, Register src) { |
|
1304 InstructionMark im(this); |
|
1305 prefix(dst, src); |
|
1306 emit_byte(0x89); |
|
1307 emit_operand(src, dst); |
|
1308 } |
|
1309 |
|
1310 void Assembler::mov64(Register dst, intptr_t imm64) { |
|
1311 InstructionMark im(this); |
|
1312 int encode = prefixq_and_encode(dst->encoding()); |
|
1313 emit_byte(0xB8 | encode); |
|
1314 emit_long64(imm64); |
|
1315 } |
|
1316 |
|
1317 void Assembler::mov_literal64(Register dst, intptr_t imm64, RelocationHolder const& rspec) { |
|
1318 InstructionMark im(this); |
|
1319 int encode = prefixq_and_encode(dst->encoding()); |
|
1320 emit_byte(0xB8 | encode); |
|
1321 emit_data64(imm64, rspec); |
|
1322 } |
|
1323 |
|
1324 void Assembler::movq(Register dst, Register src) { |
|
1325 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); |
|
1326 emit_byte(0x8B); |
|
1327 emit_byte(0xC0 | encode); |
|
1328 } |
|
1329 |
|
1330 void Assembler::movq(Register dst, Address src) { |
|
1331 InstructionMark im(this); |
|
1332 prefixq(src, dst); |
|
1333 emit_byte(0x8B); |
|
1334 emit_operand(dst, src); |
|
1335 } |
|
1336 |
|
1337 void Assembler::mov64(Address dst, intptr_t imm32) { |
|
1338 assert(is_simm32(imm32), "lost bits"); |
|
1339 InstructionMark im(this); |
|
1340 prefixq(dst); |
|
1341 emit_byte(0xC7); |
|
1342 emit_operand(rax, dst, 4); |
|
1343 emit_long(imm32); |
|
1344 } |
|
1345 |
|
1346 void Assembler::movq(Address dst, Register src) { |
|
1347 InstructionMark im(this); |
|
1348 prefixq(dst, src); |
|
1349 emit_byte(0x89); |
|
1350 emit_operand(src, dst); |
|
1351 } |
|
1352 |
|
1353 void Assembler::movsbl(Register dst, Address src) { |
|
1354 InstructionMark im(this); |
|
1355 prefix(src, dst); |
|
1356 emit_byte(0x0F); |
|
1357 emit_byte(0xBE); |
|
1358 emit_operand(dst, src); |
|
1359 } |
|
1360 |
|
1361 void Assembler::movsbl(Register dst, Register src) { |
|
1362 int encode = prefix_and_encode(dst->encoding(), src->encoding(), true); |
|
1363 emit_byte(0x0F); |
|
1364 emit_byte(0xBE); |
|
1365 emit_byte(0xC0 | encode); |
|
1366 } |
|
1367 |
|
1368 void Assembler::movswl(Register dst, Address src) { |
|
1369 InstructionMark im(this); |
|
1370 prefix(src, dst); |
|
1371 emit_byte(0x0F); |
|
1372 emit_byte(0xBF); |
|
1373 emit_operand(dst, src); |
|
1374 } |
|
1375 |
|
1376 void Assembler::movswl(Register dst, Register src) { |
|
1377 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
1378 emit_byte(0x0F); |
|
1379 emit_byte(0xBF); |
|
1380 emit_byte(0xC0 | encode); |
|
1381 } |
|
1382 |
|
1383 void Assembler::movslq(Register dst, Address src) { |
|
1384 InstructionMark im(this); |
|
1385 prefixq(src, dst); |
|
1386 emit_byte(0x63); |
|
1387 emit_operand(dst, src); |
|
1388 } |
|
1389 |
|
1390 void Assembler::movslq(Register dst, Register src) { |
|
1391 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); |
|
1392 emit_byte(0x63); |
|
1393 emit_byte(0xC0 | encode); |
|
1394 } |
|
1395 |
|
1396 void Assembler::movzbl(Register dst, Address src) { |
|
1397 InstructionMark im(this); |
|
1398 prefix(src, dst); |
|
1399 emit_byte(0x0F); |
|
1400 emit_byte(0xB6); |
|
1401 emit_operand(dst, src); |
|
1402 } |
|
1403 |
|
1404 void Assembler::movzbl(Register dst, Register src) { |
|
1405 int encode = prefix_and_encode(dst->encoding(), src->encoding(), true); |
|
1406 emit_byte(0x0F); |
|
1407 emit_byte(0xB6); |
|
1408 emit_byte(0xC0 | encode); |
|
1409 } |
|
1410 |
|
1411 void Assembler::movzwl(Register dst, Address src) { |
|
1412 InstructionMark im(this); |
|
1413 prefix(src, dst); |
|
1414 emit_byte(0x0F); |
|
1415 emit_byte(0xB7); |
|
1416 emit_operand(dst, src); |
|
1417 } |
|
1418 |
|
1419 void Assembler::movzwl(Register dst, Register src) { |
|
1420 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
1421 emit_byte(0x0F); |
|
1422 emit_byte(0xB7); |
|
1423 emit_byte(0xC0 | encode); |
|
1424 } |
|
1425 |
|
1426 void Assembler::movss(XMMRegister dst, XMMRegister src) { |
|
1427 emit_byte(0xF3); |
|
1428 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
1429 emit_byte(0x0F); |
|
1430 emit_byte(0x10); |
|
1431 emit_byte(0xC0 | encode); |
|
1432 } |
|
1433 |
|
1434 void Assembler::movss(XMMRegister dst, Address src) { |
|
1435 InstructionMark im(this); |
|
1436 emit_byte(0xF3); |
|
1437 prefix(src, dst); |
|
1438 emit_byte(0x0F); |
|
1439 emit_byte(0x10); |
|
1440 emit_operand(dst, src); |
|
1441 } |
|
1442 |
|
1443 void Assembler::movss(Address dst, XMMRegister src) { |
|
1444 InstructionMark im(this); |
|
1445 emit_byte(0xF3); |
|
1446 prefix(dst, src); |
|
1447 emit_byte(0x0F); |
|
1448 emit_byte(0x11); |
|
1449 emit_operand(src, dst); |
|
1450 } |
|
1451 |
|
1452 void Assembler::movsd(XMMRegister dst, XMMRegister src) { |
|
1453 emit_byte(0xF2); |
|
1454 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
1455 emit_byte(0x0F); |
|
1456 emit_byte(0x10); |
|
1457 emit_byte(0xC0 | encode); |
|
1458 } |
|
1459 |
|
1460 void Assembler::movsd(XMMRegister dst, Address src) { |
|
1461 InstructionMark im(this); |
|
1462 emit_byte(0xF2); |
|
1463 prefix(src, dst); |
|
1464 emit_byte(0x0F); |
|
1465 emit_byte(0x10); |
|
1466 emit_operand(dst, src); |
|
1467 } |
|
1468 |
|
1469 void Assembler::movsd(Address dst, XMMRegister src) { |
|
1470 InstructionMark im(this); |
|
1471 emit_byte(0xF2); |
|
1472 prefix(dst, src); |
|
1473 emit_byte(0x0F); |
|
1474 emit_byte(0x11); |
|
1475 emit_operand(src, dst); |
|
1476 } |
|
1477 |
|
1478 // New cpus require to use movsd and movss to avoid partial register stall |
|
1479 // when loading from memory. But for old Opteron use movlpd instead of movsd. |
|
1480 // The selection is done in MacroAssembler::movdbl() and movflt(). |
|
1481 void Assembler::movlpd(XMMRegister dst, Address src) { |
|
1482 InstructionMark im(this); |
|
1483 emit_byte(0x66); |
|
1484 prefix(src, dst); |
|
1485 emit_byte(0x0F); |
|
1486 emit_byte(0x12); |
|
1487 emit_operand(dst, src); |
|
1488 } |
|
1489 |
|
1490 void Assembler::movapd(XMMRegister dst, XMMRegister src) { |
|
1491 int dstenc = dst->encoding(); |
|
1492 int srcenc = src->encoding(); |
|
1493 emit_byte(0x66); |
|
1494 if (dstenc < 8) { |
|
1495 if (srcenc >= 8) { |
|
1496 prefix(REX_B); |
|
1497 srcenc -= 8; |
|
1498 } |
|
1499 } else { |
|
1500 if (srcenc < 8) { |
|
1501 prefix(REX_R); |
|
1502 } else { |
|
1503 prefix(REX_RB); |
|
1504 srcenc -= 8; |
|
1505 } |
|
1506 dstenc -= 8; |
|
1507 } |
|
1508 emit_byte(0x0F); |
|
1509 emit_byte(0x28); |
|
1510 emit_byte(0xC0 | dstenc << 3 | srcenc); |
|
1511 } |
|
1512 |
|
1513 void Assembler::movaps(XMMRegister dst, XMMRegister src) { |
|
1514 int dstenc = dst->encoding(); |
|
1515 int srcenc = src->encoding(); |
|
1516 if (dstenc < 8) { |
|
1517 if (srcenc >= 8) { |
|
1518 prefix(REX_B); |
|
1519 srcenc -= 8; |
|
1520 } |
|
1521 } else { |
|
1522 if (srcenc < 8) { |
|
1523 prefix(REX_R); |
|
1524 } else { |
|
1525 prefix(REX_RB); |
|
1526 srcenc -= 8; |
|
1527 } |
|
1528 dstenc -= 8; |
|
1529 } |
|
1530 emit_byte(0x0F); |
|
1531 emit_byte(0x28); |
|
1532 emit_byte(0xC0 | dstenc << 3 | srcenc); |
|
1533 } |
|
1534 |
|
1535 void Assembler::movdl(XMMRegister dst, Register src) { |
|
1536 emit_byte(0x66); |
|
1537 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
1538 emit_byte(0x0F); |
|
1539 emit_byte(0x6E); |
|
1540 emit_byte(0xC0 | encode); |
|
1541 } |
|
1542 |
|
1543 void Assembler::movdl(Register dst, XMMRegister src) { |
|
1544 emit_byte(0x66); |
|
1545 // swap src/dst to get correct prefix |
|
1546 int encode = prefix_and_encode(src->encoding(), dst->encoding()); |
|
1547 emit_byte(0x0F); |
|
1548 emit_byte(0x7E); |
|
1549 emit_byte(0xC0 | encode); |
|
1550 } |
|
1551 |
|
1552 void Assembler::movdq(XMMRegister dst, Register src) { |
|
1553 emit_byte(0x66); |
|
1554 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); |
|
1555 emit_byte(0x0F); |
|
1556 emit_byte(0x6E); |
|
1557 emit_byte(0xC0 | encode); |
|
1558 } |
|
1559 |
|
1560 void Assembler::movdq(Register dst, XMMRegister src) { |
|
1561 emit_byte(0x66); |
|
1562 // swap src/dst to get correct prefix |
|
1563 int encode = prefixq_and_encode(src->encoding(), dst->encoding()); |
|
1564 emit_byte(0x0F); |
|
1565 emit_byte(0x7E); |
|
1566 emit_byte(0xC0 | encode); |
|
1567 } |
|
1568 |
|
1569 void Assembler::pxor(XMMRegister dst, Address src) { |
|
1570 InstructionMark im(this); |
|
1571 emit_byte(0x66); |
|
1572 prefix(src, dst); |
|
1573 emit_byte(0x0F); |
|
1574 emit_byte(0xEF); |
|
1575 emit_operand(dst, src); |
|
1576 } |
|
1577 |
|
1578 void Assembler::pxor(XMMRegister dst, XMMRegister src) { |
|
1579 InstructionMark im(this); |
|
1580 emit_byte(0x66); |
|
1581 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
1582 emit_byte(0x0F); |
|
1583 emit_byte(0xEF); |
|
1584 emit_byte(0xC0 | encode); |
|
1585 } |
|
1586 |
|
1587 void Assembler::movdqa(XMMRegister dst, Address src) { |
|
1588 InstructionMark im(this); |
|
1589 emit_byte(0x66); |
|
1590 prefix(src, dst); |
|
1591 emit_byte(0x0F); |
|
1592 emit_byte(0x6F); |
|
1593 emit_operand(dst, src); |
|
1594 } |
|
1595 |
|
1596 void Assembler::movdqa(XMMRegister dst, XMMRegister src) { |
|
1597 emit_byte(0x66); |
|
1598 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); |
|
1599 emit_byte(0x0F); |
|
1600 emit_byte(0x6F); |
|
1601 emit_byte(0xC0 | encode); |
|
1602 } |
|
1603 |
|
1604 void Assembler::movdqa(Address dst, XMMRegister src) { |
|
1605 InstructionMark im(this); |
|
1606 emit_byte(0x66); |
|
1607 prefix(dst, src); |
|
1608 emit_byte(0x0F); |
|
1609 emit_byte(0x7F); |
|
1610 emit_operand(src, dst); |
|
1611 } |
|
1612 |
|
1613 void Assembler::movq(XMMRegister dst, Address src) { |
|
1614 InstructionMark im(this); |
|
1615 emit_byte(0xF3); |
|
1616 prefix(src, dst); |
|
1617 emit_byte(0x0F); |
|
1618 emit_byte(0x7E); |
|
1619 emit_operand(dst, src); |
|
1620 } |
|
1621 |
|
1622 void Assembler::movq(Address dst, XMMRegister src) { |
|
1623 InstructionMark im(this); |
|
1624 emit_byte(0x66); |
|
1625 prefix(dst, src); |
|
1626 emit_byte(0x0F); |
|
1627 emit_byte(0xD6); |
|
1628 emit_operand(src, dst); |
|
1629 } |
|
1630 |
|
1631 void Assembler::pshufd(XMMRegister dst, XMMRegister src, int mode) { |
|
1632 assert(isByte(mode), "invalid value"); |
|
1633 emit_byte(0x66); |
|
1634 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
1635 emit_byte(0x0F); |
|
1636 emit_byte(0x70); |
|
1637 emit_byte(0xC0 | encode); |
|
1638 emit_byte(mode & 0xFF); |
|
1639 } |
|
1640 |
|
1641 void Assembler::pshufd(XMMRegister dst, Address src, int mode) { |
|
1642 assert(isByte(mode), "invalid value"); |
|
1643 InstructionMark im(this); |
|
1644 emit_byte(0x66); |
|
1645 emit_byte(0x0F); |
|
1646 emit_byte(0x70); |
|
1647 emit_operand(dst, src); |
|
1648 emit_byte(mode & 0xFF); |
|
1649 } |
|
1650 |
|
1651 void Assembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) { |
|
1652 assert(isByte(mode), "invalid value"); |
|
1653 emit_byte(0xF2); |
|
1654 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
1655 emit_byte(0x0F); |
|
1656 emit_byte(0x70); |
|
1657 emit_byte(0xC0 | encode); |
|
1658 emit_byte(mode & 0xFF); |
|
1659 } |
|
1660 |
|
1661 void Assembler::pshuflw(XMMRegister dst, Address src, int mode) { |
|
1662 assert(isByte(mode), "invalid value"); |
|
1663 InstructionMark im(this); |
|
1664 emit_byte(0xF2); |
|
1665 emit_byte(0x0F); |
|
1666 emit_byte(0x70); |
|
1667 emit_operand(dst, src); |
|
1668 emit_byte(mode & 0xFF); |
|
1669 } |
|
1670 |
|
1671 void Assembler::cmovl(Condition cc, Register dst, Register src) { |
|
1672 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
1673 emit_byte(0x0F); |
|
1674 emit_byte(0x40 | cc); |
|
1675 emit_byte(0xC0 | encode); |
|
1676 } |
|
1677 |
|
1678 void Assembler::cmovl(Condition cc, Register dst, Address src) { |
|
1679 InstructionMark im(this); |
|
1680 prefix(src, dst); |
|
1681 emit_byte(0x0F); |
|
1682 emit_byte(0x40 | cc); |
|
1683 emit_operand(dst, src); |
|
1684 } |
|
1685 |
|
1686 void Assembler::cmovq(Condition cc, Register dst, Register src) { |
|
1687 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); |
|
1688 emit_byte(0x0F); |
|
1689 emit_byte(0x40 | cc); |
|
1690 emit_byte(0xC0 | encode); |
|
1691 } |
|
1692 |
|
1693 void Assembler::cmovq(Condition cc, Register dst, Address src) { |
|
1694 InstructionMark im(this); |
|
1695 prefixq(src, dst); |
|
1696 emit_byte(0x0F); |
|
1697 emit_byte(0x40 | cc); |
|
1698 emit_operand(dst, src); |
|
1699 } |
|
1700 |
|
1701 void Assembler::prefetch_prefix(Address src) { |
|
1702 prefix(src); |
|
1703 emit_byte(0x0F); |
|
1704 } |
|
1705 |
|
1706 void Assembler::prefetcht0(Address src) { |
|
1707 InstructionMark im(this); |
|
1708 prefetch_prefix(src); |
|
1709 emit_byte(0x18); |
|
1710 emit_operand(rcx, src); // 1, src |
|
1711 } |
|
1712 |
|
1713 void Assembler::prefetcht1(Address src) { |
|
1714 InstructionMark im(this); |
|
1715 prefetch_prefix(src); |
|
1716 emit_byte(0x18); |
|
1717 emit_operand(rdx, src); // 2, src |
|
1718 } |
|
1719 |
|
1720 void Assembler::prefetcht2(Address src) { |
|
1721 InstructionMark im(this); |
|
1722 prefetch_prefix(src); |
|
1723 emit_byte(0x18); |
|
1724 emit_operand(rbx, src); // 3, src |
|
1725 } |
|
1726 |
|
1727 void Assembler::prefetchnta(Address src) { |
|
1728 InstructionMark im(this); |
|
1729 prefetch_prefix(src); |
|
1730 emit_byte(0x18); |
|
1731 emit_operand(rax, src); // 0, src |
|
1732 } |
|
1733 |
|
1734 void Assembler::prefetchw(Address src) { |
|
1735 InstructionMark im(this); |
|
1736 prefetch_prefix(src); |
|
1737 emit_byte(0x0D); |
|
1738 emit_operand(rcx, src); // 1, src |
|
1739 } |
|
1740 |
|
1741 void Assembler::adcl(Register dst, int imm32) { |
|
1742 prefix(dst); |
|
1743 emit_arith(0x81, 0xD0, dst, imm32); |
|
1744 } |
|
1745 |
|
1746 void Assembler::adcl(Register dst, Address src) { |
|
1747 InstructionMark im(this); |
|
1748 prefix(src, dst); |
|
1749 emit_byte(0x13); |
|
1750 emit_operand(dst, src); |
|
1751 } |
|
1752 |
|
1753 void Assembler::adcl(Register dst, Register src) { |
|
1754 (void) prefix_and_encode(dst->encoding(), src->encoding()); |
|
1755 emit_arith(0x13, 0xC0, dst, src); |
|
1756 } |
|
1757 |
|
1758 void Assembler::adcq(Register dst, int imm32) { |
|
1759 (void) prefixq_and_encode(dst->encoding()); |
|
1760 emit_arith(0x81, 0xD0, dst, imm32); |
|
1761 } |
|
1762 |
|
1763 void Assembler::adcq(Register dst, Address src) { |
|
1764 InstructionMark im(this); |
|
1765 prefixq(src, dst); |
|
1766 emit_byte(0x13); |
|
1767 emit_operand(dst, src); |
|
1768 } |
|
1769 |
|
1770 void Assembler::adcq(Register dst, Register src) { |
|
1771 (int) prefixq_and_encode(dst->encoding(), src->encoding()); |
|
1772 emit_arith(0x13, 0xC0, dst, src); |
|
1773 } |
|
1774 |
|
1775 void Assembler::addl(Address dst, int imm32) { |
|
1776 InstructionMark im(this); |
|
1777 prefix(dst); |
|
1778 emit_arith_operand(0x81, rax, dst,imm32); |
|
1779 } |
|
1780 |
|
1781 void Assembler::addl(Address dst, Register src) { |
|
1782 InstructionMark im(this); |
|
1783 prefix(dst, src); |
|
1784 emit_byte(0x01); |
|
1785 emit_operand(src, dst); |
|
1786 } |
|
1787 |
|
1788 void Assembler::addl(Register dst, int imm32) { |
|
1789 prefix(dst); |
|
1790 emit_arith(0x81, 0xC0, dst, imm32); |
|
1791 } |
|
1792 |
|
1793 void Assembler::addl(Register dst, Address src) { |
|
1794 InstructionMark im(this); |
|
1795 prefix(src, dst); |
|
1796 emit_byte(0x03); |
|
1797 emit_operand(dst, src); |
|
1798 } |
|
1799 |
|
1800 void Assembler::addl(Register dst, Register src) { |
|
1801 (void) prefix_and_encode(dst->encoding(), src->encoding()); |
|
1802 emit_arith(0x03, 0xC0, dst, src); |
|
1803 } |
|
1804 |
|
1805 void Assembler::addq(Address dst, int imm32) { |
|
1806 InstructionMark im(this); |
|
1807 prefixq(dst); |
|
1808 emit_arith_operand(0x81, rax, dst,imm32); |
|
1809 } |
|
1810 |
|
1811 void Assembler::addq(Address dst, Register src) { |
|
1812 InstructionMark im(this); |
|
1813 prefixq(dst, src); |
|
1814 emit_byte(0x01); |
|
1815 emit_operand(src, dst); |
|
1816 } |
|
1817 |
|
1818 void Assembler::addq(Register dst, int imm32) { |
|
1819 (void) prefixq_and_encode(dst->encoding()); |
|
1820 emit_arith(0x81, 0xC0, dst, imm32); |
|
1821 } |
|
1822 |
|
1823 void Assembler::addq(Register dst, Address src) { |
|
1824 InstructionMark im(this); |
|
1825 prefixq(src, dst); |
|
1826 emit_byte(0x03); |
|
1827 emit_operand(dst, src); |
|
1828 } |
|
1829 |
|
1830 void Assembler::addq(Register dst, Register src) { |
|
1831 (void) prefixq_and_encode(dst->encoding(), src->encoding()); |
|
1832 emit_arith(0x03, 0xC0, dst, src); |
|
1833 } |
|
1834 |
|
1835 void Assembler::andl(Register dst, int imm32) { |
|
1836 prefix(dst); |
|
1837 emit_arith(0x81, 0xE0, dst, imm32); |
|
1838 } |
|
1839 |
|
1840 void Assembler::andl(Register dst, Address src) { |
|
1841 InstructionMark im(this); |
|
1842 prefix(src, dst); |
|
1843 emit_byte(0x23); |
|
1844 emit_operand(dst, src); |
|
1845 } |
|
1846 |
|
1847 void Assembler::andl(Register dst, Register src) { |
|
1848 (void) prefix_and_encode(dst->encoding(), src->encoding()); |
|
1849 emit_arith(0x23, 0xC0, dst, src); |
|
1850 } |
|
1851 |
|
1852 void Assembler::andq(Register dst, int imm32) { |
|
1853 (void) prefixq_and_encode(dst->encoding()); |
|
1854 emit_arith(0x81, 0xE0, dst, imm32); |
|
1855 } |
|
1856 |
|
1857 void Assembler::andq(Register dst, Address src) { |
|
1858 InstructionMark im(this); |
|
1859 prefixq(src, dst); |
|
1860 emit_byte(0x23); |
|
1861 emit_operand(dst, src); |
