Wed, 07 May 2008 08:06:46 -0700
6603011: RFE: Optimize long division
Summary: Transform long division by constant into multiply
Reviewed-by: never, kvn
1 /*
2 * Copyright 2003-2007 Sun Microsystems, Inc. All Rights Reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 * CA 95054 USA or visit www.sun.com if you need additional information or
21 * have any questions.
22 *
23 */
25 #include "incls/_precompiled.incl"
26 #include "incls/_assembler_x86_64.cpp.incl"
28 // Implementation of AddressLiteral
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 }
63 // Implementation of Address
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 }
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 }
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 }
120 // Implementation of Assembler
121 int AbstractAssembler::code_fill_byte() {
122 return (u_char)'\xF4'; // hlt
123 }
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.
128 bool Assembler::reachable(AddressLiteral adr) {
129 int64_t disp;
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 }
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.
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;
167 disp = (int64_t)adr._target - ((int64_t)_code_pos + sizeof(int));
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.
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 }
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 }
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.
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 }
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 }
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 }
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 }
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 }
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 }
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 }
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
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.
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);
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 }
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.
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;
550 assert(inst_mark() != NULL, "must be inside InstructionMark");
551 address next_ip = pc() + sizeof(int32_t) + rip_relative_correction;
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 }
568 // Secret local extension to Assembler::WhichOperand:
569 #define end_pc_operand (_WhichOperand_limit)
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.
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.
582 // If "which" is end_pc_operand, find the end of the instruction.
584 address ip = inst;
585 bool is_64bit = false;
587 debug_only(bool has_disp32 = false);
588 int tail_size = 0; // other random bytes (#32, #16, etc.) at end of insn
590 again_after_prefix:
591 switch (0xFF & *ip++) {
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)
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;
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;
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;
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;
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();
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;
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 return ip;
689 case 0x69: // imul r, a, #32
690 case 0xC7: // movl a, #32(oop?)
691 tail_size = 4;
692 debug_only(has_disp32 = true); // has both kinds of operands!
693 break;
695 case 0x0F: // movx..., etc.
696 switch (0xFF & *ip++) {
697 case 0x12: // movlps
698 case 0x28: // movaps
699 case 0x2E: // ucomiss
700 case 0x2F: // comiss
701 case 0x54: // andps
702 case 0x57: // xorps
703 case 0x6E: // movd
704 case 0x7E: // movd
705 case 0xAE: // ldmxcsr a
706 debug_only(has_disp32 = true); // has both kinds of operands!
707 break;
708 case 0xAD: // shrd r, a, %cl
709 case 0xAF: // imul r, a
710 case 0xBE: // movsbl r, a
711 case 0xBF: // movswl r, a
712 case 0xB6: // movzbl r, a
713 case 0xB7: // movzwl r, a
714 case REP16(0x40): // cmovl cc, r, a
715 case 0xB0: // cmpxchgb
716 case 0xB1: // cmpxchg
717 case 0xC1: // xaddl
718 case 0xC7: // cmpxchg8
719 case REP16(0x90): // setcc a
720 debug_only(has_disp32 = true);
721 // fall out of the switch to decode the address
722 break;
723 case 0xAC: // shrd r, a, #8
724 debug_only(has_disp32 = true);
725 tail_size = 1; // the imm8
726 break;
727 case REP16(0x80): // jcc rdisp32
728 if (which == end_pc_operand) return ip + 4;
729 assert(which == call32_operand, "jcc has no disp32 or imm64");
730 return ip;
731 default:
732 ShouldNotReachHere();
733 }
734 break;
736 case 0x81: // addl a, #32; addl r, #32
737 // also: orl, adcl, sbbl, andl, subl, xorl, cmpl
738 tail_size = 4;
739 debug_only(has_disp32 = true); // has both kinds of operands!
740 break;
742 case 0x83: // addl a, #8; addl r, #8
743 // also: orl, adcl, sbbl, andl, subl, xorl, cmpl
744 debug_only(has_disp32 = true); // has both kinds of operands!