|
1862 } |
|
1863 |
|
1864 void Assembler::andq(Register dst, Register src) { |
|
1865 (int) prefixq_and_encode(dst->encoding(), src->encoding()); |
|
1866 emit_arith(0x23, 0xC0, dst, src); |
|
1867 } |
|
1868 |
|
1869 void Assembler::cmpb(Address dst, int imm8) { |
|
1870 InstructionMark im(this); |
|
1871 prefix(dst); |
|
1872 emit_byte(0x80); |
|
1873 emit_operand(rdi, dst, 1); |
|
1874 emit_byte(imm8); |
|
1875 } |
|
1876 |
|
1877 void Assembler::cmpl(Address dst, int imm32) { |
|
1878 InstructionMark im(this); |
|
1879 prefix(dst); |
|
1880 emit_byte(0x81); |
|
1881 emit_operand(rdi, dst, 4); |
|
1882 emit_long(imm32); |
|
1883 } |
|
1884 |
|
1885 void Assembler::cmpl(Register dst, int imm32) { |
|
1886 prefix(dst); |
|
1887 emit_arith(0x81, 0xF8, dst, imm32); |
|
1888 } |
|
1889 |
|
1890 void Assembler::cmpl(Register dst, Register src) { |
|
1891 (void) prefix_and_encode(dst->encoding(), src->encoding()); |
|
1892 emit_arith(0x3B, 0xC0, dst, src); |
|
1893 } |
|
1894 |
|
1895 void Assembler::cmpl(Register dst, Address src) { |
|
1896 InstructionMark im(this); |
|
1897 prefix(src, dst); |
|
1898 emit_byte(0x3B); |
|
1899 emit_operand(dst, src); |
|
1900 } |
|
1901 |
|
1902 void Assembler::cmpq(Address dst, int imm32) { |
|
1903 InstructionMark im(this); |
|
1904 prefixq(dst); |
|
1905 emit_byte(0x81); |
|
1906 emit_operand(rdi, dst, 4); |
|
1907 emit_long(imm32); |
|
1908 } |
|
1909 |
|
1910 void Assembler::cmpq(Register dst, int imm32) { |
|
1911 (void) prefixq_and_encode(dst->encoding()); |
|
1912 emit_arith(0x81, 0xF8, dst, imm32); |
|
1913 } |
|
1914 |
|
1915 void Assembler::cmpq(Address dst, Register src) { |
|
1916 prefixq(dst, src); |
|
1917 emit_byte(0x3B); |
|
1918 emit_operand(src, dst); |
|
1919 } |
|
1920 |
|
1921 void Assembler::cmpq(Register dst, Register src) { |
|
1922 (void) prefixq_and_encode(dst->encoding(), src->encoding()); |
|
1923 emit_arith(0x3B, 0xC0, dst, src); |
|
1924 } |
|
1925 |
|
1926 void Assembler::cmpq(Register dst, Address src) { |
|
1927 InstructionMark im(this); |
|
1928 prefixq(src, dst); |
|
1929 emit_byte(0x3B); |
|
1930 emit_operand(dst, src); |
|
1931 } |
|
1932 |
|
1933 void Assembler::ucomiss(XMMRegister dst, XMMRegister src) { |
|
1934 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
1935 emit_byte(0x0F); |
|
1936 emit_byte(0x2E); |
|
1937 emit_byte(0xC0 | encode); |
|
1938 } |
|
1939 |
|
1940 void Assembler::ucomisd(XMMRegister dst, XMMRegister src) { |
|
1941 emit_byte(0x66); |
|
1942 ucomiss(dst, src); |
|
1943 } |
|
1944 |
|
1945 void Assembler::decl(Register dst) { |
|
1946 // Don't use it directly. Use MacroAssembler::decrementl() instead. |
|
1947 // Use two-byte form (one-byte from is a REX prefix in 64-bit mode) |
|
1948 int encode = prefix_and_encode(dst->encoding()); |
|
1949 emit_byte(0xFF); |
|
1950 emit_byte(0xC8 | encode); |
|
1951 } |
|
1952 |
|
1953 void Assembler::decl(Address dst) { |
|
1954 // Don't use it directly. Use MacroAssembler::decrementl() instead. |
|
1955 InstructionMark im(this); |
|
1956 prefix(dst); |
|
1957 emit_byte(0xFF); |
|
1958 emit_operand(rcx, dst); |
|
1959 } |
|
1960 |
|
1961 void Assembler::decq(Register dst) { |
|
1962 // Don't use it directly. Use MacroAssembler::decrementq() instead. |
|
1963 // Use two-byte form (one-byte from is a REX prefix in 64-bit mode) |
|
1964 int encode = prefixq_and_encode(dst->encoding()); |
|
1965 emit_byte(0xFF); |
|
1966 emit_byte(0xC8 | encode); |
|
1967 } |
|
1968 |
|
1969 void Assembler::decq(Address dst) { |
|
1970 // Don't use it directly. Use MacroAssembler::decrementq() instead. |
|
1971 InstructionMark im(this); |
|
1972 prefixq(dst); |
|
1973 emit_byte(0xFF); |
|
1974 emit_operand(rcx, dst); |
|
1975 } |
|
1976 |
|
1977 void Assembler::idivl(Register src) { |
|
1978 int encode = prefix_and_encode(src->encoding()); |
|
1979 emit_byte(0xF7); |
|
1980 emit_byte(0xF8 | encode); |
|
1981 } |
|
1982 |
|
1983 void Assembler::idivq(Register src) { |
|
1984 int encode = prefixq_and_encode(src->encoding()); |
|
1985 emit_byte(0xF7); |
|
1986 emit_byte(0xF8 | encode); |
|
1987 } |
|
1988 |
|
1989 void Assembler::cdql() { |
|
1990 emit_byte(0x99); |
|
1991 } |
|
1992 |
|
1993 void Assembler::cdqq() { |
|
1994 prefix(REX_W); |
|
1995 emit_byte(0x99); |
|
1996 } |
|
1997 |
|
1998 void Assembler::imull(Register dst, Register src) { |
|
1999 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
2000 emit_byte(0x0F); |
|
2001 emit_byte(0xAF); |
|
2002 emit_byte(0xC0 | encode); |
|
2003 } |
|
2004 |
|
2005 void Assembler::imull(Register dst, Register src, int value) { |
|
2006 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
2007 if (is8bit(value)) { |
|
2008 emit_byte(0x6B); |
|
2009 emit_byte(0xC0 | encode); |
|
2010 emit_byte(value); |
|
2011 } else { |
|
2012 emit_byte(0x69); |
|
2013 emit_byte(0xC0 | encode); |
|
2014 emit_long(value); |
|
2015 } |
|
2016 } |
|
2017 |
|
2018 void Assembler::imulq(Register dst, Register src) { |
|
2019 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); |
|
2020 emit_byte(0x0F); |
|
2021 emit_byte(0xAF); |
|
2022 emit_byte(0xC0 | encode); |
|
2023 } |
|
2024 |
|
2025 void Assembler::imulq(Register dst, Register src, int value) { |
|
2026 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); |
|
2027 if (is8bit(value)) { |
|
2028 emit_byte(0x6B); |
|
2029 emit_byte(0xC0 | encode); |
|
2030 emit_byte(value); |
|
2031 } else { |
|
2032 emit_byte(0x69); |
|
2033 emit_byte(0xC0 | encode); |
|
2034 emit_long(value); |
|
2035 } |
|
2036 } |
|
2037 |
|
2038 void Assembler::incl(Register dst) { |
|
2039 // Don't use it directly. Use MacroAssembler::incrementl() instead. |
|
2040 // Use two-byte form (one-byte from is a REX prefix in 64-bit mode) |
|
2041 int encode = prefix_and_encode(dst->encoding()); |
|
2042 emit_byte(0xFF); |
|
2043 emit_byte(0xC0 | encode); |
|
2044 } |
|
2045 |
|
2046 void Assembler::incl(Address dst) { |
|
2047 // Don't use it directly. Use MacroAssembler::incrementl() instead. |
|
2048 InstructionMark im(this); |
|
2049 prefix(dst); |
|
2050 emit_byte(0xFF); |
|
2051 emit_operand(rax, dst); |
|
2052 } |
|
2053 |
|
2054 void Assembler::incq(Register dst) { |
|
2055 // Don't use it directly. Use MacroAssembler::incrementq() instead. |
|
2056 // Use two-byte form (one-byte from is a REX prefix in 64-bit mode) |
|
2057 int encode = prefixq_and_encode(dst->encoding()); |
|
2058 emit_byte(0xFF); |
|
2059 emit_byte(0xC0 | encode); |
|
2060 } |
|
2061 |
|
2062 void Assembler::incq(Address dst) { |
|
2063 // Don't use it directly. Use MacroAssembler::incrementq() instead. |
|
2064 InstructionMark im(this); |
|
2065 prefixq(dst); |
|
2066 emit_byte(0xFF); |
|
2067 emit_operand(rax, dst); |
|
2068 } |
|
2069 |
|
2070 void Assembler::leal(Register dst, Address src) { |
|
2071 InstructionMark im(this); |
|
2072 emit_byte(0x67); // addr32 |
|
2073 prefix(src, dst); |
|
2074 emit_byte(0x8D); |
|
2075 emit_operand(dst, src); |
|
2076 } |
|
2077 |
|
2078 void Assembler::leaq(Register dst, Address src) { |
|
2079 InstructionMark im(this); |
|
2080 prefixq(src, dst); |
|
2081 emit_byte(0x8D); |
|
2082 emit_operand(dst, src); |
|
2083 } |
|
2084 |
|
2085 void Assembler::mull(Address src) { |
|
2086 InstructionMark im(this); |
|
2087 // was missing |
|
2088 prefix(src); |
|
2089 emit_byte(0xF7); |
|
2090 emit_operand(rsp, src); |
|
2091 } |
|
2092 |
|
2093 void Assembler::mull(Register src) { |
|
2094 // was missing |
|
2095 int encode = prefix_and_encode(src->encoding()); |
|
2096 emit_byte(0xF7); |
|
2097 emit_byte(0xE0 | encode); |
|
2098 } |
|
2099 |
|
2100 void Assembler::negl(Register dst) { |
|
2101 int encode = prefix_and_encode(dst->encoding()); |
|
2102 emit_byte(0xF7); |
|
2103 emit_byte(0xD8 | encode); |
|
2104 } |
|
2105 |
|
2106 void Assembler::negq(Register dst) { |
|
2107 int encode = prefixq_and_encode(dst->encoding()); |
|
2108 emit_byte(0xF7); |
|
2109 emit_byte(0xD8 | encode); |
|
2110 } |
|
2111 |
|
2112 void Assembler::notl(Register dst) { |
|
2113 int encode = prefix_and_encode(dst->encoding()); |
|
2114 emit_byte(0xF7); |
|
2115 emit_byte(0xD0 | encode); |
|
2116 } |
|
2117 |
|
2118 void Assembler::notq(Register dst) { |
|
2119 int encode = prefixq_and_encode(dst->encoding()); |
|
2120 emit_byte(0xF7); |
|
2121 emit_byte(0xD0 | encode); |
|
2122 } |
|
2123 |
|
2124 void Assembler::orl(Address dst, int imm32) { |
|
2125 InstructionMark im(this); |
|
2126 prefix(dst); |
|
2127 emit_byte(0x81); |
|
2128 emit_operand(rcx, dst, 4); |
|
2129 emit_long(imm32); |
|
2130 } |
|
2131 |
|
2132 void Assembler::orl(Register dst, int imm32) { |
|
2133 prefix(dst); |
|
2134 emit_arith(0x81, 0xC8, dst, imm32); |
|
2135 } |
|
2136 |
|
2137 void Assembler::orl(Register dst, Address src) { |
|
2138 InstructionMark im(this); |
|
2139 prefix(src, dst); |
|
2140 emit_byte(0x0B); |
|
2141 emit_operand(dst, src); |
|
2142 } |
|
2143 |
|
2144 void Assembler::orl(Register dst, Register src) { |
|
2145 (void) prefix_and_encode(dst->encoding(), src->encoding()); |
|
2146 emit_arith(0x0B, 0xC0, dst, src); |
|
2147 } |
|
2148 |
|
2149 void Assembler::orq(Address dst, int imm32) { |
|
2150 InstructionMark im(this); |
|
2151 prefixq(dst); |
|
2152 emit_byte(0x81); |
|
2153 emit_operand(rcx, dst, 4); |
|
2154 emit_long(imm32); |
|
2155 } |
|
2156 |
|
2157 void Assembler::orq(Register dst, int imm32) { |
|
2158 (void) prefixq_and_encode(dst->encoding()); |
|
2159 emit_arith(0x81, 0xC8, dst, imm32); |
|
2160 } |
|
2161 |
|
2162 void Assembler::orq(Register dst, Address src) { |
|
2163 InstructionMark im(this); |
|
2164 prefixq(src, dst); |
|
2165 emit_byte(0x0B); |
|
2166 emit_operand(dst, src); |
|
2167 } |
|
2168 |
|
2169 void Assembler::orq(Register dst, Register src) { |
|
2170 (void) prefixq_and_encode(dst->encoding(), src->encoding()); |
|
2171 emit_arith(0x0B, 0xC0, dst, src); |
|
2172 } |
|
2173 |
|
2174 void Assembler::rcll(Register dst, int imm8) { |
|
2175 assert(isShiftCount(imm8), "illegal shift count"); |
|
2176 int encode = prefix_and_encode(dst->encoding()); |
|
2177 if (imm8 == 1) { |
|
2178 emit_byte(0xD1); |
|
2179 emit_byte(0xD0 | encode); |
|
2180 } else { |
|
2181 emit_byte(0xC1); |
|
2182 emit_byte(0xD0 | encode); |
|
2183 emit_byte(imm8); |
|
2184 } |
|
2185 } |
|
2186 |
|
2187 void Assembler::rclq(Register dst, int imm8) { |
|
2188 assert(isShiftCount(imm8 >> 1), "illegal shift count"); |
|
2189 int encode = prefixq_and_encode(dst->encoding()); |
|
2190 if (imm8 == 1) { |
|
2191 emit_byte(0xD1); |
|
2192 emit_byte(0xD0 | encode); |
|
2193 } else { |
|
2194 emit_byte(0xC1); |
|
2195 emit_byte(0xD0 | encode); |
|
2196 emit_byte(imm8); |
|
2197 } |
|
2198 } |
|
2199 |
|
2200 void Assembler::sarl(Register dst, int imm8) { |
|
2201 int encode = prefix_and_encode(dst->encoding()); |
|
2202 assert(isShiftCount(imm8), "illegal shift count"); |
|
2203 if (imm8 == 1) { |
|
2204 emit_byte(0xD1); |
|
2205 emit_byte(0xF8 | encode); |
|
2206 } else { |
|
2207 emit_byte(0xC1); |
|
2208 emit_byte(0xF8 | encode); |
|
2209 emit_byte(imm8); |
|
2210 } |
|
2211 } |
|
2212 |
|
2213 void Assembler::sarl(Register dst) { |
|
2214 int encode = prefix_and_encode(dst->encoding()); |
|
2215 emit_byte(0xD3); |
|
2216 emit_byte(0xF8 | encode); |
|
2217 } |
|
2218 |
|
2219 void Assembler::sarq(Register dst, int imm8) { |
|
2220 assert(isShiftCount(imm8 >> 1), "illegal shift count"); |
|
2221 int encode = prefixq_and_encode(dst->encoding()); |
|
2222 if (imm8 == 1) { |
|
2223 emit_byte(0xD1); |
|
2224 emit_byte(0xF8 | encode); |
|
2225 } else { |
|
2226 emit_byte(0xC1); |
|
2227 emit_byte(0xF8 | encode); |
|
2228 emit_byte(imm8); |
|
2229 } |
|
2230 } |
|
2231 |
|
2232 void Assembler::sarq(Register dst) { |
|
2233 int encode = prefixq_and_encode(dst->encoding()); |
|
2234 emit_byte(0xD3); |
|
2235 emit_byte(0xF8 | encode); |
|
2236 } |
|
2237 |
|
2238 void Assembler::sbbl(Address dst, int imm32) { |
|
2239 InstructionMark im(this); |
|
2240 prefix(dst); |
|
2241 emit_arith_operand(0x81, rbx, dst, imm32); |
|
2242 } |
|
2243 |
|
2244 void Assembler::sbbl(Register dst, int imm32) { |
|
2245 prefix(dst); |
|
2246 emit_arith(0x81, 0xD8, dst, imm32); |
|
2247 } |
|
2248 |
|
2249 void Assembler::sbbl(Register dst, Address src) { |
|
2250 InstructionMark im(this); |
|
2251 prefix(src, dst); |
|
2252 emit_byte(0x1B); |
|
2253 emit_operand(dst, src); |
|
2254 } |
|
2255 |
|
2256 void Assembler::sbbl(Register dst, Register src) { |
|
2257 (void) prefix_and_encode(dst->encoding(), src->encoding()); |
|
2258 emit_arith(0x1B, 0xC0, dst, src); |
|
2259 } |
|
2260 |
|
2261 void Assembler::sbbq(Address dst, int imm32) { |
|
2262 InstructionMark im(this); |
|
2263 prefixq(dst); |
|
2264 emit_arith_operand(0x81, rbx, dst, imm32); |
|
2265 } |
|
2266 |
|
2267 void Assembler::sbbq(Register dst, int imm32) { |
|
2268 (void) prefixq_and_encode(dst->encoding()); |
|
2269 emit_arith(0x81, 0xD8, dst, imm32); |
|
2270 } |
|
2271 |
|
2272 void Assembler::sbbq(Register dst, Address src) { |
|
2273 InstructionMark im(this); |
|
2274 prefixq(src, dst); |
|
2275 emit_byte(0x1B); |
|
2276 emit_operand(dst, src); |
|
2277 } |
|
2278 |
|
2279 void Assembler::sbbq(Register dst, Register src) { |
|
2280 (void) prefixq_and_encode(dst->encoding(), src->encoding()); |
|
2281 emit_arith(0x1B, 0xC0, dst, src); |
|
2282 } |
|
2283 |
|
2284 void Assembler::shll(Register dst, int imm8) { |
|
2285 assert(isShiftCount(imm8), "illegal shift count"); |
|
2286 int encode = prefix_and_encode(dst->encoding()); |
|
2287 if (imm8 == 1 ) { |
|
2288 emit_byte(0xD1); |
|
2289 emit_byte(0xE0 | encode); |
|
2290 } else { |
|
2291 emit_byte(0xC1); |
|
2292 emit_byte(0xE0 | encode); |
|
2293 emit_byte(imm8); |
|
2294 } |
|
2295 } |
|
2296 |
|
2297 void Assembler::shll(Register dst) { |
|
2298 int encode = prefix_and_encode(dst->encoding()); |
|
2299 emit_byte(0xD3); |
|
2300 emit_byte(0xE0 | encode); |
|
2301 } |
|
2302 |
|
2303 void Assembler::shlq(Register dst, int imm8) { |
|
2304 assert(isShiftCount(imm8 >> 1), "illegal shift count"); |
|
2305 int encode = prefixq_and_encode(dst->encoding()); |
|
2306 if (imm8 == 1) { |
|
2307 emit_byte(0xD1); |
|
2308 emit_byte(0xE0 | encode); |
|
2309 } else { |
|
2310 emit_byte(0xC1); |
|
2311 emit_byte(0xE0 | encode); |
|
2312 emit_byte(imm8); |
|
2313 } |
|
2314 } |
|
2315 |
|
2316 void Assembler::shlq(Register dst) { |
|
2317 int encode = prefixq_and_encode(dst->encoding()); |
|
2318 emit_byte(0xD3); |
|
2319 emit_byte(0xE0 | encode); |
|
2320 } |
|
2321 |
|
2322 void Assembler::shrl(Register dst, int imm8) { |
|
2323 assert(isShiftCount(imm8), "illegal shift count"); |
|
2324 int encode = prefix_and_encode(dst->encoding()); |
|
2325 emit_byte(0xC1); |
|
2326 emit_byte(0xE8 | encode); |
|
2327 emit_byte(imm8); |
|
2328 } |
|
2329 |
|
2330 void Assembler::shrl(Register dst) { |
|
2331 int encode = prefix_and_encode(dst->encoding()); |
|
2332 emit_byte(0xD3); |
|
2333 emit_byte(0xE8 | encode); |
|
2334 } |
|
2335 |
|
2336 void Assembler::shrq(Register dst, int imm8) { |
|
2337 assert(isShiftCount(imm8 >> 1), "illegal shift count"); |
|
2338 int encode = prefixq_and_encode(dst->encoding()); |
|
2339 emit_byte(0xC1); |
|
2340 emit_byte(0xE8 | encode); |
|
2341 emit_byte(imm8); |
|
2342 } |
|
2343 |
|
2344 void Assembler::shrq(Register dst) { |
|
2345 int encode = prefixq_and_encode(dst->encoding()); |
|
2346 emit_byte(0xD3); |
|
2347 emit_byte(0xE8 | encode); |
|
2348 } |
|
2349 |
|
2350 void Assembler::subl(Address dst, int imm32) { |
|
2351 InstructionMark im(this); |
|
2352 prefix(dst); |
|
2353 if (is8bit(imm32)) { |
|
2354 emit_byte(0x83); |
|
2355 emit_operand(rbp, dst, 1); |
|
2356 emit_byte(imm32 & 0xFF); |
|
2357 } else { |
|
2358 emit_byte(0x81); |
|
2359 emit_operand(rbp, dst, 4); |
|
2360 emit_long(imm32); |
|
2361 } |
|
2362 } |
|
2363 |
|
2364 void Assembler::subl(Register dst, int imm32) { |
|
2365 prefix(dst); |
|
2366 emit_arith(0x81, 0xE8, dst, imm32); |
|
2367 } |
|
2368 |
|
2369 void Assembler::subl(Address dst, Register src) { |
|
2370 InstructionMark im(this); |
|
2371 prefix(dst, src); |
|
2372 emit_byte(0x29); |
|
2373 emit_operand(src, dst); |
|
2374 } |
|
2375 |
|
2376 void Assembler::subl(Register dst, Address src) { |
|
2377 InstructionMark im(this); |
|
2378 prefix(src, dst); |
|
2379 emit_byte(0x2B); |
|
2380 emit_operand(dst, src); |
|
2381 } |
|
2382 |
|
2383 void Assembler::subl(Register dst, Register src) { |
|
2384 (void) prefix_and_encode(dst->encoding(), src->encoding()); |
|
2385 emit_arith(0x2B, 0xC0, dst, src); |
|
2386 } |
|
2387 |
|
2388 void Assembler::subq(Address dst, int imm32) { |
|
2389 InstructionMark im(this); |
|
2390 prefixq(dst); |
|
2391 if (is8bit(imm32)) { |
|
2392 emit_byte(0x83); |
|
2393 emit_operand(rbp, dst, 1); |
|
2394 emit_byte(imm32 & 0xFF); |
|
2395 } else { |
|
2396 emit_byte(0x81); |
|
2397 emit_operand(rbp, dst, 4); |
|
2398 emit_long(imm32); |
|
2399 } |
|
2400 } |
|
2401 |
|
2402 void Assembler::subq(Register dst, int imm32) { |
|
2403 (void) prefixq_and_encode(dst->encoding()); |
|
2404 emit_arith(0x81, 0xE8, dst, imm32); |
|
2405 } |
|
2406 |
|
2407 void Assembler::subq(Address dst, Register src) { |
|
2408 InstructionMark im(this); |
|
2409 prefixq(dst, src); |
|
2410 emit_byte(0x29); |
|
2411 emit_operand(src, dst); |
|
2412 } |
|
2413 |
|
2414 void Assembler::subq(Register dst, Address src) { |
|
2415 InstructionMark im(this); |
|
2416 prefixq(src, dst); |
|
2417 emit_byte(0x2B); |
|
2418 emit_operand(dst, src); |
|
2419 } |
|
2420 |
|
2421 void Assembler::subq(Register dst, Register src) { |
|
2422 (void) prefixq_and_encode(dst->encoding(), src->encoding()); |
|
2423 emit_arith(0x2B, 0xC0, dst, src); |
|
2424 } |
|
2425 |
|
2426 void Assembler::testb(Register dst, int imm8) { |
|
2427 (void) prefix_and_encode(dst->encoding(), true); |
|
2428 emit_arith_b(0xF6, 0xC0, dst, imm8); |
|
2429 } |
|
2430 |
|
2431 void Assembler::testl(Register dst, int imm32) { |
|
2432 // not using emit_arith because test |
|
2433 // doesn't support sign-extension of |
|
2434 // 8bit operands |
|
2435 int encode = dst->encoding(); |
|
2436 if (encode == 0) { |
|
2437 emit_byte(0xA9); |
|
2438 } else { |
|
2439 encode = prefix_and_encode(encode); |
|
2440 emit_byte(0xF7); |
|
2441 emit_byte(0xC0 | encode); |
|
2442 } |
|
2443 emit_long(imm32); |
|
2444 } |
|
2445 |
|
2446 void Assembler::testl(Register dst, Register src) { |
|
2447 (void) prefix_and_encode(dst->encoding(), src->encoding()); |
|
2448 emit_arith(0x85, 0xC0, dst, src); |
|
2449 } |
|
2450 |
|
2451 void Assembler::testq(Register dst, int imm32) { |
|
2452 // not using emit_arith because test |
|
2453 // doesn't support sign-extension of |
|
2454 // 8bit operands |
|
2455 int encode = dst->encoding(); |
|
2456 if (encode == 0) { |
|
2457 prefix(REX_W); |
|
2458 emit_byte(0xA9); |
|
2459 } else { |
|
2460 encode = prefixq_and_encode(encode); |
|
2461 emit_byte(0xF7); |
|
2462 emit_byte(0xC0 | encode); |
|
2463 } |
|
2464 emit_long(imm32); |
|
2465 } |
|
2466 |
|
2467 void Assembler::testq(Register dst, Register src) { |
|
2468 (void) prefixq_and_encode(dst->encoding(), src->encoding()); |
|
2469 emit_arith(0x85, 0xC0, dst, src); |
|
2470 } |
|
2471 |
|
2472 void Assembler::xaddl(Address dst, Register src) { |
|
2473 InstructionMark im(this); |
|
2474 prefix(dst, src); |
|
2475 emit_byte(0x0F); |
|
2476 emit_byte(0xC1); |
|
2477 emit_operand(src, dst); |
|
2478 } |
|
2479 |
|
2480 void Assembler::xaddq(Address dst, Register src) { |
|
2481 InstructionMark im(this); |
|
2482 prefixq(dst, src); |
|
2483 emit_byte(0x0F); |
|
2484 emit_byte(0xC1); |
|
2485 emit_operand(src, dst); |
|
2486 } |
|
2487 |
|
2488 void Assembler::xorl(Register dst, int imm32) { |
|
2489 prefix(dst); |
|
2490 emit_arith(0x81, 0xF0, dst, imm32); |
|
2491 } |
|
2492 |
|
2493 void Assembler::xorl(Register dst, Register src) { |
|
2494 (void) prefix_and_encode(dst->encoding(), src->encoding()); |
|
2495 emit_arith(0x33, 0xC0, dst, src); |
|
2496 } |
|
2497 |
|
2498 void Assembler::xorl(Register dst, Address src) { |
|
2499 InstructionMark im(this); |
|
2500 prefix(src, dst); |
|
2501 emit_byte(0x33); |
|
2502 emit_operand(dst, src); |
|
2503 } |
|
2504 |
|
2505 void Assembler::xorq(Register dst, int imm32) { |
|
2506 (void) prefixq_and_encode(dst->encoding()); |
|
2507 emit_arith(0x81, 0xF0, dst, imm32); |
|
2508 } |
|
2509 |
|
2510 void Assembler::xorq(Register dst, Register src) { |
|
2511 (void) prefixq_and_encode(dst->encoding(), src->encoding()); |
|
2512 emit_arith(0x33, 0xC0, dst, src); |
|
2513 } |
|
2514 |
|
2515 void Assembler::xorq(Register dst, Address src) { |
|
2516 InstructionMark im(this); |
|
2517 prefixq(src, dst); |
|
2518 emit_byte(0x33); |
|
2519 emit_operand(dst, src); |
|
2520 } |
|
2521 |
|
2522 void Assembler::bswapl(Register reg) { |
|
2523 int encode = prefix_and_encode(reg->encoding()); |
|
2524 emit_byte(0x0F); |
|
2525 emit_byte(0xC8 | encode); |
|
2526 } |
|
2527 |
|
2528 void Assembler::bswapq(Register reg) { |
|
2529 int encode = prefixq_and_encode(reg->encoding()); |
|
2530 emit_byte(0x0F); |
|
2531 emit_byte(0xC8 | encode); |
|
2532 } |
|
2533 |
|
2534 void Assembler::lock() { |
|
2535 emit_byte(0xF0); |
|
2536 } |
|
2537 |
|
2538 void Assembler::xchgl(Register dst, Address src) { |
|
2539 InstructionMark im(this); |
|
2540 prefix(src, dst); |
|
2541 emit_byte(0x87); |
|
2542 emit_operand(dst, src); |
|
2543 } |
|
2544 |
|
2545 void Assembler::xchgl(Register dst, Register src) { |
|
2546 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
2547 emit_byte(0x87); |
|
2548 emit_byte(0xc0 | encode); |
|
2549 } |
|
2550 |
|
2551 void Assembler::xchgq(Register dst, Address src) { |
|
2552 InstructionMark im(this); |
|
2553 prefixq(src, dst); |
|
2554 emit_byte(0x87); |
|
2555 emit_operand(dst, src); |
|
2556 } |
|
2557 |
|
2558 void Assembler::xchgq(Register dst, Register src) { |
|
2559 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); |
|
2560 emit_byte(0x87); |
|
2561 emit_byte(0xc0 | encode); |
|
2562 } |
|
2563 |
|
2564 void Assembler::cmpxchgl(Register reg, Address adr) { |
|
2565 InstructionMark im(this); |
|
2566 prefix(adr, reg); |
|
2567 emit_byte(0x0F); |
|
2568 emit_byte(0xB1); |
|
2569 emit_operand(reg, adr); |
|
2570 } |
|
2571 |
|
2572 void Assembler::cmpxchgq(Register reg, Address adr) { |
|
2573 InstructionMark im(this); |
|
2574 prefixq(adr, reg); |
|
2575 emit_byte(0x0F); |
|
2576 emit_byte(0xB1); |
|
2577 emit_operand(reg, adr); |
|
2578 } |
|
2579 |
|
2580 void Assembler::hlt() { |
|
2581 emit_byte(0xF4); |
|
2582 } |
|
2583 |
|
2584 |
|
2585 void Assembler::addr_nop_4() { |
|
2586 // 4 bytes: NOP DWORD PTR [EAX+0] |
|
2587 emit_byte(0x0F); |
|
2588 emit_byte(0x1F); |
|
2589 emit_byte(0x40); // emit_rm(cbuf, 0x1, EAX_enc, EAX_enc); |
|
2590 emit_byte(0); // 8-bits offset (1 byte) |
|
2591 } |
|
2592 |
|
2593 void Assembler::addr_nop_5() { |
|
2594 // 5 bytes: NOP DWORD PTR [EAX+EAX*0+0] 8-bits offset |
|
2595 emit_byte(0x0F); |
|
2596 emit_byte(0x1F); |
|
2597 emit_byte(0x44); // emit_rm(cbuf, 0x1, EAX_enc, 0x4); |
|
2598 emit_byte(0x00); // emit_rm(cbuf, 0x0, EAX_enc, EAX_enc); |
|
2599 emit_byte(0); // 8-bits offset (1 byte) |
|
2600 } |
|
2601 |
|
2602 void Assembler::addr_nop_7() { |
|
2603 // 7 bytes: NOP DWORD PTR [EAX+0] 32-bits offset |
|
2604 emit_byte(0x0F); |
|
2605 emit_byte(0x1F); |
|
2606 emit_byte(0x80); // emit_rm(cbuf, 0x2, EAX_enc, EAX_enc); |
|
2607 emit_long(0); // 32-bits offset (4 bytes) |
|
2608 } |
|
2609 |
|
2610 void Assembler::addr_nop_8() { |
|
2611 // 8 bytes: NOP DWORD PTR [EAX+EAX*0+0] 32-bits offset |
|
2612 emit_byte(0x0F); |
|
2613 emit_byte(0x1F); |
|
2614 emit_byte(0x84); // emit_rm(cbuf, 0x2, EAX_enc, 0x4); |
|
2615 emit_byte(0x00); // emit_rm(cbuf, 0x0, EAX_enc, EAX_enc); |
|
2616 emit_long(0); // 32-bits offset (4 bytes) |
|
2617 } |
|
2618 |
|
2619 void Assembler::nop(int i) { |
|
2620 assert(i > 0, " "); |
|
2621 if (UseAddressNop && VM_Version::is_intel()) { |
|
2622 // |
|
2623 // Using multi-bytes nops "0x0F 0x1F [address]" for Intel |
|
2624 // 1: 0x90 |
|
2625 // 2: 0x66 0x90 |
|
2626 // 3: 0x66 0x66 0x90 (don't use "0x0F 0x1F 0x00" - need patching safe padding) |
|
2627 // 4: 0x0F 0x1F 0x40 0x00 |
|
2628 // 5: 0x0F 0x1F 0x44 0x00 0x00 |
|
2629 // 6: 0x66 0x0F 0x1F 0x44 0x00 0x00 |
|
2630 // 7: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 |
|
2631 // 8: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 |
|
2632 // 9: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 |
|
2633 // 10: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 |
|
2634 // 11: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 |
|
2635 |
|
2636 // The rest coding is Intel specific - don't use consecutive address nops |
|
2637 |
|
2638 // 12: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90 |
|
2639 // 13: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90 |
|
2640 // 14: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90 |
|
2641 // 15: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90 |
|
2642 |
|
2643 while(i >= 15) { |
|
2644 // For Intel don't generate consecutive addess nops (mix with regular nops) |
|
2645 i -= 15; |
|
2646 emit_byte(0x66); // size prefix |
|
2647 emit_byte(0x66); // size prefix |
|
2648 emit_byte(0x66); // size prefix |
|
2649 addr_nop_8(); |
|
2650 emit_byte(0x66); // size prefix |
|
2651 emit_byte(0x66); // size prefix |
|
2652 emit_byte(0x66); // size prefix |
|
2653 emit_byte(0x90); // nop |
|
2654 } |
|
2655 switch (i) { |
|
2656 case 14: |
|
2657 emit_byte(0x66); // size prefix |
|
2658 case 13: |
|
2659 emit_byte(0x66); // size prefix |
|
2660 case 12: |
|
2661 addr_nop_8(); |
|
2662 emit_byte(0x66); // size prefix |
|
2663 emit_byte(0x66); // size prefix |
|
2664 emit_byte(0x66); // size prefix |
|
2665 emit_byte(0x90); // nop |
|
2666 break; |
|
2667 case 11: |
|
2668 emit_byte(0x66); // size prefix |
|
2669 case 10: |
|
2670 emit_byte(0x66); // size prefix |
|
2671 case 9: |
|
2672 emit_byte(0x66); // size prefix |
|
2673 case 8: |
|
2674 addr_nop_8(); |
|
2675 break; |
|
2676 case 7: |
|
2677 addr_nop_7(); |
|
2678 break; |
|
2679 case 6: |
|
2680 emit_byte(0x66); // size prefix |
|
2681 case 5: |
|
2682 addr_nop_5(); |
|
2683 break; |
|
2684 case 4: |
|
2685 addr_nop_4(); |
|
2686 break; |
|
2687 case 3: |
|
2688 // Don't use "0x0F 0x1F 0x00" - need patching safe padding |
|
2689 emit_byte(0x66); // size prefix |
|
2690 case 2: |
|
2691 emit_byte(0x66); // size prefix |
|
2692 case 1: |
|
2693 emit_byte(0x90); // nop |
|
2694 break; |
|
2695 default: |
|
2696 assert(i == 0, " "); |
|
2697 } |
|
2698 return; |
|
2699 } |
|
2700 if (UseAddressNop && VM_Version::is_amd()) { |
|
2701 // |
|
2702 // Using multi-bytes nops "0x0F 0x1F [address]" for AMD. |
|
2703 // 1: 0x90 |
|
2704 // 2: 0x66 0x90 |
|
2705 // 3: 0x66 0x66 0x90 (don't use "0x0F 0x1F 0x00" - need patching safe padding) |
|
2706 // 4: 0x0F 0x1F 0x40 0x00 |
|
2707 // 5: 0x0F 0x1F 0x44 0x00 0x00 |
|
2708 // 6: 0x66 0x0F 0x1F 0x44 0x00 0x00 |
|
2709 // 7: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 |
|
2710 // 8: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 |
|
2711 // 9: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 |
|
2712 // 10: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 |
|
2713 // 11: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 |
|
2714 |
|
2715 // The rest coding is AMD specific - use consecutive address nops |
|
2716 |
|
2717 // 12: 0x66 0x0F 0x1F 0x44 0x00 0x00 0x66 0x0F 0x1F 0x44 0x00 0x00 |
|
2718 // 13: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 0x66 0x0F 0x1F 0x44 0x00 0x00 |
|
2719 // 14: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 |
|
2720 // 15: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 |
|
2721 // 16: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 |
|
2722 // Size prefixes (0x66) are added for larger sizes |
|
2723 |
|
2724 while(i >= 22) { |
|
2725 i -= 11; |
|
2726 emit_byte(0x66); // size prefix |
|
2727 emit_byte(0x66); // size prefix |
|
2728 emit_byte(0x66); // size prefix |
|
2729 addr_nop_8(); |
|
2730 } |
|
2731 // Generate first nop for size between 21-12 |
|
2732 switch (i) { |
|
2733 case 21: |
|
2734 i -= 1; |
|
2735 emit_byte(0x66); // size prefix |
|
2736 case 20: |
|
2737 case 19: |
|
2738 i -= 1; |
|
2739 emit_byte(0x66); // size prefix |
|
2740 case 18: |
|
2741 case 17: |
|
2742 i -= 1; |
|
2743 emit_byte(0x66); // size prefix |
|
2744 case 16: |
|
2745 case 15: |
|
2746 i -= 8; |
|
2747 addr_nop_8(); |
|
2748 break; |
|
2749 case 14: |
|
2750 case 13: |
|
2751 i -= 7; |
|
2752 addr_nop_7(); |
|
2753 break; |
|
2754 case 12: |
|
2755 i -= 6; |
|
2756 emit_byte(0x66); // size prefix |
|
2757 addr_nop_5(); |
|
2758 break; |
|
2759 default: |
|
2760 assert(i < 12, " "); |
|
2761 } |
|
2762 |
|
2763 // Generate second nop for size between 11-1 |
|
2764 switch (i) { |
|
2765 case 11: |
|
2766 emit_byte(0x66); // size prefix |
|
2767 case 10: |
|
2768 emit_byte(0x66); // size prefix |
|
2769 case 9: |
|
2770 emit_byte(0x66); // size prefix |
|
2771 case 8: |
|
2772 addr_nop_8(); |
|
2773 break; |
|
2774 case 7: |
|
2775 addr_nop_7(); |
|
2776 break; |
|
2777 case 6: |
|
2778 emit_byte(0x66); // size prefix |
|
2779 case 5: |
|
2780 addr_nop_5(); |
|
2781 break; |
|
2782 case 4: |
|
2783 addr_nop_4(); |
|
2784 break; |
|
2785 case 3: |
|
2786 // Don't use "0x0F 0x1F 0x00" - need patching safe padding |
|
2787 emit_byte(0x66); // size prefix |
|
2788 case 2: |
|
2789 emit_byte(0x66); // size prefix |
|
2790 case 1: |
|
2791 emit_byte(0x90); // nop |
|
2792 break; |
|
2793 default: |
|
2794 assert(i == 0, " "); |
|
2795 } |
|
2796 return; |
|
2797 } |
|
2798 |
|
2799 // Using nops with size prefixes "0x66 0x90". |
|
2800 // From AMD Optimization Guide: |
|
2801 // 1: 0x90 |
|
2802 // 2: 0x66 0x90 |
|
2803 // 3: 0x66 0x66 0x90 |
|
2804 // 4: 0x66 0x66 0x66 0x90 |
|
2805 // 5: 0x66 0x66 0x90 0x66 0x90 |
|
2806 // 6: 0x66 0x66 0x90 0x66 0x66 0x90 |
|
2807 // 7: 0x66 0x66 0x66 0x90 0x66 0x66 0x90 |
|
2808 // 8: 0x66 0x66 0x66 0x90 0x66 0x66 0x66 0x90 |
|
2809 // 9: 0x66 0x66 0x90 0x66 0x66 0x90 0x66 0x66 0x90 |
|
2810 // 10: 0x66 0x66 0x66 0x90 0x66 0x66 0x90 0x66 0x66 0x90 |
|
2811 // |
|
2812 while(i > 12) { |
|
2813 i -= 4; |
|
2814 emit_byte(0x66); // size prefix |
|
2815 emit_byte(0x66); |
|
2816 emit_byte(0x66); |
|
2817 emit_byte(0x90); // nop |
|
2818 } |
|
2819 // 1 - 12 nops |
|
2820 if(i > 8) { |
|
2821 if(i > 9) { |
|
2822 i -= 1; |
|
2823 emit_byte(0x66); |
|
2824 } |
|
2825 i -= 3; |
|
2826 emit_byte(0x66); |
|
2827 emit_byte(0x66); |
|
2828 emit_byte(0x90); |
|
2829 } |
|
2830 // 1 - 8 nops |
|
2831 if(i > 4) { |
|
2832 if(i > 6) { |
|
2833 i -= 1; |
|
2834 emit_byte(0x66); |
|
2835 } |
|
2836 i -= 3; |
|
2837 emit_byte(0x66); |
|
2838 emit_byte(0x66); |
|
2839 emit_byte(0x90); |
|
2840 } |
|
2841 switch (i) { |
|
2842 case 4: |
|
2843 emit_byte(0x66); |
|
2844 case 3: |
|
2845 emit_byte(0x66); |
|
2846 case 2: |
|
2847 emit_byte(0x66); |
|
2848 case 1: |
|
2849 emit_byte(0x90); |
|
2850 break; |
|
2851 default: |
|
2852 assert(i == 0, " "); |
|
2853 } |
|
2854 } |
|
2855 |
|
2856 void Assembler::ret(int imm16) { |
|
2857 if (imm16 == 0) { |
|
2858 emit_byte(0xC3); |
|
2859 } else { |
|
2860 emit_byte(0xC2); |
|
2861 emit_word(imm16); |
|
2862 } |
|
2863 } |
|
2864 |
|
2865 // copies a single word from [esi] to [edi] |
|
2866 void Assembler::smovl() { |
|
2867 emit_byte(0xA5); |
|
2868 } |
|
2869 |
|
2870 // copies data from [rsi] to [rdi] using rcx words (m32) |
|
2871 void Assembler::rep_movl() { |
|
2872 // REP |
|
2873 emit_byte(0xF3); |
|
2874 // MOVSL |
|
2875 emit_byte(0xA5); |
|
2876 } |
|
2877 |
|
2878 // copies data from [rsi] to [rdi] using rcx double words (m64) |
|
2879 void Assembler::rep_movq() { |
|
2880 // REP |
|
2881 emit_byte(0xF3); |
|
2882 // MOVSQ |
|
2883 prefix(REX_W); |
|
2884 emit_byte(0xA5); |
|
2885 } |
|
2886 |
|
2887 // sets rcx double words (m64) with rax value at [rdi] |
|
2888 void Assembler::rep_set() { |
|
2889 // REP |
|
2890 emit_byte(0xF3); |
|
2891 // STOSQ |
|
2892 prefix(REX_W); |
|
2893 emit_byte(0xAB); |
|
2894 } |
|
2895 |
|
2896 // scans rcx double words (m64) at [rdi] for occurance of rax |
|
2897 void Assembler::repne_scanq() { |
|
2898 // REPNE/REPNZ |
|
2899 emit_byte(0xF2); |
|
2900 // SCASQ |
|
2901 prefix(REX_W); |
|
2902 emit_byte(0xAF); |
|
2903 } |
|
2904 |
|
2905 void Assembler::repne_scanl() { |
|
2906 // REPNE/REPNZ |
|
2907 emit_byte(0xF2); |
|
2908 // SCASL |
|
2909 emit_byte(0xAF); |
|
2910 } |
|
2911 |
|
2912 |
|
2913 void Assembler::setb(Condition cc, Register dst) { |
|
2914 assert(0 <= cc && cc < 16, "illegal cc"); |
|
2915 int encode = prefix_and_encode(dst->encoding(), true); |
|
2916 emit_byte(0x0F); |
|
2917 emit_byte(0x90 | cc); |
|
2918 emit_byte(0xC0 | encode); |
|
2919 } |
|
2920 |
|
2921 void Assembler::clflush(Address adr) { |
|
2922 prefix(adr); |
|
2923 emit_byte(0x0F); |
|
2924 emit_byte(0xAE); |
|
2925 emit_operand(rdi, adr); |
|
2926 } |
|
2927 |
|
2928 void Assembler::call(Label& L, relocInfo::relocType rtype) { |
|
2929 if (L.is_bound()) { |
|
2930 const int long_size = 5; |
|
2931 int offs = (int)( target(L) - pc() ); |
|
2932 assert(offs <= 0, "assembler error"); |
|
2933 InstructionMark im(this); |
|
2934 // 1110 1000 #32-bit disp |
|
2935 emit_byte(0xE8); |
|
2936 emit_data(offs - long_size, rtype, disp32_operand); |
|
2937 } else { |
|
2938 InstructionMark im(this); |
|
2939 // 1110 1000 #32-bit disp |
|
2940 L.add_patch_at(code(), locator()); |
|
2941 |
|
2942 emit_byte(0xE8); |
|
2943 emit_data(int(0), rtype, disp32_operand); |
|
2944 } |
|
2945 } |
|
2946 |
|
2947 void Assembler::call_literal(address entry, RelocationHolder const& rspec) { |
|
2948 assert(entry != NULL, "call most probably wrong"); |
|
2949 InstructionMark im(this); |
|
2950 emit_byte(0xE8); |
|
2951 intptr_t disp = entry - (_code_pos + sizeof(int32_t)); |
|
2952 assert(is_simm32(disp), "must be 32bit offset (call2)"); |
|
2953 // Technically, should use call32_operand, but this format is |
|
2954 // implied by the fact that we're emitting a call instruction. |
|
2955 emit_data((int) disp, rspec, disp32_operand); |
|
2956 } |
|
2957 |
|
2958 |
|
2959 void Assembler::call(Register dst) { |
|
2960 // This was originally using a 32bit register encoding |
|
2961 // and surely we want 64bit! |
|
2962 // this is a 32bit encoding but in 64bit mode the default |
|
2963 // operand size is 64bit so there is no need for the |
|
2964 // wide prefix. So prefix only happens if we use the |
|
2965 // new registers. Much like push/pop. |
|
2966 int encode = prefixq_and_encode(dst->encoding()); |
|
2967 emit_byte(0xFF); |
|
2968 emit_byte(0xD0 | encode); |
|
2969 } |
|
2970 |
|
2971 void Assembler::call(Address adr) { |