745 tail_size = 1;
746 break;
748 case 0x9B:
749 switch (0xFF & *ip++) {
750 case 0xD9: // fnstcw a
751 debug_only(has_disp32 = true);
752 break;
753 default:
754 ShouldNotReachHere();
755 }
756 break;
758 case REP4(0x00): // addb a, r; addl a, r; addb r, a; addl r, a
759 case REP4(0x10): // adc...
760 case REP4(0x20): // and...
761 case REP4(0x30): // xor...
762 case REP4(0x08): // or...
763 case REP4(0x18): // sbb...
764 case REP4(0x28): // sub...
765 case 0xF7: // mull a
766 case 0x87: // xchg r, a
767 debug_only(has_disp32 = true);
768 break;
769 case REP4(0x38): // cmp...
770 case 0x8D: // lea r, a
771 case 0x85: // test r, a
772 debug_only(has_disp32 = true); // has both kinds of operands!
773 break;
775 case 0xC1: // sal a, #8; sar a, #8; shl a, #8; shr a, #8
776 case 0xC6: // movb a, #8
777 case 0x80: // cmpb a, #8
778 case 0x6B: // imul r, a, #8
779 debug_only(has_disp32 = true); // has both kinds of operands!
780 tail_size = 1; // the imm8
781 break;
783 case 0xE8: // call rdisp32
784 case 0xE9: // jmp rdisp32
785 if (which == end_pc_operand) return ip + 4;
786 assert(which == call32_operand, "call has no disp32 or imm32");
787 return ip;
789 case 0xD1: // sal a, 1; sar a, 1; shl a, 1; shr a, 1
790 case 0xD3: // sal a, %cl; sar a, %cl; shl a, %cl; shr a, %cl
791 case 0xD9: // fld_s a; fst_s a; fstp_s a; fldcw a
792 case 0xDD: // fld_d a; fst_d a; fstp_d a
793 case 0xDB: // fild_s a; fistp_s a; fld_x a; fstp_x a
794 case 0xDF: // fild_d a; fistp_d a
795 case 0xD8: // fadd_s a; fsubr_s a; fmul_s a; fdivr_s a; fcomp_s a
796 case 0xDC: // fadd_d a; fsubr_d a; fmul_d a; fdivr_d a; fcomp_d a
797 case 0xDE: // faddp_d a; fsubrp_d a; fmulp_d a; fdivrp_d a; fcompp_d a
798 debug_only(has_disp32 = true);
799 break;
801 case 0xF3: // For SSE
802 case 0xF2: // For SSE2
803 switch (0xFF & *ip++) {
804 case REX:
805 case REX_B:
806 case REX_X:
807 case REX_XB:
808 case REX_R:
809 case REX_RB:
810 case REX_RX:
811 case REX_RXB:
812 case REX_W:
813 case REX_WB:
814 case REX_WX:
815 case REX_WXB:
816 case REX_WR:
817 case REX_WRB:
818 case REX_WRX:
819 case REX_WRXB:
820 ip++;
821 default:
822 ip++;
823 }
824 debug_only(has_disp32 = true); // has both kinds of operands!