|
2972 InstructionMark im(this); |
|
2973 prefix(adr); |
|
2974 emit_byte(0xFF); |
|
2975 emit_operand(rdx, adr); |
|
2976 } |
|
2977 |
|
2978 void Assembler::jmp(Register reg) { |
|
2979 int encode = prefix_and_encode(reg->encoding()); |
|
2980 emit_byte(0xFF); |
|
2981 emit_byte(0xE0 | encode); |
|
2982 } |
|
2983 |
|
2984 void Assembler::jmp(Address adr) { |
|
2985 InstructionMark im(this); |
|
2986 prefix(adr); |
|
2987 emit_byte(0xFF); |
|
2988 emit_operand(rsp, adr); |
|
2989 } |
|
2990 |
|
2991 void Assembler::jmp_literal(address dest, RelocationHolder const& rspec) { |
|
2992 InstructionMark im(this); |
|
2993 emit_byte(0xE9); |
|
2994 assert(dest != NULL, "must have a target"); |
|
2995 intptr_t disp = dest - (_code_pos + sizeof(int32_t)); |
|
2996 assert(is_simm32(disp), "must be 32bit offset (jmp)"); |
|
2997 emit_data(disp, rspec.reloc(), call32_operand); |
|
2998 } |
|
2999 |
|
3000 void Assembler::jmp(Label& L, relocInfo::relocType rtype) { |
|
3001 if (L.is_bound()) { |
|
3002 address entry = target(L); |
|
3003 assert(entry != NULL, "jmp most probably wrong"); |
|
3004 InstructionMark im(this); |
|
3005 const int short_size = 2; |
|
3006 const int long_size = 5; |
|
3007 intptr_t offs = entry - _code_pos; |
|
3008 if (rtype == relocInfo::none && is8bit(offs - short_size)) { |
|
3009 emit_byte(0xEB); |
|
3010 emit_byte((offs - short_size) & 0xFF); |
|
3011 } else { |
|
3012 emit_byte(0xE9); |
|
3013 emit_long(offs - long_size); |
|
3014 } |
|
3015 } else { |
|
3016 // By default, forward jumps are always 32-bit displacements, since |
|
3017 // we can't yet know where the label will be bound. If you're sure that |
|
3018 // the forward jump will not run beyond 256 bytes, use jmpb to |
|
3019 // force an 8-bit displacement. |
|
3020 InstructionMark im(this); |
|
3021 relocate(rtype); |
|
3022 L.add_patch_at(code(), locator()); |
|
3023 emit_byte(0xE9); |
|
3024 emit_long(0); |
|
3025 } |
|
3026 } |
|
3027 |
|
3028 void Assembler::jmpb(Label& L) { |
|
3029 if (L.is_bound()) { |
|
3030 const int short_size = 2; |
|
3031 address entry = target(L); |
|
3032 assert(is8bit((entry - _code_pos) + short_size), |
|
3033 "Dispacement too large for a short jmp"); |
|
3034 assert(entry != NULL, "jmp most probably wrong"); |
|
3035 intptr_t offs = entry - _code_pos; |
|
3036 emit_byte(0xEB); |
|
3037 emit_byte((offs - short_size) & 0xFF); |
|
3038 } else { |
|
3039 InstructionMark im(this); |
|
3040 L.add_patch_at(code(), locator()); |
|
3041 emit_byte(0xEB); |
|
3042 emit_byte(0); |
|
3043 } |
|
3044 } |
|
3045 |
|
3046 void Assembler::jcc(Condition cc, Label& L, relocInfo::relocType rtype) { |
|
3047 InstructionMark im(this); |
|
3048 relocate(rtype); |
|
3049 assert((0 <= cc) && (cc < 16), "illegal cc"); |
|
3050 if (L.is_bound()) { |
|
3051 address dst = target(L); |
|
3052 assert(dst != NULL, "jcc most probably wrong"); |
|
3053 |
|
3054 const int short_size = 2; |
|
3055 const int long_size = 6; |
|
3056 intptr_t offs = (intptr_t)dst - (intptr_t)_code_pos; |
|
3057 if (rtype == relocInfo::none && is8bit(offs - short_size)) { |
|
3058 // 0111 tttn #8-bit disp |
|
3059 emit_byte(0x70 | cc); |
|
3060 emit_byte((offs - short_size) & 0xFF); |
|
3061 } else { |
|
3062 // 0000 1111 1000 tttn #32-bit disp |
|
3063 assert(is_simm32(offs - long_size), |
|
3064 "must be 32bit offset (call4)"); |
|
3065 emit_byte(0x0F); |
|
3066 emit_byte(0x80 | cc); |
|
3067 emit_long(offs - long_size); |
|
3068 } |
|
3069 } else { |
|
3070 // Note: could eliminate cond. jumps to this jump if condition |
|
3071 // is the same however, seems to be rather unlikely case. |
|
3072 // Note: use jccb() if label to be bound is very close to get |
|
3073 // an 8-bit displacement |
|
3074 L.add_patch_at(code(), locator()); |
|
3075 emit_byte(0x0F); |
|
3076 emit_byte(0x80 | cc); |
|
3077 emit_long(0); |
|
3078 } |
|
3079 } |
|
3080 |
|
3081 void Assembler::jccb(Condition cc, Label& L) { |
|
3082 if (L.is_bound()) { |
|
3083 const int short_size = 2; |
|
3084 const int long_size = 6; |
|
3085 address entry = target(L); |
|
3086 assert(is8bit((intptr_t)entry - ((intptr_t)_code_pos + short_size)), |
|
3087 "Dispacement too large for a short jmp"); |
|
3088 intptr_t offs = (intptr_t)entry - (intptr_t)_code_pos; |
|
3089 // 0111 tttn #8-bit disp |
|
3090 emit_byte(0x70 | cc); |
|
3091 emit_byte((offs - short_size) & 0xFF); |
|
3092 } else { |
|
3093 InstructionMark im(this); |
|
3094 L.add_patch_at(code(), locator()); |
|
3095 emit_byte(0x70 | cc); |
|
3096 emit_byte(0); |
|
3097 } |
|
3098 } |
|
3099 |
|
3100 // FP instructions |
|
3101 |
|
3102 void Assembler::fxsave(Address dst) { |
|
3103 prefixq(dst); |
|
3104 emit_byte(0x0F); |
|
3105 emit_byte(0xAE); |
|
3106 emit_operand(as_Register(0), dst); |
|
3107 } |
|
3108 |
|
3109 void Assembler::fxrstor(Address src) { |
|
3110 prefixq(src); |
|
3111 emit_byte(0x0F); |
|
3112 emit_byte(0xAE); |
|
3113 emit_operand(as_Register(1), src); |
|
3114 } |
|
3115 |
|
3116 void Assembler::ldmxcsr(Address src) { |
|
3117 InstructionMark im(this); |
|
3118 prefix(src); |
|
3119 emit_byte(0x0F); |
|
3120 emit_byte(0xAE); |
|
3121 emit_operand(as_Register(2), src); |
|
3122 } |
|
3123 |
|
3124 void Assembler::stmxcsr(Address dst) { |
|
3125 InstructionMark im(this); |
|
3126 prefix(dst); |
|
3127 emit_byte(0x0F); |
|
3128 emit_byte(0xAE); |
|
3129 emit_operand(as_Register(3), dst); |
|
3130 } |
|
3131 |
|
3132 void Assembler::addss(XMMRegister dst, XMMRegister src) { |
|
3133 emit_byte(0xF3); |
|
3134 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3135 emit_byte(0x0F); |
|
3136 emit_byte(0x58); |
|
3137 emit_byte(0xC0 | encode); |
|
3138 } |
|
3139 |
|
3140 void Assembler::addss(XMMRegister dst, Address src) { |
|
3141 InstructionMark im(this); |
|
3142 emit_byte(0xF3); |
|
3143 prefix(src, dst); |
|
3144 emit_byte(0x0F); |
|
3145 emit_byte(0x58); |
|
3146 emit_operand(dst, src); |
|
3147 } |
|
3148 |
|
3149 void Assembler::subss(XMMRegister dst, XMMRegister src) { |
|
3150 emit_byte(0xF3); |
|
3151 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3152 emit_byte(0x0F); |
|
3153 emit_byte(0x5C); |
|
3154 emit_byte(0xC0 | encode); |
|
3155 } |
|
3156 |
|
3157 void Assembler::subss(XMMRegister dst, Address src) { |
|
3158 InstructionMark im(this); |
|
3159 emit_byte(0xF3); |
|
3160 prefix(src, dst); |
|
3161 emit_byte(0x0F); |
|
3162 emit_byte(0x5C); |
|
3163 emit_operand(dst, src); |
|
3164 } |
|
3165 |
|
3166 void Assembler::mulss(XMMRegister dst, XMMRegister src) { |
|
3167 emit_byte(0xF3); |
|
3168 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3169 emit_byte(0x0F); |
|
3170 emit_byte(0x59); |
|
3171 emit_byte(0xC0 | encode); |
|
3172 } |
|
3173 |
|
3174 void Assembler::mulss(XMMRegister dst, Address src) { |
|
3175 InstructionMark im(this); |
|
3176 emit_byte(0xF3); |
|
3177 prefix(src, dst); |
|
3178 emit_byte(0x0F); |
|
3179 emit_byte(0x59); |
|
3180 emit_operand(dst, src); |
|
3181 } |
|
3182 |
|
3183 void Assembler::divss(XMMRegister dst, XMMRegister src) { |
|
3184 emit_byte(0xF3); |
|
3185 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3186 emit_byte(0x0F); |
|
3187 emit_byte(0x5E); |
|
3188 emit_byte(0xC0 | encode); |
|
3189 } |
|
3190 |
|
3191 void Assembler::divss(XMMRegister dst, Address src) { |
|
3192 InstructionMark im(this); |
|
3193 emit_byte(0xF3); |
|
3194 prefix(src, dst); |
|
3195 emit_byte(0x0F); |
|
3196 emit_byte(0x5E); |
|
3197 emit_operand(dst, src); |
|
3198 } |
|
3199 |
|
3200 void Assembler::addsd(XMMRegister dst, XMMRegister src) { |
|
3201 emit_byte(0xF2); |
|
3202 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3203 emit_byte(0x0F); |
|
3204 emit_byte(0x58); |
|
3205 emit_byte(0xC0 | encode); |
|
3206 } |
|
3207 |
|
3208 void Assembler::addsd(XMMRegister dst, Address src) { |
|
3209 InstructionMark im(this); |
|
3210 emit_byte(0xF2); |
|
3211 prefix(src, dst); |
|
3212 emit_byte(0x0F); |
|
3213 emit_byte(0x58); |
|
3214 emit_operand(dst, src); |
|
3215 } |
|
3216 |
|
3217 void Assembler::subsd(XMMRegister dst, XMMRegister src) { |
|
3218 emit_byte(0xF2); |
|
3219 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3220 emit_byte(0x0F); |
|
3221 emit_byte(0x5C); |
|
3222 emit_byte(0xC0 | encode); |
|
3223 } |
|
3224 |
|
3225 void Assembler::subsd(XMMRegister dst, Address src) { |
|
3226 InstructionMark im(this); |
|
3227 emit_byte(0xF2); |
|
3228 prefix(src, dst); |
|
3229 emit_byte(0x0F); |
|
3230 emit_byte(0x5C); |
|
3231 emit_operand(dst, src); |
|
3232 } |
|
3233 |
|
3234 void Assembler::mulsd(XMMRegister dst, XMMRegister src) { |
|
3235 emit_byte(0xF2); |
|
3236 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3237 emit_byte(0x0F); |
|
3238 emit_byte(0x59); |
|
3239 emit_byte(0xC0 | encode); |
|
3240 } |
|
3241 |
|
3242 void Assembler::mulsd(XMMRegister dst, Address src) { |
|
3243 InstructionMark im(this); |
|
3244 emit_byte(0xF2); |
|
3245 prefix(src, dst); |
|
3246 emit_byte(0x0F); |
|
3247 emit_byte(0x59); |
|
3248 emit_operand(dst, src); |
|
3249 } |
|
3250 |
|
3251 void Assembler::divsd(XMMRegister dst, XMMRegister src) { |
|
3252 emit_byte(0xF2); |
|
3253 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3254 emit_byte(0x0F); |
|
3255 emit_byte(0x5E); |
|
3256 emit_byte(0xC0 | encode); |
|
3257 } |
|
3258 |
|
3259 void Assembler::divsd(XMMRegister dst, Address src) { |
|
3260 InstructionMark im(this); |
|
3261 emit_byte(0xF2); |
|
3262 prefix(src, dst); |
|
3263 emit_byte(0x0F); |
|
3264 emit_byte(0x5E); |
|
3265 emit_operand(dst, src); |
|
3266 } |
|
3267 |
|
3268 void Assembler::sqrtsd(XMMRegister dst, XMMRegister src) { |
|
3269 emit_byte(0xF2); |
|
3270 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3271 emit_byte(0x0F); |
|
3272 emit_byte(0x51); |
|
3273 emit_byte(0xC0 | encode); |
|
3274 } |
|
3275 |
|
3276 void Assembler::sqrtsd(XMMRegister dst, Address src) { |
|
3277 InstructionMark im(this); |
|
3278 emit_byte(0xF2); |
|
3279 prefix(src, dst); |
|
3280 emit_byte(0x0F); |
|
3281 emit_byte(0x51); |
|
3282 emit_operand(dst, src); |
|
3283 } |
|
3284 |
|
3285 void Assembler::xorps(XMMRegister dst, XMMRegister src) { |
|
3286 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3287 emit_byte(0x0F); |
|
3288 emit_byte(0x57); |
|
3289 emit_byte(0xC0 | encode); |
|
3290 } |
|
3291 |
|
3292 void Assembler::xorps(XMMRegister dst, Address src) { |
|
3293 InstructionMark im(this); |
|
3294 prefix(src, dst); |
|
3295 emit_byte(0x0F); |
|
3296 emit_byte(0x57); |
|
3297 emit_operand(dst, src); |
|
3298 } |
|
3299 |
|
3300 void Assembler::xorpd(XMMRegister dst, XMMRegister src) { |
|
3301 emit_byte(0x66); |
|
3302 xorps(dst, src); |
|
3303 } |
|
3304 |
|
3305 void Assembler::xorpd(XMMRegister dst, Address src) { |
|
3306 InstructionMark im(this); |
|
3307 emit_byte(0x66); |
|
3308 prefix(src, dst); |
|
3309 emit_byte(0x0F); |
|
3310 emit_byte(0x57); |
|
3311 emit_operand(dst, src); |
|
3312 } |
|
3313 |
|
3314 void Assembler::cvtsi2ssl(XMMRegister dst, Register src) { |
|
3315 emit_byte(0xF3); |
|
3316 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3317 emit_byte(0x0F); |
|
3318 emit_byte(0x2A); |
|
3319 emit_byte(0xC0 | encode); |
|
3320 } |
|
3321 |
|
3322 void Assembler::cvtsi2ssq(XMMRegister dst, Register src) { |
|
3323 emit_byte(0xF3); |
|
3324 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); |
|
3325 emit_byte(0x0F); |
|
3326 emit_byte(0x2A); |
|
3327 emit_byte(0xC0 | encode); |
|
3328 } |
|
3329 |
|
3330 void Assembler::cvtsi2sdl(XMMRegister dst, Register src) { |
|
3331 emit_byte(0xF2); |
|
3332 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3333 emit_byte(0x0F); |
|
3334 emit_byte(0x2A); |
|
3335 emit_byte(0xC0 | encode); |
|
3336 } |
|
3337 |
|
3338 void Assembler::cvtsi2sdq(XMMRegister dst, Register src) { |
|
3339 emit_byte(0xF2); |
|
3340 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); |
|
3341 emit_byte(0x0F); |
|
3342 emit_byte(0x2A); |
|
3343 emit_byte(0xC0 | encode); |
|
3344 } |
|
3345 |
|
3346 void Assembler::cvttss2sil(Register dst, XMMRegister src) { |
|
3347 emit_byte(0xF3); |
|
3348 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3349 emit_byte(0x0F); |
|
3350 emit_byte(0x2C); |
|
3351 emit_byte(0xC0 | encode); |
|
3352 } |
|
3353 |
|
3354 void Assembler::cvttss2siq(Register dst, XMMRegister src) { |
|
3355 emit_byte(0xF3); |
|
3356 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); |
|
3357 emit_byte(0x0F); |
|
3358 emit_byte(0x2C); |
|
3359 emit_byte(0xC0 | encode); |
|
3360 } |
|
3361 |
|
3362 void Assembler::cvttsd2sil(Register dst, XMMRegister src) { |
|
3363 emit_byte(0xF2); |
|
3364 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3365 emit_byte(0x0F); |
|
3366 emit_byte(0x2C); |
|
3367 emit_byte(0xC0 | encode); |
|
3368 } |
|
3369 |
|
3370 void Assembler::cvttsd2siq(Register dst, XMMRegister src) { |
|
3371 emit_byte(0xF2); |
|
3372 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); |
|
3373 emit_byte(0x0F); |
|
3374 emit_byte(0x2C); |
|
3375 emit_byte(0xC0 | encode); |
|
3376 } |
|
3377 |
|
3378 void Assembler::cvtss2sd(XMMRegister dst, XMMRegister src) { |
|
3379 emit_byte(0xF3); |
|
3380 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3381 emit_byte(0x0F); |
|
3382 emit_byte(0x5A); |
|
3383 emit_byte(0xC0 | encode); |
|
3384 } |
|
3385 |
|
3386 void Assembler::cvtdq2pd(XMMRegister dst, XMMRegister src) { |
|
3387 emit_byte(0xF3); |
|
3388 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3389 emit_byte(0x0F); |
|
3390 emit_byte(0xE6); |
|
3391 emit_byte(0xC0 | encode); |
|
3392 } |
|
3393 |
|
3394 void Assembler::cvtdq2ps(XMMRegister dst, XMMRegister src) { |
|
3395 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3396 emit_byte(0x0F); |
|
3397 emit_byte(0x5B); |
|
3398 emit_byte(0xC0 | encode); |
|
3399 } |
|
3400 |
|
3401 void Assembler::cvtsd2ss(XMMRegister dst, XMMRegister src) { |
|
3402 emit_byte(0xF2); |
|
3403 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3404 emit_byte(0x0F); |
|
3405 emit_byte(0x5A); |
|
3406 emit_byte(0xC0 | encode); |
|
3407 } |
|
3408 |
|
3409 void Assembler::punpcklbw(XMMRegister dst, XMMRegister src) { |
|
3410 emit_byte(0x66); |
|
3411 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
|
3412 emit_byte(0x0F); |
|
3413 emit_byte(0x60); |
|
3414 emit_byte(0xC0 | encode); |
|
3415 } |
|
3416 |
|
3417 // Implementation of MacroAssembler |
|
3418 |
|
3419 // On 32 bit it returns a vanilla displacement on 64 bit is a rip relative displacement |
|
3420 Address MacroAssembler::as_Address(AddressLiteral adr) { |
|
3421 assert(!adr.is_lval(), "must be rval"); |
|
3422 assert(reachable(adr), "must be"); |
|
3423 return Address((int)(intptr_t)(adr.target() - pc()), adr.target(), adr.reloc()); |
|
3424 } |
|
3425 |
|
3426 Address MacroAssembler::as_Address(ArrayAddress adr) { |
|
3427 #ifdef _LP64 |
|
3428 AddressLiteral base = adr.base(); |
|
3429 lea(rscratch1, base); |
|
3430 Address index = adr.index(); |
|
3431 assert(index._disp == 0, "must not have disp"); // maybe it can? |
|
3432 Address array(rscratch1, index._index, index._scale, index._disp); |
|
3433 return array; |
|
3434 #else |
|
3435 return Address::make_array(adr); |
|
3436 #endif // _LP64 |
|
3437 |
|
3438 } |
|
3439 |
|
3440 void MacroAssembler::fat_nop() { |
|
3441 // A 5 byte nop that is safe for patching (see patch_verified_entry) |
|
3442 // Recommened sequence from 'Software Optimization Guide for the AMD |
|
3443 // Hammer Processor' |
|
3444 emit_byte(0x66); |
|
3445 emit_byte(0x66); |
|
3446 emit_byte(0x90); |
|
3447 emit_byte(0x66); |
|
3448 emit_byte(0x90); |
|
3449 } |
|
3450 |
|
3451 static Assembler::Condition reverse[] = { |
|
3452 Assembler::noOverflow /* overflow = 0x0 */ , |
|
3453 Assembler::overflow /* noOverflow = 0x1 */ , |
|
3454 Assembler::aboveEqual /* carrySet = 0x2, below = 0x2 */ , |
|
3455 Assembler::below /* aboveEqual = 0x3, carryClear = 0x3 */ , |
|
3456 Assembler::notZero /* zero = 0x4, equal = 0x4 */ , |
|
3457 Assembler::zero /* notZero = 0x5, notEqual = 0x5 */ , |
|
3458 Assembler::above /* belowEqual = 0x6 */ , |
|
3459 Assembler::belowEqual /* above = 0x7 */ , |
|
3460 Assembler::positive /* negative = 0x8 */ , |
|
3461 Assembler::negative /* positive = 0x9 */ , |
|
3462 Assembler::noParity /* parity = 0xa */ , |
|
3463 Assembler::parity /* noParity = 0xb */ , |
|
3464 Assembler::greaterEqual /* less = 0xc */ , |
|
3465 Assembler::less /* greaterEqual = 0xd */ , |
|
3466 Assembler::greater /* lessEqual = 0xe */ , |
|
3467 Assembler::lessEqual /* greater = 0xf, */ |
|
3468 |
|
3469 }; |
|
3470 |
|
3471 // 32bit can do a case table jump in one instruction but we no longer allow the base |
|
3472 // to be installed in the Address class |
|
3473 void MacroAssembler::jump(ArrayAddress entry) { |
|
3474 #ifdef _LP64 |
|
3475 lea(rscratch1, entry.base()); |
|
3476 Address dispatch = entry.index(); |
|
3477 assert(dispatch._base == noreg, "must be"); |
|
3478 dispatch._base = rscratch1; |
|
3479 jmp(dispatch); |
|
3480 #else |
|
3481 jmp(as_Address(entry)); |
|
3482 #endif // _LP64 |
|
3483 } |
|
3484 |
|
3485 void MacroAssembler::jump(AddressLiteral dst) { |
|
3486 if (reachable(dst)) { |
|
3487 jmp_literal(dst.target(), dst.rspec()); |
|
3488 } else { |
|
3489 lea(rscratch1, dst); |
|
3490 jmp(rscratch1); |
|
3491 } |
|
3492 } |
|
3493 |
|
3494 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst) { |
|
3495 if (reachable(dst)) { |
|
3496 InstructionMark im(this); |
|
3497 relocate(dst.reloc()); |
|
3498 const int short_size = 2; |
|
3499 const int long_size = 6; |
|
3500 int offs = (intptr_t)dst.target() - ((intptr_t)_code_pos); |
|
3501 if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) { |
|
3502 // 0111 tttn #8-bit disp |
|
3503 emit_byte(0x70 | cc); |
|
3504 emit_byte((offs - short_size) & 0xFF); |
|
3505 } else { |
|
3506 // 0000 1111 1000 tttn #32-bit disp |
|
3507 emit_byte(0x0F); |
|
3508 emit_byte(0x80 | cc); |
|
3509 emit_long(offs - long_size); |
|
3510 } |
|
3511 } else { |
|
3512 #ifdef ASSERT |
|
3513 warning("reversing conditional branch"); |
|
3514 #endif /* ASSERT */ |
|
3515 Label skip; |
|
3516 jccb(reverse[cc], skip); |
|
3517 lea(rscratch1, dst); |
|
3518 Assembler::jmp(rscratch1); |
|
3519 bind(skip); |
|
3520 } |
|
3521 } |
|
3522 |
|
3523 // Wouldn't need if AddressLiteral version had new name |
|
3524 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) { |
|
3525 Assembler::call(L, rtype); |
|
3526 } |
|
3527 |
|
3528 // Wouldn't need if AddressLiteral version had new name |
|
3529 void MacroAssembler::call(Register entry) { |
|
3530 Assembler::call(entry); |
|
3531 } |
|
3532 |
|
3533 void MacroAssembler::call(AddressLiteral entry) { |
|
3534 if (reachable(entry)) { |
|
3535 Assembler::call_literal(entry.target(), entry.rspec()); |
|
3536 } else { |
|
3537 lea(rscratch1, entry); |
|
3538 Assembler::call(rscratch1); |
|
3539 } |
|
3540 } |
|
3541 |
|
3542 void MacroAssembler::cmp8(AddressLiteral src1, int8_t src2) { |
|
3543 if (reachable(src1)) { |
|
3544 cmpb(as_Address(src1), src2); |
|
3545 } else { |
|
3546 lea(rscratch1, src1); |
|
3547 cmpb(Address(rscratch1, 0), src2); |
|
3548 } |
|
3549 } |
|
3550 |
|
3551 void MacroAssembler::cmp32(AddressLiteral src1, int32_t src2) { |
|
3552 if (reachable(src1)) { |
|
3553 cmpl(as_Address(src1), src2); |
|
3554 } else { |
|
3555 lea(rscratch1, src1); |
|
3556 cmpl(Address(rscratch1, 0), src2); |
|
3557 } |
|
3558 } |
|
3559 |
|
3560 void MacroAssembler::cmp32(Register src1, AddressLiteral src2) { |
|
3561 if (reachable(src2)) { |
|
3562 cmpl(src1, as_Address(src2)); |
|
3563 } else { |
|
3564 lea(rscratch1, src2); |
|
3565 cmpl(src1, Address(rscratch1, 0)); |
|
3566 } |
|
3567 } |
|
3568 |
|
3569 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2) { |
|
3570 #ifdef _LP64 |
|
3571 if (src2.is_lval()) { |
|
3572 movptr(rscratch1, src2); |
|
3573 Assembler::cmpq(src1, rscratch1); |
|
3574 } else if (reachable(src2)) { |
|
3575 cmpq(src1, as_Address(src2)); |
|
3576 } else { |
|
3577 lea(rscratch1, src2); |
|
3578 Assembler::cmpq(src1, Address(rscratch1, 0)); |
|
3579 } |
|
3580 #else |
|
3581 if (src2.is_lval()) { |
|
3582 cmp_literal32(src1, (int32_t) src2.target(), src2.rspec()); |
|
3583 } else { |
|
3584 cmpl(src1, as_Address(src2)); |
|
3585 } |
|
3586 #endif // _LP64 |
|
3587 } |
|
3588 |
|
3589 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2) { |
|
3590 assert(src2.is_lval(), "not a mem-mem compare"); |
|
3591 #ifdef _LP64 |
|
3592 // moves src2's literal address |
|
3593 movptr(rscratch1, src2); |
|
3594 Assembler::cmpq(src1, rscratch1); |
|
3595 #else |
|
3596 cmp_literal32(src1, (int32_t) src2.target(), src2.rspec()); |
|
3597 #endif // _LP64 |
|
3598 } |
|
3599 |
|
3600 void MacroAssembler::cmp64(Register src1, AddressLiteral src2) { |
|
3601 assert(!src2.is_lval(), "should use cmpptr"); |
|
3602 |
|
3603 if (reachable(src2)) { |
|
3604 #ifdef _LP64 |
|
3605 cmpq(src1, as_Address(src2)); |
|
3606 #else |
|
3607 ShouldNotReachHere(); |
|
3608 #endif // _LP64 |
|
3609 } else { |
|
3610 lea(rscratch1, src2); |
|
3611 Assembler::cmpq(src1, Address(rscratch1, 0)); |
|
3612 } |
|
3613 } |
|
3614 |
|
3615 void MacroAssembler::cmpxchgptr(Register reg, AddressLiteral adr) { |
|
3616 if (reachable(adr)) { |
|
3617 #ifdef _LP64 |
|
3618 cmpxchgq(reg, as_Address(adr)); |
|
3619 #else |
|
3620 cmpxchgl(reg, as_Address(adr)); |
|
3621 #endif // _LP64 |
|
3622 } else { |
|
3623 lea(rscratch1, adr); |
|
3624 cmpxchgq(reg, Address(rscratch1, 0)); |
|
3625 } |
|
3626 } |
|
3627 |
|
3628 void MacroAssembler::incrementl(AddressLiteral dst) { |
|
3629 if (reachable(dst)) { |
|
3630 incrementl(as_Address(dst)); |
|
3631 } else { |
|
3632 lea(rscratch1, dst); |
|
3633 incrementl(Address(rscratch1, 0)); |
|
3634 } |
|
3635 } |
|
3636 |
|
3637 void MacroAssembler::incrementl(ArrayAddress dst) { |
|
3638 incrementl(as_Address(dst)); |
|
3639 } |
|
3640 |
|
3641 void MacroAssembler::lea(Register dst, Address src) { |
|
3642 #ifdef _LP64 |
|
3643 leaq(dst, src); |
|
3644 #else |
|
3645 leal(dst, src); |
|
3646 #endif // _LP64 |
|
3647 } |
|
3648 |
|
3649 void MacroAssembler::lea(Register dst, AddressLiteral src) { |
|
3650 #ifdef _LP64 |
|
3651 mov_literal64(dst, (intptr_t)src.target(), src.rspec()); |
|
3652 #else |
|
3653 mov_literal32(dst, (intptr_t)src.target(), src.rspec()); |
|
3654 #endif // _LP64 |
|
3655 } |
|
3656 |
|
3657 void MacroAssembler::mov32(AddressLiteral dst, Register src) { |
|
3658 if (reachable(dst)) { |
|
3659 movl(as_Address(dst), src); |
|
3660 } else { |
|
3661 lea(rscratch1, dst); |
|
3662 movl(Address(rscratch1, 0), src); |
|
3663 } |
|
3664 } |
|
3665 |
|
3666 void MacroAssembler::mov32(Register dst, AddressLiteral src) { |
|
3667 if (reachable(src)) { |
|
3668 movl(dst, as_Address(src)); |
|
3669 } else { |
|
3670 lea(rscratch1, src); |
|
3671 movl(dst, Address(rscratch1, 0)); |
|
3672 } |
|
3673 } |
|
3674 |
|
3675 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src) { |
|
3676 if (reachable(src)) { |
|
3677 if (UseXmmLoadAndClearUpper) { |
|
3678 movsd (dst, as_Address(src)); |
|
3679 } else { |
|
3680 movlpd(dst, as_Address(src)); |
|
3681 } |
|
3682 } else { |
|
3683 lea(rscratch1, src); |
|
3684 if (UseXmmLoadAndClearUpper) { |
|
3685 movsd (dst, Address(rscratch1, 0)); |
|
3686 } else { |
|
3687 movlpd(dst, Address(rscratch1, 0)); |
|
3688 } |
|
3689 } |
|
3690 } |
|
3691 |
|
3692 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src) { |
|
3693 if (reachable(src)) { |
|
3694 movss(dst, as_Address(src)); |
|
3695 } else { |
|
3696 lea(rscratch1, src); |
|
3697 movss(dst, Address(rscratch1, 0)); |
|
3698 } |
|
3699 } |
|
3700 |
|
3701 void MacroAssembler::movoop(Register dst, jobject obj) { |
|
3702 mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate()); |
|
3703 } |
|
3704 |
|
3705 void MacroAssembler::movoop(Address dst, jobject obj) { |
|
3706 mov_literal64(rscratch1, (intptr_t)obj, oop_Relocation::spec_for_immediate()); |
|
3707 movq(dst, rscratch1); |
|
3708 } |
|
3709 |
|
3710 void MacroAssembler::movptr(Register dst, AddressLiteral src) { |
|
3711 #ifdef _LP64 |
|
3712 if (src.is_lval()) { |
|
3713 mov_literal64(dst, (intptr_t)src.target(), src.rspec()); |
|
3714 } else { |
|
3715 if (reachable(src)) { |
|
3716 movq(dst, as_Address(src)); |
|
3717 } else { |
|
3718 lea(rscratch1, src); |
|
3719 movq(dst, Address(rscratch1,0)); |
|
3720 } |
|
3721 } |
|
3722 #else |
|
3723 if (src.is_lval()) { |
|
3724 mov_literal32(dst, (intptr_t)src.target(), src.rspec()); |
|
3725 } else { |
|
3726 movl(dst, as_Address(src)); |
|
3727 } |
|
3728 #endif // LP64 |
|
3729 } |
|
3730 |
|
3731 void MacroAssembler::movptr(ArrayAddress dst, Register src) { |
|
3732 #ifdef _LP64 |
|
3733 movq(as_Address(dst), src); |
|
3734 #else |
|
3735 movl(as_Address(dst), src); |
|
3736 #endif // _LP64 |
|
3737 } |
|
3738 |
|
3739 void MacroAssembler::pushoop(jobject obj) { |
|
3740 #ifdef _LP64 |
|
3741 movoop(rscratch1, obj); |
|
3742 pushq(rscratch1); |
|
3743 #else |
|
3744 push_literal32((int32_t)obj, oop_Relocation::spec_for_immediate()); |
|
3745 #endif // _LP64 |
|
3746 } |
|
3747 |
|
3748 void MacroAssembler::pushptr(AddressLiteral src) { |
|
3749 #ifdef _LP64 |
|
3750 lea(rscratch1, src); |
|
3751 if (src.is_lval()) { |
|
3752 pushq(rscratch1); |
|
3753 } else { |
|
3754 pushq(Address(rscratch1, 0)); |
|
3755 } |
|
3756 #else |
|
3757 if (src.is_lval()) { |
|
3758 push_literal((int32_t)src.target(), src.rspec()); |
|
3759 else { |
|
3760 pushl(as_Address(src)); |
|
3761 } |
|
3762 #endif // _LP64 |
|
3763 } |
|
3764 |
|
3765 void MacroAssembler::ldmxcsr(AddressLiteral src) { |
|
3766 if (reachable(src)) { |
|
3767 Assembler::ldmxcsr(as_Address(src)); |
|
3768 } else { |
|
3769 lea(rscratch1, src); |
|
3770 Assembler::ldmxcsr(Address(rscratch1, 0)); |
|
3771 } |
|
3772 } |
|
3773 |
|
3774 void MacroAssembler::movlpd(XMMRegister dst, AddressLiteral src) { |
|
3775 if (reachable(src)) { |
|
3776 movlpd(dst, as_Address(src)); |
|
3777 } else { |
|
3778 lea(rscratch1, src); |
|
3779 movlpd(dst, Address(rscratch1, 0)); |
|
3780 } |
|
3781 } |
|
3782 |
|
3783 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src) { |
|
3784 if (reachable(src)) { |
|
3785 movss(dst, as_Address(src)); |
|
3786 } else { |
|
3787 lea(rscratch1, src); |
|
3788 movss(dst, Address(rscratch1, 0)); |
|
3789 } |
|
3790 } |
|
3791 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src) { |
|
3792 if (reachable(src)) { |
|
3793 xorpd(dst, as_Address(src)); |
|
3794 } else { |
|
3795 lea(rscratch1, src); |
|
3796 xorpd(dst, Address(rscratch1, 0)); |
|
3797 } |
|
3798 } |
|
3799 |
|
3800 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src) { |
|
3801 if (reachable(src)) { |
|
3802 xorps(dst, as_Address(src)); |
|
3803 } else { |
|
3804 lea(rscratch1, src); |
|
3805 xorps(dst, Address(rscratch1, 0)); |
|
3806 } |
|
3807 } |
|
3808 |
|
3809 void MacroAssembler::null_check(Register reg, int offset) { |
|
3810 if (needs_explicit_null_check(offset)) { |
|
3811 // provoke OS NULL exception if reg = NULL by |
|
3812 // accessing M[reg] w/o changing any (non-CC) registers |
|
3813 cmpq(rax, Address(reg, 0)); |
|
3814 // Note: should probably use testl(rax, Address(reg, 0)); |
|
3815 // may be shorter code (however, this version of |
|
3816 // testl needs to be implemented first) |
|
3817 } else { |
|
3818 // nothing to do, (later) access of M[reg + offset] |
|
3819 // will provoke OS NULL exception if reg = NULL |
|
3820 } |
|
3821 } |
|
3822 |
|
3823 int MacroAssembler::load_unsigned_byte(Register dst, Address src) { |
|
3824 int off = offset(); |
|
3825 movzbl(dst, src); |
|
3826 return off; |
|
3827 } |
|
3828 |
|
3829 int MacroAssembler::load_unsigned_word(Register dst, Address src) { |
|
3830 int off = offset(); |
|
3831 movzwl(dst, src); |
|
3832 return off; |
|
3833 } |
|
3834 |
|
3835 int MacroAssembler::load_signed_byte(Register dst, Address src) { |
|
3836 int off = offset(); |
|
3837 movsbl(dst, src); |
|
3838 return off; |
|
3839 } |
|
3840 |
|
3841 int MacroAssembler::load_signed_word(Register dst, Address src) { |
|
3842 int off = offset(); |
|
3843 movswl(dst, src); |
|
3844 return off; |
|
3845 } |
|
3846 |
|
3847 void MacroAssembler::incrementl(Register reg, int value) { |
|
3848 if (value == min_jint) { addl(reg, value); return; } |
|
3849 if (value < 0) { decrementl(reg, -value); return; } |
|
3850 if (value == 0) { ; return; } |
|
3851 if (value == 1 && UseIncDec) { incl(reg) ; return; } |
|
3852 /* else */ { addl(reg, value) ; return; } |
|
3853 } |
|
3854 |
|
3855 void MacroAssembler::decrementl(Register reg, int value) { |
|
3856 if (value == min_jint) { subl(reg, value); return; } |
|
3857 if (value < 0) { incrementl(reg, -value); return; } |
|
3858 if (value == 0) { ; return; } |
|
3859 if (value == 1 && UseIncDec) { decl(reg) ; return; } |
|
3860 /* else */ { subl(reg, value) ; return; } |
|
3861 } |
|
3862 |
|
3863 void MacroAssembler::incrementq(Register reg, int value) { |
|
3864 if (value == min_jint) { addq(reg, value); return; } |
|
3865 if (value < 0) { decrementq(reg, -value); return; } |
|
3866 if (value == 0) { ; return; } |
|
3867 if (value == 1 && UseIncDec) { incq(reg) ; return; } |
|
3868 /* else */ { addq(reg, value) ; return; } |
|
3869 } |
|
3870 |
|
3871 void MacroAssembler::decrementq(Register reg, int value) { |
|
3872 if (value == min_jint) { subq(reg, value); return; } |
|
3873 if (value < 0) { incrementq(reg, -value); return; } |
|
3874 if (value == 0) { ; return; } |
|
3875 if (value == 1 && UseIncDec) { decq(reg) ; return; } |
|
3876 /* else */ { subq(reg, value) ; return; } |
|
3877 } |
|
3878 |
|
3879 void MacroAssembler::incrementl(Address dst, int value) { |
|
3880 if (value == min_jint) { addl(dst, value); return; } |
|
3881 if (value < 0) { decrementl(dst, -value); return; } |
|
3882 if (value == 0) { ; return; } |
|
3883 if (value == 1 && UseIncDec) { incl(dst) ; return; } |
|
3884 /* else */ { addl(dst, value) ; return; } |
|
3885 } |
|
3886 |
|
3887 void MacroAssembler::decrementl(Address dst, int value) { |
|
3888 if (value == min_jint) { subl(dst, value); return; } |
|
3889 if (value < 0) { incrementl(dst, -value); return; } |
|
3890 if (value == 0) { ; return; } |
|
3891 if (value == 1 && UseIncDec) { decl(dst) ; return; } |
|
3892 /* else */ { subl(dst, value) ; return; } |
|
3893 } |
|
3894 |
|
3895 void MacroAssembler::incrementq(Address dst, int value) { |
|
3896 if (value == min_jint) { addq(dst, value); return; } |
|
3897 if (value < 0) { decrementq(dst, -value); return; } |
|
3898 if (value == 0) { ; return; } |
|
3899 if (value == 1 && UseIncDec) { incq(dst) ; return; } |
|
3900 /* else */ { addq(dst, value) ; return; } |
|
3901 } |
|
3902 |
|
3903 void MacroAssembler::decrementq(Address dst, int value) { |
|
3904 if (value == min_jint) { subq(dst, value); return; } |
|
3905 if (value < 0) { incrementq(dst, -value); return; } |
|
3906 if (value == 0) { ; return; } |
|
3907 if (value == 1 && UseIncDec) { decq(dst) ; return; } |
|
3908 /* else */ { subq(dst, value) ; return; } |
|
3909 } |
|
3910 |
|
3911 void MacroAssembler::align(int modulus) { |
|
3912 if (offset() % modulus != 0) { |
|
3913 nop(modulus - (offset() % modulus)); |
|
3914 } |
|
3915 } |
|
3916 |
|
3917 void MacroAssembler::enter() { |
|
3918 pushq(rbp); |
|
3919 movq(rbp, rsp); |
|
3920 } |
|
3921 |
|
3922 void MacroAssembler::leave() { |
|
3923 emit_byte(0xC9); // LEAVE |
|
3924 } |
|
3925 |
|
3926 // C++ bool manipulation |
|
3927 |
|
3928 void MacroAssembler::movbool(Register dst, Address src) { |
|
3929 if(sizeof(bool) == 1) |
|
3930 movb(dst, src); |
|
3931 else if(sizeof(bool) == 2) |
|
3932 movw(dst, src); |
|
3933 else if(sizeof(bool) == 4) |
|
3934 movl(dst, src); |
|
3935 else { |
|
3936 // unsupported |
|
3937 ShouldNotReachHere(); |
|
3938 } |
|
3939 } |
|
3940 |
|
3941 void MacroAssembler::movbool(Address dst, bool boolconst) { |
|
3942 if(sizeof(bool) == 1) |
|
3943 movb(dst, (int) boolconst); |
|
3944 else if(sizeof(bool) == 2) |
|
3945 movw(dst, (int) boolconst); |
|
3946 else if(sizeof(bool) == 4) |
|
3947 movl(dst, (int) boolconst); |
|
3948 else { |
|
3949 // unsupported |
|
3950 ShouldNotReachHere(); |
|
3951 } |
|
3952 } |
|
3953 |
|
3954 void MacroAssembler::movbool(Address dst, Register src) { |
|
3955 if(sizeof(bool) == 1) |
|
3956 movb(dst, src); |
|
3957 else if(sizeof(bool) == 2) |
|
3958 movw(dst, src); |
|
3959 else if(sizeof(bool) == 4) |
|
3960 movl(dst, src); |
|
3961 else { |
|
3962 // unsupported |
|
3963 ShouldNotReachHere(); |
|
3964 } |
|
3965 } |
|
3966 |
|
3967 void MacroAssembler::testbool(Register dst) { |
|
3968 if(sizeof(bool) == 1) |
|
3969 testb(dst, (int) 0xff); |
|
3970 else if(sizeof(bool) == 2) { |
|
3971 // need testw impl |
|
3972 ShouldNotReachHere(); |
|
3973 } else if(sizeof(bool) == 4) |
|
3974 testl(dst, dst); |
|
3975 else { |
|
3976 // unsupported |
|
3977 ShouldNotReachHere(); |
|
3978 } |
|
3979 } |
|
3980 |
|
3981 void MacroAssembler::set_last_Java_frame(Register last_java_sp, |
|
3982 Register last_java_fp, |
|
3983 address last_java_pc) { |
|
3984 // determine last_java_sp register |
|
3985 if (!last_java_sp->is_valid()) { |
|
3986 last_java_sp = rsp; |
|
3987 } |
|
3988 |
|
3989 // last_java_fp is optional |
|
3990 if (last_java_fp->is_valid()) { |
|
3991 movq(Address(r15_thread, JavaThread::last_Java_fp_offset()), |
|
3992 last_java_fp); |
|
3993 } |
|
3994 |
|
3995 // last_java_pc is optional |
|
3996 if (last_java_pc != NULL) { |
|
3997 Address java_pc(r15_thread, |
|
3998 JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset()); |
|
3999 lea(rscratch1, InternalAddress(last_java_pc)); |
|
4000 movq(java_pc, rscratch1); |
|
4001 } |
|
4002 |
|
4003 movq(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp); |
|
4004 } |
|
4005 |
|
4006 void MacroAssembler::reset_last_Java_frame(bool clear_fp, |
|
4007 bool clear_pc) { |
|
4008 // we must set sp to zero to clear frame |
|
4009 movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD); |
|
4010 // must clear fp, so that compiled frames are not confused; it is |
|
4011 // possible that we need it only for debugging |
|
4012 if (clear_fp) { |
|
4013 movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD); |
|
4014 } |
|
4015 |
|
4016 if (clear_pc) { |
|
4017 movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD); |
|
4018 } |
|
4019 } |
|
4020 |
|
4021 |
|
4022 // Implementation of call_VM versions |
|
4023 |
|
4024 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) { |
|
4025 Label L, E; |
|
4026 |
|
4027 #ifdef _WIN64 |
|
4028 // Windows always allocates space for it's register args |
|
4029 assert(num_args <= 4, "only register arguments supported"); |
|
4030 subq(rsp, frame::arg_reg_save_area_bytes); |
|
4031 #endif |
|
4032 |
|
4033 // Align stack if necessary |
|
4034 testl(rsp, 15); |
|
4035 jcc(Assembler::zero, L); |
|
4036 |
|
4037 subq(rsp, 8); |
|
4038 { |
|
4039 call(RuntimeAddress(entry_point)); |
|
4040 } |
|
4041 addq(rsp, 8); |
|
4042 jmp(E); |
|
4043 |
|
4044 bind(L); |
|
4045 { |
|
4046 call(RuntimeAddress(entry_point)); |
|
4047 } |
|
4048 |
|
4049 bind(E); |
|
4050 |
|
4051 #ifdef _WIN64 |
|
4052 // restore stack pointer |
|
4053 addq(rsp, frame::arg_reg_save_area_bytes); |
|
4054 #endif |
|
4055 |
|
4056 } |
|
4057 |
|
4058 |
|
4059 void MacroAssembler::call_VM_base(Register oop_result, |
|
4060 Register java_thread, |
|
4061 Register last_java_sp, |
|
4062 address entry_point, |
|
4063 int num_args, |
|
4064 bool check_exceptions) { |
|
4065 // determine last_java_sp register |
|
4066 if (!last_java_sp->is_valid()) { |
|
4067 last_java_sp = rsp; |
|
4068 } |
|
4069 |
|
4070 // debugging support |
|
4071 assert(num_args >= 0, "cannot have negative number of arguments"); |
|
4072 assert(r15_thread != oop_result, |
|
4073 "cannot use the same register for java_thread & oop_result"); |
|
4074 assert(r15_thread != last_java_sp, |
|
4075 "cannot use the same register for java_thread & last_java_sp"); |
|
4076 |
|
4077 // set last Java frame before call |
|
4078 |
|
4079 // This sets last_Java_fp which is only needed from interpreted frames |
|
4080 // and should really be done only from the interp_masm version before |
|
4081 // calling the underlying call_VM. That doesn't happen yet so we set |
|
4082 // last_Java_fp here even though some callers don't need it and |
|
4083 // also clear it below. |
|
4084 set_last_Java_frame(last_java_sp, rbp, NULL); |
|
4085 |
|
4086 { |
|
4087 Label L, E; |
|
4088 |
|
4089 // Align stack if necessary |
|
4090 #ifdef _WIN64 |
|
4091 assert(num_args <= 4, "only register arguments supported"); |
|
4092 // Windows always allocates space for it's register args |
|
4093 subq(rsp, frame::arg_reg_save_area_bytes); |
|
4094 #endif |
|
4095 testl(rsp, 15); |
|
4096 jcc(Assembler::zero, L); |
|
4097 |
|
4098 subq(rsp, 8); |
|
4099 { |
|
4100 call(RuntimeAddress(entry_point)); |
|
4101 } |
|
4102 addq(rsp, 8); |
|
4103 jmp(E); |
|
4104 |
|
4105 |
|
4106 bind(L); |
|
4107 { |
|
4108 call(RuntimeAddress(entry_point)); |
|
4109 } |
|
4110 |
|
4111 bind(E); |
|
4112 |
|
4113 #ifdef _WIN64 |
|
4114 // restore stack pointer |
|
4115 addq(rsp, frame::arg_reg_save_area_bytes); |
|
4116 #endif |
|
4117 } |
|
4118 |
|
4119 #ifdef ASSERT |
|
4120 pushq(rax); |
|
4121 { |
|
4122 Label L; |
|
4123 get_thread(rax); |
|
4124 cmpq(r15_thread, rax); |
|
4125 jcc(Assembler::equal, L); |
|
4126 stop("MacroAssembler::call_VM_base: register not callee saved?"); |
|
4127 bind(L); |
|
4128 } |
|
4129 popq(rax); |
|
4130 #endif |
|
4131 |
|
4132 // reset last Java frame |
|
4133 // This really shouldn't have to clear fp set note above at the |
|
4134 // call to set_last_Java_frame |
|
4135 reset_last_Java_frame(true, false); |
|
4136 |
|
4137 check_and_handle_popframe(noreg); |
|
4138 check_and_handle_earlyret(noreg); |
|
4139 |
|
4140 if (check_exceptions) { |
|
4141 cmpq(Address(r15_thread, Thread::pending_exception_offset()), (int) NULL); |
|
4142 // This used to conditionally jump to forward_exception however it is |
|
4143 // possible if we relocate that the branch will not reach. So we must jump |
|
4144 // around so we can always reach |
|
4145 Label ok; |
|
4146 jcc(Assembler::equal, ok); |
|
4147 jump(RuntimeAddress(StubRoutines::forward_exception_entry())); |
|
4148 bind(ok); |
|
4149 } |
|
4150 |
|
4151 // get oop result if there is one and reset the value in the thread |
|
4152 if (oop_result->is_valid()) { |
|
4153 movq(oop_result, Address(r15_thread, JavaThread::vm_result_offset())); |
|
4154 movptr(Address(r15_thread, JavaThread::vm_result_offset()), NULL_WORD); |
|
4155 verify_oop(oop_result, "broken oop in call_VM_base"); |
|
4156 } |
|
4157 } |
|
4158 |
|
4159 void MacroAssembler::check_and_handle_popframe(Register java_thread) {} |
|
4160 void MacroAssembler::check_and_handle_earlyret(Register java_thread) {} |
|
4161 |
|
4162 void MacroAssembler::call_VM_helper(Register oop_result, |
|
4163 address entry_point, |
|
4164 int num_args, |
|
4165 bool check_exceptions) { |
|
4166 // Java thread becomes first argument of C function |
|
4167 movq(c_rarg0, r15_thread); |
|
4168 |
|
4169 // We've pushed one address, correct last_Java_sp |
|
4170 leaq(rax, Address(rsp, wordSize)); |
|
4171 |
|
4172 call_VM_base(oop_result, noreg, rax, entry_point, num_args, |
|
4173 check_exceptions); |
|
4174 } |
|
4175 |
|
4176 |
|
4177 void MacroAssembler::call_VM(Register oop_result, |
|
4178 address entry_point, |
|
4179 bool check_exceptions) { |
|
4180 Label C, E; |
|
4181 Assembler::call(C, relocInfo::none); |
|
4182 jmp(E); |
|
4183 |
|
4184 bind(C); |
|
4185 call_VM_helper(oop_result, entry_point, 0, check_exceptions); |
|
4186 ret(0); |
|
4187 |
|
4188 bind(E); |
|
4189 } |
|
4190 |
|
4191 |
|
4192 void MacroAssembler::call_VM(Register oop_result, |
|
4193 address entry_point, |
|
4194 Register arg_1, |
|
4195 bool check_exceptions) { |
|
4196 assert(rax != arg_1, "smashed argument"); |
|
4197 assert(c_rarg0 != arg_1, "smashed argument"); |
|
4198 |
|
4199 Label C, E; |
|
4200 Assembler::call(C, relocInfo::none); |
|
4201 jmp(E); |
|
4202 |
|
4203 bind(C); |
|
4204 // c_rarg0 is reserved for thread |
|
4205 if (c_rarg1 != arg_1) { |
|
4206 movq(c_rarg1, arg_1); |
|
4207 } |
|
4208 call_VM_helper(oop_result, entry_point, 1, check_exceptions); |
|
4209 ret(0); |
|
4210 |
|
4211 bind(E); |
|
4212 } |
|
4213 |
|
4214 void MacroAssembler::call_VM(Register oop_result, |
|
4215 address entry_point, |
|
4216 Register arg_1, |
|
4217 Register arg_2, |
|
4218 bool check_exceptions) { |
|
4219 assert(rax != arg_1, "smashed argument"); |
|
4220 assert(rax != arg_2, "smashed argument"); |
|
4221 assert(c_rarg0 != arg_1, "smashed argument"); |
|
4222 assert(c_rarg0 != arg_2, "smashed argument"); |
|
4223 assert(c_rarg1 != arg_2, "smashed argument"); |
|
4224 assert(c_rarg2 != arg_1, "smashed argument"); |
|
4225 |
|
4226 Label C, E; |
|
4227 Assembler::call(C, relocInfo::none); |
|
4228 jmp(E); |
|
4229 |
|
4230 bind(C); |
|
4231 // c_rarg0 is reserved for thread |
|
4232 if (c_rarg1 != arg_1) { |
|
4233 movq(c_rarg1, arg_1); |
|
4234 } |
|
4235 if (c_rarg2 != arg_2) { |
|
4236 movq(c_rarg2, arg_2); |
|
4237 } |
|
4238 call_VM_helper(oop_result, entry_point, 2, check_exceptions); |
|
4239 ret(0); |
|
4240 |
|
4241 bind(E); |
|
4242 } |
|
4243 |
|
4244 |
|
4245 void MacroAssembler::call_VM(Register oop_result, |
|
4246 address entry_point, |
|
4247 Register arg_1, |
|
4248 Register arg_2, |
|
4249 Register arg_3, |
|
4250 bool check_exceptions) { |
|
4251 assert(rax != arg_1, "smashed argument"); |
|
4252 assert(rax != arg_2, "smashed argument"); |
|
4253 assert(rax != arg_3, "smashed argument"); |
|
4254 assert(c_rarg0 != arg_1, "smashed argument"); |
|
4255 assert(c_rarg0 != arg_2, "smashed argument"); |
|
4256 assert(c_rarg0 != arg_3, "smashed argument"); |
|
4257 assert(c_rarg1 != arg_2, "smashed argument"); |
|
4258 assert(c_rarg1 != arg_3, "smashed argument"); |
|
4259 assert(c_rarg2 != arg_1, "smashed argument"); |
|
4260 assert(c_rarg2 != arg_3, "smashed argument"); |
|
4261 assert(c_rarg3 != arg_1, "smashed argument"); |
|
4262 assert(c_rarg3 != arg_2, "smashed argument"); |
|
4263 |
|
4264 Label C, E; |
|
4265 Assembler::call(C, relocInfo::none); |
|
4266 jmp(E); |
|
4267 |
|
4268 bind(C); |
|
4269 // c_rarg0 is reserved for thread |
|
4270 if (c_rarg1 != arg_1) { |
|
4271 movq(c_rarg1, arg_1); |
|
4272 } |
|
4273 if (c_rarg2 != arg_2) { |
|
4274 movq(c_rarg2, arg_2); |
|
4275 } |
|
4276 if (c_rarg3 != arg_3) { |
|
4277 movq(c_rarg3, arg_3); |
|
4278 } |
|
4279 call_VM_helper(oop_result, entry_point, 3, check_exceptions); |
|
4280 ret(0); |
|
4281 |
|
4282 bind(E); |
|
4283 } |
|
4284 |
|
4285 void MacroAssembler::call_VM(Register oop_result, |
|
4286 Register last_java_sp, |
|
4287 address entry_point, |
|
4288 int num_args, |
|
4289 bool check_exceptions) { |
|
4290 call_VM_base(oop_result, noreg, last_java_sp, entry_point, num_args, |
|
4291 check_exceptions); |
|
4292 } |
|
4293 |
|
4294 void MacroAssembler::call_VM(Register oop_result, |
|
4295 Register last_java_sp, |
|
4296 address entry_point, |
|
4297 Register arg_1, |
|
4298 bool check_exceptions) { |
|
4299 assert(c_rarg0 != arg_1, "smashed argument"); |
|
4300 assert(c_rarg1 != last_java_sp, "smashed argument"); |
|
4301 // c_rarg0 is reserved for thread |
|
4302 if (c_rarg1 != arg_1) { |
|
4303 movq(c_rarg1, arg_1); |
|
4304 } |
|
4305 call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions); |
|
4306 } |
|
4307 |
|
4308 void MacroAssembler::call_VM(Register oop_result, |
|
4309 Register last_java_sp, |
|
4310 address entry_point, |
|
4311 Register arg_1, |
|
4312 Register arg_2, |
|
4313 bool check_exceptions) { |
|
4314 assert(c_rarg0 != arg_1, "smashed argument"); |
|
4315 assert(c_rarg0 != arg_2, "smashed argument"); |
|
4316 assert(c_rarg1 != arg_2, "smashed argument"); |
|
4317 assert(c_rarg1 != last_java_sp, "smashed argument"); |
|
4318 assert(c_rarg2 != arg_1, "smashed argument"); |
|
4319 assert(c_rarg2 != last_java_sp, "smashed argument"); |
|
4320 // c_rarg0 is reserved for thread |
|
4321 if (c_rarg1 != arg_1) { |
|
4322 movq(c_rarg1, arg_1); |
|
4323 } |
|
4324 if (c_rarg2 != arg_2) { |
|
4325 movq(c_rarg2, arg_2); |
|
4326 } |
|
4327 call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions); |
|
4328 } |
|
4329 |
|
4330 |
|
4331 void MacroAssembler::call_VM(Register oop_result, |
|
4332 Register last_java_sp, |
|
4333 address entry_point, |
|
4334 Register arg_1, |
|
4335 Register arg_2, |
|
4336 Register arg_3, |
|
4337 bool check_exceptions) { |
|
4338 assert(c_rarg0 != arg_1, "smashed argument"); |
|
4339 assert(c_rarg0 != arg_2, "smashed argument"); |
|
4340 assert(c_rarg0 != arg_3, "smashed argument"); |
|
4341 assert(c_rarg1 != arg_2, "smashed argument"); |
|
4342 assert(c_rarg1 != arg_3, "smashed argument"); |
|
4343 assert(c_rarg1 != last_java_sp, "smashed argument"); |
|
4344 assert(c_rarg2 != arg_1, "smashed argument"); |
|
4345 assert(c_rarg2 != arg_3, "smashed argument"); |
|
4346 assert(c_rarg2 != last_java_sp, "smashed argument"); |
|
4347 assert(c_rarg3 != arg_1, "smashed argument"); |
|
4348 assert(c_rarg3 != arg_2, "smashed argument"); |
|
4349 assert(c_rarg3 != last_java_sp, "smashed argument"); |
|
4350 // c_rarg0 is reserved for thread |
|
4351 if (c_rarg1 != arg_1) { |
|
4352 movq(c_rarg1, arg_1); |
|
4353 } |
|
4354 if (c_rarg2 != arg_2) { |
|
4355 movq(c_rarg2, arg_2); |
|
4356 } |
|
4357 if (c_rarg3 != arg_3) { |
|
4358 movq(c_rarg2, arg_3); |
|
4359 } |
|
4360 call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions); |
|
4361 } |
|
4362 |
|
4363 void MacroAssembler::call_VM_leaf(address entry_point, int num_args) { |
|
4364 call_VM_leaf_base(entry_point, num_args); |
|
4365 } |
|
4366 |
|
4367 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_1) { |
|
4368 if (c_rarg0 != arg_1) { |
|
4369 movq(c_rarg0, arg_1); |
|
4370 } |
|
4371 call_VM_leaf(entry_point, 1); |
|
4372 } |
|
4373 |
|
4374 void MacroAssembler::call_VM_leaf(address entry_point, |
|
4375 Register arg_1, |
|
4376 Register arg_2) { |
|
4377 assert(c_rarg0 != arg_2, "smashed argument"); |
|
4378 assert(c_rarg1 != arg_1, "smashed argument"); |
|
4379 if (c_rarg0 != arg_1) { |
|
4380 movq(c_rarg0, arg_1); |
|
4381 } |
|
4382 if (c_rarg1 != arg_2) { |
|
4383 movq(c_rarg1, arg_2); |
|
4384 } |
|
4385 call_VM_leaf(entry_point, 2); |
|
4386 } |
|
4387 |
|
4388 void MacroAssembler::call_VM_leaf(address entry_point, |
|
4389 Register arg_1, |
|
4390 Register arg_2, |
|
4391 Register arg_3) { |
|
4392 assert(c_rarg0 != arg_2, "smashed argument"); |
|
4393 assert(c_rarg0 != arg_3, "smashed argument"); |
|
4394 assert(c_rarg1 != arg_1, "smashed argument"); |
|
4395 assert(c_rarg1 != arg_3, "smashed argument"); |
|
4396 assert(c_rarg2 != arg_1, "smashed argument"); |
|
4397 assert(c_rarg2 != arg_2, "smashed argument"); |
|
4398 if (c_rarg0 != arg_1) { |
|
4399 movq(c_rarg0, arg_1); |
|
4400 } |
|
4401 if (c_rarg1 != arg_2) { |
|
4402 movq(c_rarg1, arg_2); |
|
4403 } |
|
4404 if (c_rarg2 != arg_3) { |
|
4405 movq(c_rarg2, arg_3); |
|
4406 } |
|
4407 call_VM_leaf(entry_point, 3); |
|
4408 } |
|
4409 |
|
4410 |
|
4411 // Calls to C land |
|
4412 // |
|
4413 // When entering C land, the rbp & rsp of the last Java frame have to |
|
4414 // be recorded in the (thread-local) JavaThread object. When leaving C |
|
4415 // land, the last Java fp has to be reset to 0. This is required to |
|
4416 // allow proper stack traversal. |
|
4417 void MacroAssembler::store_check(Register obj) { |
|
4418 // Does a store check for the oop in register obj. The content of |
|
4419 // register obj is destroyed afterwards. |
|
4420 store_check_part_1(obj); |
|
4421 store_check_part_2(obj); |
|
4422 } |
|
4423 |
|
4424 void MacroAssembler::store_check(Register obj, Address dst) { |
|
4425 store_check(obj); |
|
4426 } |
|
4427 |
|
4428 // split the store check operation so that other instructions can be |
|
4429 // scheduled inbetween |
|
4430 void MacroAssembler::store_check_part_1(Register obj) { |
|
4431 BarrierSet* bs = Universe::heap()->barrier_set(); |
|
4432 assert(bs->kind() == BarrierSet::CardTableModRef, "Wrong barrier set kind"); |
|
4433 shrq(obj, CardTableModRefBS::card_shift); |
|
4434 } |
|
4435 |
|
4436 void MacroAssembler::store_check_part_2(Register obj) { |
|
4437 BarrierSet* bs = Universe::heap()->barrier_set(); |
|
4438 assert(bs->kind() == BarrierSet::CardTableModRef, "Wrong barrier set kind"); |
|
4439 CardTableModRefBS* ct = (CardTableModRefBS*)bs; |
|
4440 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); |
|
4441 |
|
4442 // The calculation for byte_map_base is as follows: |
|
4443 // byte_map_base = _byte_map - (uintptr_t(low_bound) >> card_shift); |
|
4444 // So this essentially converts an address to a displacement and |
|
4445 // it will never need to be relocated. On 64bit however the value may be too |
|
4446 // large for a 32bit displacement |
|
4447 |
|
4448 intptr_t disp = (intptr_t) ct->byte_map_base; |
|
4449 if (is_simm32(disp)) { |
|
4450 Address cardtable(noreg, obj, Address::times_1, disp); |
|
4451 movb(cardtable, 0); |
|
4452 } else { |
|
4453 // By doing it as an ExternalAddress disp could be converted to a rip-relative |
|
4454 // displacement and done in a single instruction given favorable mapping and |
|
4455 // a smarter version of as_Address. Worst case it is two instructions which |
|
4456 // is no worse off then loading disp into a register and doing as a simple |
|
4457 // Address() as above. |
|
4458 // We can't do as ExternalAddress as the only style since if disp == 0 we'll |
|
4459 // assert since NULL isn't acceptable in a reloci (see 6644928). In any case |
|
4460 // in some cases we'll get a single instruction version. |
|
4461 |
|
4462 ExternalAddress cardtable((address)disp); |
|
4463 Address index(noreg, obj, Address::times_1); |
|
4464 movb(as_Address(ArrayAddress(cardtable, index)), 0); |
|
4465 } |
|
4466 |
|
4467 } |
|
4468 |
|
4469 void MacroAssembler::c2bool(Register x) { |
|
4470 // implements x == 0 ? 0 : 1 |
|
4471 // note: must only look at least-significant byte of x |
|
4472 // since C-style booleans are stored in one byte |
|
4473 // only! (was bug) |
|
4474 andl(x, 0xFF); |
|
4475 setb(Assembler::notZero, x); |
|
4476 } |
|
4477 |
|
4478 int MacroAssembler::corrected_idivl(Register reg) { |
|
4479 // Full implementation of Java idiv and irem; checks for special |
|
4480 // case as described in JVM spec., p.243 & p.271. The function |
|
4481 // returns the (pc) offset of the idivl instruction - may be needed |
|
4482 // for implicit exceptions. |
|
4483 // |
|
4484 // normal case special case |
|
4485 // |
|
4486 // input : eax: dividend min_int |
|
4487 // reg: divisor (may not be eax/edx) -1 |
|
4488 // |
|
4489 // output: eax: quotient (= eax idiv reg) min_int |
|
4490 // edx: remainder (= eax irem reg) 0 |
|
4491 assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register"); |
|
4492 const int min_int = 0x80000000; |
|
4493 Label normal_case, special_case; |
|
4494 |
|
4495 // check for special case |
|
4496 cmpl(rax, min_int); |
|
4497 jcc(Assembler::notEqual, normal_case); |
|
4498 xorl(rdx, rdx); // prepare edx for possible special case (where |
|
4499 // remainder = 0) |
|
4500 cmpl(reg, -1); |
|
4501 jcc(Assembler::equal, special_case); |
|
4502 |
|
4503 // handle normal case |
|
4504 bind(normal_case); |
|
4505 cdql(); |
|
4506 int idivl_offset = offset(); |
|
4507 idivl(reg); |
|
4508 |
|
4509 // normal and special case exit |
|
4510 bind(special_case); |
|
4511 |
|
4512 return idivl_offset; |
|
4513 } |
|
4514 |
|
4515 int MacroAssembler::corrected_idivq(Register reg) { |
|
4516 // Full implementation of Java ldiv and lrem; checks for special |
|
4517 // case as described in JVM spec., p.243 & p.271. The function |
|
4518 // returns the (pc) offset of the idivl instruction - may be needed |
|
4519 // for implicit exceptions. |
|
4520 // |
|
4521 // normal case special case |
|
4522 // |
|
4523 // input : rax: dividend min_long |
|
4524 // reg: divisor (may not be eax/edx) -1 |
|
4525 // |
|
4526 // output: rax: quotient (= rax idiv reg) min_long |
|
4527 // rdx: remainder (= rax irem reg) 0 |
|
4528 assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register"); |
|
4529 static const int64_t min_long = 0x8000000000000000; |
|
4530 Label normal_case, special_case; |
|
4531 |
|
4532 // check for special case |
|
4533 cmp64(rax, ExternalAddress((address) &min_long)); |
|
4534 jcc(Assembler::notEqual, normal_case); |
|
4535 xorl(rdx, rdx); // prepare rdx for possible special case (where |
|
4536 // remainder = 0) |
|
4537 cmpq(reg, -1); |
|
4538 jcc(Assembler::equal, special_case); |
|
4539 |
|
4540 // handle normal case |
|
4541 bind(normal_case); |
|
4542 cdqq(); |
|
4543 int idivq_offset = offset(); |
|
4544 idivq(reg); |
|
4545 |
|
4546 // normal and special case exit |
|
4547 bind(special_case); |
|
4548 |
|
4549 return idivq_offset; |
|
4550 } |
|
4551 |
|
4552 void MacroAssembler::push_IU_state() { |
|
4553 pushfq(); // Push flags first because pushaq kills them |
|
4554 subq(rsp, 8); // Make sure rsp stays 16-byte aligned |
|
4555 pushaq(); |
|
4556 } |
|
4557 |
|
4558 void MacroAssembler::pop_IU_state() { |
|
4559 popaq(); |
|
4560 addq(rsp, 8); |
|
4561 popfq(); |
|
4562 } |
|
4563 |
|
4564 void MacroAssembler::push_FPU_state() { |
|
4565 subq(rsp, FPUStateSizeInWords * wordSize); |
|
4566 fxsave(Address(rsp, 0)); |
|
4567 } |
|
4568 |
|
4569 void MacroAssembler::pop_FPU_state() { |
|
4570 fxrstor(Address(rsp, 0)); |
|
4571 addq(rsp, FPUStateSizeInWords * wordSize); |
|
4572 } |
|
4573 |
|
4574 // Save Integer and Float state |
|
4575 // Warning: Stack must be 16 byte aligned |
|
4576 void MacroAssembler::push_CPU_state() { |
|
4577 push_IU_state(); |
|
4578 push_FPU_state(); |
|
4579 } |
|
4580 |
|
4581 void MacroAssembler::pop_CPU_state() { |
|
4582 pop_FPU_state(); |
|
4583 pop_IU_state(); |
|
4584 } |
|
4585 |
|
4586 void MacroAssembler::sign_extend_short(Register reg) { |
|
4587 movswl(reg, reg); |
|
4588 } |
|
4589 |
|
4590 void MacroAssembler::sign_extend_byte(Register reg) { |
|
4591 movsbl(reg, reg); |
|
4592 } |
|
4593 |
|
4594 void MacroAssembler::division_with_shift(Register reg, int shift_value) { |
|
4595 assert (shift_value > 0, "illegal shift value"); |
|
4596 Label _is_positive; |
|
4597 testl (reg, reg); |
|
4598 jcc (Assembler::positive, _is_positive); |
|
4599 int offset = (1 << shift_value) - 1 ; |
|
4600 |
|
4601 if (offset == 1) { |
|
4602 incrementl(reg); |
|
4603 } else { |
|
4604 addl(reg, offset); |
|
4605 } |
|
4606 |
|
4607 bind (_is_positive); |
|
4608 sarl(reg, shift_value); |
|
4609 } |
|
4610 |
|
4611 void MacroAssembler::round_to_l(Register reg, int modulus) { |
|
4612 addl(reg, modulus - 1); |
|
4613 andl(reg, -modulus); |
|
4614 } |
|
4615 |
|
4616 void MacroAssembler::round_to_q(Register reg, int modulus) { |
|
4617 addq(reg, modulus - 1); |
|
4618 andq(reg, -modulus); |
|
4619 } |
|
4620 |
|
4621 void MacroAssembler::verify_oop(Register reg, const char* s) { |
|
4622 if (!VerifyOops) { |
|
4623 return; |
|
4624 } |
|
4625 |
|
4626 // Pass register number to verify_oop_subroutine |
|
4627 char* b = new char[strlen(s) + 50]; |
|
4628 sprintf(b, "verify_oop: %s: %s", reg->name(), s); |
|
4629 |
|
4630 pushq(rax); // save rax, restored by receiver |
|
4631 |
|
4632 // pass args on stack, only touch rax |
|
4633 pushq(reg); |
|
4634 // avoid using pushptr, as it modifies scratch registers |
|
4635 // and our contract is not to modify anything |
|
4636 ExternalAddress buffer((address)b); |
|
4637 movptr(rax, buffer.addr()); |
|
4638 pushq(rax); |
|
4639 |
|
4640 // call indirectly to solve generation ordering problem |
|
4641 movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address())); |
|
4642 call(rax); // no alignment requirement |
|
4643 // everything popped by receiver |
|
4644 } |
|
4645 |
|
4646 void MacroAssembler::verify_oop_addr(Address addr, const char* s) { |
|
4647 if (!VerifyOops) return; |
|
4648 // Pass register number to verify_oop_subroutine |
|
4649 char* b = new char[strlen(s) + 50]; |
|
4650 sprintf(b, "verify_oop_addr: %s", s); |
|
4651 pushq(rax); // save rax |
|
4652 movq(addr, rax); |
|
4653 pushq(rax); // pass register argument |
|
4654 |
|
4655 |
|
4656 // avoid using pushptr, as it modifies scratch registers |
|
4657 // and our contract is not to modify anything |
|
4658 ExternalAddress buffer((address)b); |
|
4659 movptr(rax, buffer.addr()); |
|
4660 pushq(rax); |
|
4661 |
|
4662 // call indirectly to solve generation ordering problem |
|
4663 movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address())); |
|
4664 call(rax); // no alignment requirement |
|
4665 // everything popped by receiver |
|
4666 } |
|
4667 |
|
4668 |
|
4669 void MacroAssembler::stop(const char* msg) { |
|
4670 address rip = pc(); |
|
4671 pushaq(); // get regs on stack |
|
4672 lea(c_rarg0, ExternalAddress((address) msg)); |
|
4673 lea(c_rarg1, InternalAddress(rip)); |
|
4674 movq(c_rarg2, rsp); // pass pointer to regs array |
|
4675 andq(rsp, -16); // align stack as required by ABI |
|
4676 call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug))); |
|
4677 hlt(); |
|
4678 } |
|
4679 |
|
4680 void MacroAssembler::warn(const char* msg) { |
|
4681 pushq(r12); |
|
4682 movq(r12, rsp); |
|
4683 andq(rsp, -16); // align stack as required by push_CPU_state and call |
|
4684 |
|
4685 push_CPU_state(); // keeps alignment at 16 bytes |
|
4686 lea(c_rarg0, ExternalAddress((address) msg)); |
|
4687 call_VM_leaf(CAST_FROM_FN_PTR(address, warning), c_rarg0); |
|
4688 pop_CPU_state(); |
|
4689 |
|
4690 movq(rsp, r12); |
|
4691 popq(r12); |
|
4692 } |
|
4693 |
|
4694 #ifndef PRODUCT |
|
4695 extern "C" void findpc(intptr_t x); |
|
4696 #endif |
|
4697 |
|
4698 void MacroAssembler::debug(char* msg, int64_t pc, int64_t regs[]) { |
|
4699 // In order to get locks to work, we need to fake a in_VM state |
|
4700 if (ShowMessageBoxOnError ) { |
|
4701 JavaThread* thread = JavaThread::current(); |
|
4702 JavaThreadState saved_state = thread->thread_state(); |
|
4703 thread->set_thread_state(_thread_in_vm); |
|
4704 #ifndef PRODUCT |
|
4705 if (CountBytecodes || TraceBytecodes || StopInterpreterAt) { |
|
4706 ttyLocker ttyl; |
|
4707 BytecodeCounter::print(); |
|
4708 } |
|
4709 #endif |
|
4710 // To see where a verify_oop failed, get $ebx+40/X for this frame. |
|
4711 // XXX correct this offset for amd64 |
|
4712 // This is the value of eip which points to where verify_oop will return. |
|
4713 if (os::message_box(msg, "Execution stopped, print registers?")) { |
|
4714 ttyLocker ttyl; |
|
4715 tty->print_cr("rip = 0x%016lx", pc); |
|
4716 #ifndef PRODUCT |
|
4717 tty->cr(); |
|
4718 findpc(pc); |
|
4719 tty->cr(); |
|
4720 #endif |
|
4721 tty->print_cr("rax = 0x%016lx", regs[15]); |
|
4722 tty->print_cr("rbx = 0x%016lx", regs[12]); |
|
4723 tty->print_cr("rcx = 0x%016lx", regs[14]); |
|
4724 tty->print_cr("rdx = 0x%016lx", regs[13]); |
|
4725 tty->print_cr("rdi = 0x%016lx", regs[8]); |
|
4726 tty->print_cr("rsi = 0x%016lx", regs[9]); |
|
4727 tty->print_cr("rbp = 0x%016lx", regs[10]); |
|
4728 tty->print_cr("rsp = 0x%016lx", regs[11]); |
|
4729 tty->print_cr("r8 = 0x%016lx", regs[7]); |
|
4730 tty->print_cr("r9 = 0x%016lx", regs[6]); |
|
4731 tty->print_cr("r10 = 0x%016lx", regs[5]); |
|
4732 tty->print_cr("r11 = 0x%016lx", regs[4]); |
|
4733 tty->print_cr("r12 = 0x%016lx", regs[3]); |
|
4734 tty->print_cr("r13 = 0x%016lx", regs[2]); |
|
4735 tty->print_cr("r14 = 0x%016lx", regs[1]); |
|
4736 tty->print_cr("r15 = 0x%016lx", regs[0]); |
|
4737 BREAKPOINT; |
|
4738 } |
|
4739 ThreadStateTransition::transition(thread, _thread_in_vm, saved_state); |
|
4740 } else { |
|
4741 ttyLocker ttyl; |
|
4742 ::tty->print_cr("=============== DEBUG MESSAGE: %s ================\n", |
|
4743 msg); |
|
4744 } |
|
4745 } |
|
4746 |
|
4747 void MacroAssembler::os_breakpoint() { |
|
4748 // instead of directly emitting a breakpoint, call os:breakpoint for |
|
4749 // better debugability |
|
4750 // This shouldn't need alignment, it's an empty function |
|
4751 call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint))); |
|
4752 } |
|
4753 |
|
4754 // Write serialization page so VM thread can do a pseudo remote membar. |
|
4755 // We use the current thread pointer to calculate a thread specific |
|
4756 // offset to write to within the page. This minimizes bus traffic |
|
4757 // due to cache line collision. |
|
4758 void MacroAssembler::serialize_memory(Register thread, |
|
4759 Register tmp) { |
|
4760 |
|
4761 movl(tmp, thread); |
|
4762 shrl(tmp, os::get_serialize_page_shift_count()); |
|
4763 andl(tmp, (os::vm_page_size() - sizeof(int))); |
|
4764 |
|
4765 Address index(noreg, tmp, Address::times_1); |
|
4766 ExternalAddress page(os::get_memory_serialize_page()); |
|
4767 |
|
4768 movptr(ArrayAddress(page, index), tmp); |
|
4769 } |
|
4770 |
|
4771 void MacroAssembler::verify_tlab() { |
|
4772 #ifdef ASSERT |
|
4773 if (UseTLAB) { |
|
4774 Label next, ok; |
|
4775 Register t1 = rsi; |
|
4776 |
|
4777 pushq(t1); |
|
4778 |
|
4779 movq(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset()))); |
|
4780 cmpq(t1, Address(r15_thread, in_bytes(JavaThread::tlab_start_offset()))); |
|
4781 jcc(Assembler::aboveEqual, next); |
|
4782 stop("assert(top >= start)"); |
|
4783 should_not_reach_here(); |
|
4784 |
|
4785 bind(next); |
|
4786 movq(t1, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset()))); |
|
4787 cmpq(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset()))); |
|
4788 jcc(Assembler::aboveEqual, ok); |
|
4789 stop("assert(top <= end)"); |
|
4790 should_not_reach_here(); |
|
4791 |
|
4792 bind(ok); |
|
4793 |
|
4794 popq(t1); |
|
4795 } |
|
4796 #endif |
|
4797 } |
|
4798 |
|
4799 // Defines obj, preserves var_size_in_bytes |
|
4800 void MacroAssembler::eden_allocate(Register obj, |
|
4801 Register var_size_in_bytes, |
|
4802 int con_size_in_bytes, |
|
4803 Register t1, |
|
4804 Label& slow_case) { |
|
4805 assert(obj == rax, "obj must be in rax for cmpxchg"); |
|
4806 assert_different_registers(obj, var_size_in_bytes, t1); |
|
4807 Register end = t1; |
|
4808 Label retry; |
|
4809 bind(retry); |
|
4810 ExternalAddress heap_top((address) Universe::heap()->top_addr()); |
|
4811 movptr(obj, heap_top); |
|
4812 if (var_size_in_bytes == noreg) { |
|
4813 leaq(end, Address(obj, con_size_in_bytes)); |
|
4814 } else { |
|
4815 leaq(end, Address(obj, var_size_in_bytes, Address::times_1)); |
|
4816 } |
|
4817 // if end < obj then we wrapped around => object too long => slow case |
|
4818 cmpq(end, obj); |
|
4819 jcc(Assembler::below, slow_case); |
|
4820 cmpptr(end, ExternalAddress((address) Universe::heap()->end_addr())); |
|
4821 |
|
4822 jcc(Assembler::above, slow_case); |
|
4823 // Compare obj with the top addr, and if still equal, store the new |
|
4824 // top addr in end at the address of the top addr pointer. Sets ZF |
|
4825 // if was equal, and clears it otherwise. Use lock prefix for |
|
4826 // atomicity on MPs. |
|
4827 if (os::is_MP()) { |
|
4828 lock(); |
|
4829 } |
|
4830 cmpxchgptr(end, heap_top); |
|
4831 // if someone beat us on the allocation, try again, otherwise continue |
|
4832 jcc(Assembler::notEqual, retry); |
|
4833 } |
|
4834 |
|
4835 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes. |
|
4836 void MacroAssembler::tlab_allocate(Register obj, |
|
4837 Register var_size_in_bytes, |
|
4838 int con_size_in_bytes, |
|
4839 Register t1, |
|
4840 Register t2, |
|
4841 Label& slow_case) { |
|
4842 assert_different_registers(obj, t1, t2); |
|
4843 assert_different_registers(obj, var_size_in_bytes, t1); |
|
4844 Register end = t2; |
|
4845 |
|
4846 verify_tlab(); |
|
4847 |
|
4848 movq(obj, Address(r15_thread, JavaThread::tlab_top_offset())); |
|
4849 if (var_size_in_bytes == noreg) { |
|
4850 leaq(end, Address(obj, con_size_in_bytes)); |
|
4851 } else { |
|
4852 leaq(end, Address(obj, var_size_in_bytes, Address::times_1)); |
|
4853 } |
|
4854 cmpq(end, Address(r15_thread, JavaThread::tlab_end_offset())); |
|
4855 jcc(Assembler::above, slow_case); |
|
4856 |
|
4857 // update the tlab top pointer |
|
4858 movq(Address(r15_thread, JavaThread::tlab_top_offset()), end); |
|
4859 |
|
4860 // recover var_size_in_bytes if necessary |
|
4861 if (var_size_in_bytes == end) { |
|
4862 subq(var_size_in_bytes, obj); |
|
4863 } |
|
4864 verify_tlab(); |
|
4865 } |
|
4866 |
|
4867 // Preserves rbx and rdx. |
|
4868 void MacroAssembler::tlab_refill(Label& retry, |
|
4869 Label& try_eden, |
|
4870 Label& slow_case) { |
|
4871 Register top = rax; |
|
4872 Register t1 = rcx; |
|
4873 Register t2 = rsi; |
|
4874 Register t3 = r10; |
|
4875 Register thread_reg = r15_thread; |
|
4876 assert_different_registers(top, thread_reg, t1, t2, t3, |
|
4877 /* preserve: */ rbx, rdx); |
|
4878 Label do_refill, discard_tlab; |
|
4879 |
|
4880 if (CMSIncrementalMode || !Universe::heap()->supports_inline_contig_alloc()) { |
|
4881 // No allocation in the shared eden. |
|
4882 jmp(slow_case); |
|
4883 } |
|
4884 |
|
4885 movq(top, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset()))); |
|
4886 movq(t1, Address(thread_reg, in_bytes(JavaThread::tlab_end_offset()))); |
|
4887 |
|
4888 // calculate amount of free space |
|
4889 subq(t1, top); |
|
4890 shrq(t1, LogHeapWordSize); |
|
4891 |
|
4892 // Retain tlab and allocate object in shared space if |
|
4893 // the amount free in the tlab is too large to discard. |
|
4894 cmpq(t1, Address(thread_reg, // size_t |
|
4895 in_bytes(JavaThread::tlab_refill_waste_limit_offset()))); |
|
4896 jcc(Assembler::lessEqual, discard_tlab); |
|
4897 |
|
4898 // Retain |
|
4899 mov64(t2, ThreadLocalAllocBuffer::refill_waste_limit_increment()); |
|
4900 addq(Address(thread_reg, // size_t |
|
4901 in_bytes(JavaThread::tlab_refill_waste_limit_offset())), |
|
4902 t2); |
|
4903 if (TLABStats) { |
|
4904 // increment number of slow_allocations |
|
4905 addl(Address(thread_reg, // unsigned int |
|
4906 in_bytes(JavaThread::tlab_slow_allocations_offset())), |
|
4907 1); |
|
4908 } |
|
4909 jmp(try_eden); |
|
4910 |
|
4911 bind(discard_tlab); |
|
4912 if (TLABStats) { |
|
4913 // increment number of refills |
|
4914 addl(Address(thread_reg, // unsigned int |
|
4915 in_bytes(JavaThread::tlab_number_of_refills_offset())), |
|
4916 1); |
|
4917 // accumulate wastage -- t1 is amount free in tlab |
|
4918 addl(Address(thread_reg, // unsigned int |
|
4919 in_bytes(JavaThread::tlab_fast_refill_waste_offset())), |
|
4920 t1); |
|
4921 } |
|
4922 |
|
4923 // if tlab is currently allocated (top or end != null) then |
|
4924 // fill [top, end + alignment_reserve) with array object |
|
4925 testq(top, top); |
|
4926 jcc(Assembler::zero, do_refill); |
|
4927 |
|
4928 // set up the mark word |
|
4929 mov64(t3, (int64_t) markOopDesc::prototype()->copy_set_hash(0x2)); |
|
4930 movq(Address(top, oopDesc::mark_offset_in_bytes()), t3); |
|
4931 // set the length to the remaining space |
|
4932 subq(t1, typeArrayOopDesc::header_size(T_INT)); |
|
4933 addq(t1, (int)ThreadLocalAllocBuffer::alignment_reserve()); |
|
4934 shlq(t1, log2_intptr(HeapWordSize / sizeof(jint))); |
|
4935 movq(Address(top, arrayOopDesc::length_offset_in_bytes()), t1); |
|
4936 // set klass to intArrayKlass |
|
4937 movptr(t1, ExternalAddress((address) Universe::intArrayKlassObj_addr())); |
|
4938 // store klass last. concurrent gcs assumes klass length is valid if |
|
4939 // klass field is not null. |
|
4940 store_klass(top, t1); |
|
4941 |
|
4942 // refill the tlab with an eden allocation |
|
4943 bind(do_refill); |
|
4944 movq(t1, Address(thread_reg, in_bytes(JavaThread::tlab_size_offset()))); |
|
4945 shlq(t1, LogHeapWordSize); |
|
4946 // add object_size ?? |
|
4947 eden_allocate(top, t1, 0, t2, slow_case); |
|
4948 |
|
4949 // Check that t1 was preserved in eden_allocate. |
|
4950 #ifdef ASSERT |
|
4951 if (UseTLAB) { |
|
4952 Label ok; |
|
4953 Register tsize = rsi; |
|
4954 assert_different_registers(tsize, thread_reg, t1); |
|
4955 pushq(tsize); |
|
4956 movq(tsize, Address(thread_reg, in_bytes(JavaThread::tlab_size_offset()))); |
|
4957 shlq(tsize, LogHeapWordSize); |
|
4958 cmpq(t1, tsize); |
|
4959 jcc(Assembler::equal, ok); |
|
4960 stop("assert(t1 != tlab size)"); |
|
4961 should_not_reach_here(); |
|
4962 |
|
4963 bind(ok); |
|
4964 popq(tsize); |
|
4965 } |
|
4966 #endif |
|
4967 movq(Address(thread_reg, in_bytes(JavaThread::tlab_start_offset())), top); |
|
4968 movq(Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())), top); |
|
4969 addq(top, t1); |
|
4970 subq(top, (int)ThreadLocalAllocBuffer::alignment_reserve_in_bytes()); |
|
4971 movq(Address(thread_reg, in_bytes(JavaThread::tlab_end_offset())), top); |
|
4972 verify_tlab(); |
|
4973 jmp(retry); |
|
4974 } |
|
4975 |
|
4976 |
|
4977 int MacroAssembler::biased_locking_enter(Register lock_reg, Register obj_reg, Register swap_reg, Register tmp_reg, |
|
4978 bool swap_reg_contains_mark, |
|
4979 Label& done, Label* slow_case, |
|
4980 BiasedLockingCounters* counters) { |
|
4981 assert(UseBiasedLocking, "why call this otherwise?"); |
|
4982 assert(swap_reg == rax, "swap_reg must be rax for cmpxchgq"); |
|
4983 assert(tmp_reg != noreg, "tmp_reg must be supplied"); |
|
4984 assert_different_registers(lock_reg, obj_reg, swap_reg, tmp_reg); |
|
4985 assert(markOopDesc::age_shift == markOopDesc::lock_bits + markOopDesc::biased_lock_bits, "biased locking makes assumptions about bit layout"); |
|
4986 Address mark_addr (obj_reg, oopDesc::mark_offset_in_bytes()); |
|
4987 Address saved_mark_addr(lock_reg, 0); |
|
4988 |
|
4989 if (PrintBiasedLockingStatistics && counters == NULL) |
|
4990 counters = BiasedLocking::counters(); |
|
4991 |
|
4992 // Biased locking |
|
4993 // See whether the lock is currently biased toward our thread and |
|
4994 // whether the epoch is still valid |
|
4995 // Note that the runtime guarantees sufficient alignment of JavaThread |
|
4996 // pointers to allow age to be placed into low bits |
|
4997 // First check to see whether biasing is even enabled for this object |
|
4998 Label cas_label; |
|
4999 int null_check_offset = -1; |
|
5000 if (!swap_reg_contains_mark) { |
|
5001 null_check_offset = offset(); |
|
5002 movq(swap_reg, mark_addr); |
|
5003 } |
|
5004 movq(tmp_reg, swap_reg); |
|
5005 andq(tmp_reg, markOopDesc::biased_lock_mask_in_place); |
|
5006 cmpq(tmp_reg, markOopDesc::biased_lock_pattern); |
|
5007 jcc(Assembler::notEqual, cas_label); |
|
5008 // The bias pattern is present in the object's header. Need to check |
|
5009 // whether the bias owner and the epoch are both still current. |
|
5010 load_prototype_header(tmp_reg, obj_reg); |
|
5011 orq(tmp_reg, r15_thread); |
|
5012 xorq(tmp_reg, swap_reg); |
|
5013 andq(tmp_reg, ~((int) markOopDesc::age_mask_in_place)); |
|
5014 if (counters != NULL) { |
|
5015 cond_inc32(Assembler::zero, |
|
5016 ExternalAddress((address) counters->anonymously_biased_lock_entry_count_addr())); |
|
5017 } |
|
5018 jcc(Assembler::equal, done); |
|
5019 |
|
5020 Label try_revoke_bias; |
|
5021 Label try_rebias; |
|
5022 |
|
5023 // At this point we know that the header has the bias pattern and |
|
5024 // that we are not the bias owner in the current epoch. We need to |
|
5025 // figure out more details about the state of the header in order to |
|
5026 // know what operations can be legally performed on the object's |
|
5027 // header. |
|
5028 |
|
5029 // If the low three bits in the xor result aren't clear, that means |
|
5030 // the prototype header is no longer biased and we have to revoke |
|
5031 // the bias on this object. |
|
5032 testq(tmp_reg, markOopDesc::biased_lock_mask_in_place); |
|
5033 jcc(Assembler::notZero, try_revoke_bias); |
|
5034 |
|
5035 // Biasing is still enabled for this data type. See whether the |
|
5036 // epoch of the current bias is still valid, meaning that the epoch |
|
5037 // bits of the mark word are equal to the epoch bits of the |
|
5038 // prototype header. (Note that the prototype header's epoch bits |
|
5039 // only change at a safepoint.) If not, attempt to rebias the object |
|
5040 // toward the current thread. Note that we must be absolutely sure |
|
5041 // that the current epoch is invalid in order to do this because |
|
5042 // otherwise the manipulations it performs on the mark word are |
|
5043 // illegal. |
|
5044 testq(tmp_reg, markOopDesc::epoch_mask_in_place); |
|
5045 jcc(Assembler::notZero, try_rebias); |
|
5046 |
|
5047 // The epoch of the current bias is still valid but we know nothing |
|
5048 // about the owner; it might be set or it might be clear. Try to |
|
5049 // acquire the bias of the object using an atomic operation. If this |
|
5050 // fails we will go in to the runtime to revoke the object's bias. |
|
5051 // Note that we first construct the presumed unbiased header so we |
|
5052 // don't accidentally blow away another thread's valid bias. |
|
5053 andq(swap_reg, |
|
5054 markOopDesc::biased_lock_mask_in_place | markOopDesc::age_mask_in_place | markOopDesc::epoch_mask_in_place); |
|
5055 movq(tmp_reg, swap_reg); |
|
5056 orq(tmp_reg, r15_thread); |
|
5057 if (os::is_MP()) { |
|
5058 lock(); |
|
5059 } |
|
5060 cmpxchgq(tmp_reg, Address(obj_reg, 0)); |
|
5061 // If the biasing toward our thread failed, this means that |
|
5062 // another thread succeeded in biasing it toward itself and we |
|
5063 // need to revoke that bias. The revocation will occur in the |
|
5064 // interpreter runtime in the slow case. |
|
5065 if (counters != NULL) { |
|
5066 cond_inc32(Assembler::zero, |
|
5067 ExternalAddress((address) counters->anonymously_biased_lock_entry_count_addr())); |
|
5068 } |
|
5069 if (slow_case != NULL) { |
|
5070 jcc(Assembler::notZero, *slow_case); |
|
5071 } |
|
5072 jmp(done); |
|
5073 |
|
5074 bind(try_rebias); |
|
5075 // At this point we know the epoch has expired, meaning that the |
|
5076 // current "bias owner", if any, is actually invalid. Under these |
|
5077 // circumstances _only_, we are allowed to use the current header's |
|
5078 // value as the comparison value when doing the cas to acquire the |
|
5079 // bias in the current epoch. In other words, we allow transfer of |
|
5080 // the bias from one thread to another directly in this situation. |
|
5081 // |
|
5082 // FIXME: due to a lack of registers we currently blow away the age |
|
5083 // bits in this situation. Should attempt to preserve them. |
|
5084 load_prototype_header(tmp_reg, obj_reg); |
|
5085 orq(tmp_reg, r15_thread); |
|
5086 if (os::is_MP()) { |
|
5087 lock(); |
|
5088 } |
|
5089 cmpxchgq(tmp_reg, Address(obj_reg, 0)); |
|
5090 // If the biasing toward our thread failed, then another thread |
|
5091 // succeeded in biasing it toward itself and we need to revoke that |
|
5092 // bias. The revocation will occur in the runtime in the slow case. |
|
5093 if (counters != NULL) { |
|
5094 cond_inc32(Assembler::zero, |
|
5095 ExternalAddress((address) counters->rebiased_lock_entry_count_addr())); |
|
5096 } |
|
5097 if (slow_case != NULL) { |
|
5098 jcc(Assembler::notZero, *slow_case); |
|
5099 } |
|
5100 jmp(done); |
|
5101 |
|
5102 bind(try_revoke_bias); |
|
5103 // The prototype mark in the klass doesn't have the bias bit set any |
|
5104 // more, indicating that objects of this data type are not supposed |
|
5105 // to be biased any more. We are going to try to reset the mark of |
|
5106 // this object to the prototype value and fall through to the |
|
5107 // CAS-based locking scheme. Note that if our CAS fails, it means |
|
5108 // that another thread raced us for the privilege of revoking the |
|
5109 // bias of this particular object, so it's okay to continue in the |
|
5110 // normal locking code. |
|
5111 // |
|
5112 // FIXME: due to a lack of registers we currently blow away the age |
|
5113 // bits in this situation. Should attempt to preserve them. |
|
5114 load_prototype_header(tmp_reg, obj_reg); |
|
5115 if (os::is_MP()) { |
|
5116 lock(); |
|
5117 } |
|
5118 cmpxchgq(tmp_reg, Address(obj_reg, 0)); |
|
5119 // Fall through to the normal CAS-based lock, because no matter what |
|
5120 // the result of the above CAS, some thread must have succeeded in |
|
5121 // removing the bias bit from the object's header. |
|
5122 if (counters != NULL) { |
|
5123 cond_inc32(Assembler::zero, |
|
5124 ExternalAddress((address) counters->revoked_lock_entry_count_addr())); |
|
5125 } |
|
5126 |
|
5127 bind(cas_label); |
|
5128 |
|
5129 return null_check_offset; |
|
5130 } |
|
5131 |
|
5132 |
|
5133 void MacroAssembler::biased_locking_exit(Register obj_reg, Register temp_reg, Label& done) { |
|
5134 assert(UseBiasedLocking, "why call this otherwise?"); |
|
5135 |
|
5136 // Check for biased locking unlock case, which is a no-op |
|
5137 // Note: we do not have to check the thread ID for two reasons. |
|
5138 // First, the interpreter checks for IllegalMonitorStateException at |
|
5139 // a higher level. Second, if the bias was revoked while we held the |
|
5140 // lock, the object could not be rebiased toward another thread, so |
|
5141 // the bias bit would be clear. |
|
5142 movq(temp_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); |
|
5143 andq(temp_reg, markOopDesc::biased_lock_mask_in_place); |
|
5144 cmpq(temp_reg, markOopDesc::biased_lock_pattern); |
|
5145 jcc(Assembler::equal, done); |
|
5146 } |
|
5147 |
|
5148 |
|
5149 void MacroAssembler::load_klass(Register dst, Register src) { |
|
5150 if (UseCompressedOops) { |
|
5151 movl(dst, Address(src, oopDesc::klass_offset_in_bytes())); |
|
5152 decode_heap_oop_not_null(dst); |
|
5153 } else { |
|
5154 movq(dst, Address(src, oopDesc::klass_offset_in_bytes())); |
|
5155 } |
|
5156 } |
|
5157 |
|
5158 void MacroAssembler::load_prototype_header(Register dst, Register src) { |
|
5159 if (UseCompressedOops) { |
|
5160 movl(dst, Address(src, oopDesc::klass_offset_in_bytes())); |
|
5161 movq(dst, Address(r12_heapbase, dst, Address::times_8, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); |
|
5162 } else { |
|
5163 movq(dst, Address(src, oopDesc::klass_offset_in_bytes())); |
|
5164 movq(dst, Address(dst, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); |
|
5165 } |
|
5166 } |
|
5167 |
|
5168 void MacroAssembler::store_klass(Register dst, Register src) { |
|
5169 if (UseCompressedOops) { |
|
5170 encode_heap_oop_not_null(src); |
|
5171 movl(Address(dst, oopDesc::klass_offset_in_bytes()), src); |
|
5172 } else { |
|
5173 movq(Address(dst, oopDesc::klass_offset_in_bytes()), src); |
|
5174 } |
|
5175 } |
|
5176 |
|
5177 void MacroAssembler::store_klass_gap(Register dst, Register src) { |
|
5178 if (UseCompressedOops) { |
|
5179 // Store to klass gap in destination |
|
5180 movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src); |
|
5181 } |
|
5182 } |
|
5183 |
|
5184 void MacroAssembler::load_heap_oop(Register dst, Address src) { |
|
5185 if (UseCompressedOops) { |
|
5186 movl(dst, src); |
|
5187 decode_heap_oop(dst); |
|
5188 } else { |
|
5189 movq(dst, src); |
|
5190 } |
|
5191 } |
|
5192 |
|
5193 void MacroAssembler::store_heap_oop(Address dst, Register src) { |
|
5194 if (UseCompressedOops) { |
|
5195 assert(!dst.uses(src), "not enough registers"); |
|
5196 encode_heap_oop(src); |
|
5197 movl(dst, src); |
|
5198 } else { |
|
5199 movq(dst, src); |
|
5200 } |
|
5201 } |
|
5202 |
|
5203 // Algorithm must match oop.inline.hpp encode_heap_oop. |
|
5204 void MacroAssembler::encode_heap_oop(Register r) { |
|
5205 assert (UseCompressedOops, "should be compressed"); |
|
5206 #ifdef ASSERT |
|
5207 if (CheckCompressedOops) { |
|
5208 Label ok; |
|
5209 pushq(rscratch1); // cmpptr trashes rscratch1 |
|
5210 cmpptr(r12_heapbase, ExternalAddress((address)Universe::heap_base_addr())); |
|
5211 jcc(Assembler::equal, ok); |
|
5212 stop("MacroAssembler::encode_heap_oop: heap base corrupted?"); |
|
5213 bind(ok); |
|
5214 popq(rscratch1); |
|
5215 } |
|
5216 #endif |
|
5217 verify_oop(r, "broken oop in encode_heap_oop"); |
|
5218 testq(r, r); |
|
5219 cmovq(Assembler::equal, r, r12_heapbase); |
|
5220 subq(r, r12_heapbase); |
|
5221 shrq(r, LogMinObjAlignmentInBytes); |
|
5222 } |
|
5223 |
|
5224 void MacroAssembler::encode_heap_oop_not_null(Register r) { |
|
5225 assert (UseCompressedOops, "should be compressed"); |
|
5226 #ifdef ASSERT |
|
5227 if (CheckCompressedOops) { |
|
5228 Label ok; |
|
5229 testq(r, r); |
|
5230 jcc(Assembler::notEqual, ok); |
|
5231 stop("null oop passed to encode_heap_oop_not_null"); |
|
5232 bind(ok); |
|
5233 } |
|
5234 #endif |
|
5235 verify_oop(r, "broken oop in encode_heap_oop_not_null"); |
|
5236 subq(r, r12_heapbase); |
|
5237 shrq(r, LogMinObjAlignmentInBytes); |
|
5238 } |
|
5239 |
|
5240 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) { |
|
5241 assert (UseCompressedOops, "should be compressed"); |
|
5242 #ifdef ASSERT |
|
5243 if (CheckCompressedOops) { |
|
5244 Label ok; |
|
5245 testq(src, src); |
|
5246 jcc(Assembler::notEqual, ok); |
|
5247 stop("null oop passed to encode_heap_oop_not_null2"); |
|
5248 bind(ok); |
|
5249 } |
|
5250 #endif |
|
5251 verify_oop(src, "broken oop in encode_heap_oop_not_null2"); |
|
5252 if (dst != src) { |
|
5253 movq(dst, src); |
|
5254 } |
|
5255 subq(dst, r12_heapbase); |
|
5256 shrq(dst, LogMinObjAlignmentInBytes); |
|
5257 } |
|
5258 |
|
5259 void MacroAssembler::decode_heap_oop(Register r) { |
|
5260 assert (UseCompressedOops, "should be compressed"); |
|
5261 #ifdef ASSERT |
|
5262 if (CheckCompressedOops) { |
|
5263 Label ok; |
|
5264 pushq(rscratch1); |
|
5265 cmpptr(r12_heapbase, |
|
5266 ExternalAddress((address)Universe::heap_base_addr())); |
|
5267 jcc(Assembler::equal, ok); |
|
5268 stop("MacroAssembler::decode_heap_oop: heap base corrupted?"); |
|
5269 bind(ok); |
|
5270 popq(rscratch1); |
|
5271 } |
|
5272 #endif |
|
5273 |
|
5274 Label done; |
|
5275 shlq(r, LogMinObjAlignmentInBytes); |
|
5276 jccb(Assembler::equal, done); |
|
5277 addq(r, r12_heapbase); |
|
5278 #if 0 |
|
5279 // alternate decoding probably a wash. |
|
5280 testq(r, r); |
|
5281 jccb(Assembler::equal, done); |
|
5282 leaq(r, Address(r12_heapbase, r, Address::times_8, 0)); |
|
5283 #endif |
|
5284 bind(done); |
|
5285 verify_oop(r, "broken oop in decode_heap_oop"); |
|
5286 } |
|
5287 |
|
5288 void MacroAssembler::decode_heap_oop_not_null(Register r) { |
|
5289 assert (UseCompressedOops, "should only be used for compressed headers"); |
|
5290 // Cannot assert, unverified entry point counts instructions (see .ad file) |
|
5291 // vtableStubs also counts instructions in pd_code_size_limit. |
|
5292 // Also do not verify_oop as this is called by verify_oop. |
|
5293 assert(Address::times_8 == LogMinObjAlignmentInBytes, "decode alg wrong"); |
|
5294 leaq(r, Address(r12_heapbase, r, Address::times_8, 0)); |
|
5295 } |
|
5296 |
|
5297 void MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) { |
|
5298 assert (UseCompressedOops, "should only be used for compressed headers"); |
|
5299 // Cannot assert, unverified entry point counts instructions (see .ad file) |
|
5300 // vtableStubs also counts instructions in pd_code_size_limit. |
|
5301 // Also do not verify_oop as this is called by verify_oop. |
|
5302 assert(Address::times_8 == LogMinObjAlignmentInBytes, "decode alg wrong"); |
|
5303 leaq(dst, Address(r12_heapbase, src, Address::times_8, 0)); |
|
5304 } |
|
5305 |
|
5306 void MacroAssembler::set_narrow_oop(Register dst, jobject obj) { |
|
5307 assert(oop_recorder() != NULL, "this assembler needs an OopRecorder"); |
|
5308 int oop_index = oop_recorder()->find_index(obj); |
|
5309 RelocationHolder rspec = oop_Relocation::spec(oop_index); |
|
5310 |
|
5311 // movl dst,obj |
|
5312 InstructionMark im(this); |
|
5313 int encode = prefix_and_encode(dst->encoding()); |
|
5314 emit_byte(0xB8 | encode); |
|
5315 emit_data(oop_index, rspec, narrow_oop_operand); |
|
5316 } |
|
5317 |
|
5318 |
|
5319 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) { |
|
5320 switch (cond) { |
|
5321 // Note some conditions are synonyms for others |
|
5322 case Assembler::zero: return Assembler::notZero; |
|
5323 case Assembler::notZero: return Assembler::zero; |
|
5324 case Assembler::less: return Assembler::greaterEqual; |
|
5325 case Assembler::lessEqual: return Assembler::greater; |
|
5326 case Assembler::greater: return Assembler::lessEqual; |
|
5327 case Assembler::greaterEqual: return Assembler::less; |
|
5328 case Assembler::below: return Assembler::aboveEqual; |
|
5329 case Assembler::belowEqual: return Assembler::above; |
|
5330 case Assembler::above: return Assembler::belowEqual; |
|
5331 case Assembler::aboveEqual: return Assembler::below; |
|
5332 case Assembler::overflow: return Assembler::noOverflow; |
|
5333 case Assembler::noOverflow: return Assembler::overflow; |
|
5334 case Assembler::negative: return Assembler::positive; |
|
5335 case Assembler::positive: return Assembler::negative; |
|
5336 case Assembler::parity: return Assembler::noParity; |
|
5337 case Assembler::noParity: return Assembler::parity; |
|
5338 } |
|
5339 ShouldNotReachHere(); return Assembler::overflow; |
|
5340 } |
|
5341 |
|
5342 |
|
5343 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr) { |
|
5344 Condition negated_cond = negate_condition(cond); |
|
5345 Label L; |
|
5346 jcc(negated_cond, L); |
|
5347 atomic_incl(counter_addr); |
|
5348 bind(L); |
|
5349 } |
|
5350 |
|
5351 void MacroAssembler::atomic_incl(AddressLiteral counter_addr) { |
|
5352 pushfq(); |
|
5353 if (os::is_MP()) |
|
5354 lock(); |
|
5355 incrementl(counter_addr); |
|
5356 popfq(); |
|
5357 } |
|
5358 |
|
5359 SkipIfEqual::SkipIfEqual( |
|
5360 MacroAssembler* masm, const bool* flag_addr, bool value) { |
|
5361 _masm = masm; |
|
5362 _masm->cmp8(ExternalAddress((address)flag_addr), value); |
|
5363 _masm->jcc(Assembler::equal, _label); |
|
5364 } |
|
5365 |
|
5366 SkipIfEqual::~SkipIfEqual() { |
|
5367 _masm->bind(_label); |
|
5368 } |
|
5369 |
|
5370 void MacroAssembler::bang_stack_size(Register size, Register tmp) { |
|
5371 movq(tmp, rsp); |
|
5372 // Bang stack for total size given plus shadow page size. |
|
5373 // Bang one page at a time because large size can bang beyond yellow and |
|
5374 // red zones. |
|
5375 Label loop; |
|
5376 bind(loop); |
|
5377 movl(Address(tmp, (-os::vm_page_size())), size ); |
|
5378 subq(tmp, os::vm_page_size()); |
|
5379 subl(size, os::vm_page_size()); |
|
5380 jcc(Assembler::greater, loop); |
|
5381 |
|
5382 // Bang down shadow pages too. |
|
5383 // The -1 because we already subtracted 1 page. |
|
5384 for (int i = 0; i< StackShadowPages-1; i++) { |
|
5385 movq(Address(tmp, (-i*os::vm_page_size())), size ); |
|
5386 } |
|
5387 } |
|
5388 |
|
5389 void MacroAssembler::reinit_heapbase() { |
|
5390 if (UseCompressedOops) { |
|
5391 movptr(r12_heapbase, ExternalAddress((address)Universe::heap_base_addr())); |
|
5392 } |
|
5393 } |
|