825 break;
827 default:
828 ShouldNotReachHere();
830 #undef REP8
831 #undef REP16
832 }
834 assert(which != call32_operand, "instruction is not a call, jmp, or jcc");
835 assert(which != imm64_operand, "instruction is not a movq reg, imm64");
836 assert(which != disp32_operand || has_disp32, "instruction has no disp32 field");
838 // parse the output of emit_operand
839 int op2 = 0xFF & *ip++;
840 int base = op2 & 0x07;
841 int op3 = -1;
842 const int b100 = 4;
843 const int b101 = 5;
844 if (base == b100 && (op2 >> 6) != 3) {
845 op3 = 0xFF & *ip++;
846 base = op3 & 0x07; // refetch the base
847 }
848 // now ip points at the disp (if any)
850 switch (op2 >> 6) {
851 case 0:
852 // [00 reg 100][ss index base]
853 // [00 reg 100][00 100 esp]
854 // [00 reg base]
855 // [00 reg 100][ss index 101][disp32]
856 // [00 reg 101] [disp32]
858 if (base == b101) {
859 if (which == disp32_operand)
860 return ip; // caller wants the disp32
861 ip += 4; // skip the disp32
862 }
863 break;
865 case 1:
866 // [01 reg 100][ss index base][disp8]
867 // [01 reg 100][00 100 esp][disp8]
868 // [01 reg base] [disp8]
869 ip += 1; // skip the disp8
870 break;
872 case 2:
873 // [10 reg 100][ss index base][disp32]
874 // [10 reg 100][00 100 esp][disp32]
875 // [10 reg base] [disp32]
876 if (which == disp32_operand)
877 return ip; // caller wants the disp32
878 ip += 4; // skip the disp32
879 break;
881 case 3:
882 // [11 reg base] (not a memory addressing mode)
883 break;
884 }
886 if (which == end_pc_operand) {
887 return ip + tail_size;
888 }
890 assert(0, "fix locate_operand");
891 return ip;
892 }
894 address Assembler::locate_next_instruction(address inst) {
895 // Secretly share code with locate_operand:
896 return locate_operand(inst, end_pc_operand);
897 }
899 #ifdef ASSERT
900 void Assembler::check_relocation(RelocationHolder const& rspec, int format) {
901 address inst = inst_mark();
902 assert(inst != NULL && inst < pc(),
903 "must point to beginning of instruction");
904 address opnd;
906 Relocation* r = rspec.reloc();
907 if (r->type() == relocInfo::none) {
908 return;
909 } else if (r->is_call() || format == call32_operand) {
910 opnd = locate_operand(inst, call32_operand);
911 } else if (r->is_data()) {
912 assert(format == imm64_operand || format == disp32_operand, "format ok");
913 opnd = locate_operand(inst, (WhichOperand) format);
914 } else {
915 assert(format == 0, "cannot specify a format");
916 return;
917 }
918 assert(opnd == pc(), "must put operand where relocs can find it");
919 }
920 #endif
922 int Assembler::prefix_and_encode(int reg_enc, bool byteinst) {
923 if (reg_enc >= 8) {
924 prefix(REX_B);
925 reg_enc -= 8;
926 } else if (byteinst && reg_enc >= 4) {
927 prefix(REX);
928 }
929 return reg_enc;
930 }
932 int Assembler::prefixq_and_encode(int reg_enc) {
933 if (reg_enc < 8) {
934 prefix(REX_W);
935 } else {
936 prefix(REX_WB);
937 reg_enc -= 8;
938 }
939 return reg_enc;
940 }
942 int Assembler::prefix_and_encode(int dst_enc, int src_enc, bool byteinst) {
943 if (dst_enc < 8) {
944 if (src_enc >= 8) {
945 prefix(REX_B);
946 src_enc -= 8;
947 } else if (byteinst && src_enc >= 4) {
948 prefix(REX);
949 }
950 } else {
951 if (src_enc < 8) {
952 prefix(REX_R);
953 } else {
954 prefix(REX_RB);
955 src_enc -= 8;
956 }
957 dst_enc -= 8;
958 }
959 return dst_enc << 3 | src_enc;
960 }
962 int Assembler::prefixq_and_encode(int dst_enc, int src_enc) {
963 if (dst_enc < 8) {
964 if (src_enc < 8) {
965 prefix(REX_W);
966 } else {
967 prefix(REX_WB);
968 src_enc -= 8;
969 }
970 } else {
971 if (src_enc < 8) {
972 prefix(REX_WR);
973 } else {
974 prefix(REX_WRB);
975 src_enc -= 8;
976 }
977 dst_enc -= 8;
978 }
979 return dst_enc << 3 | src_enc;
980 }
982 void Assembler::prefix(Register reg) {
983 if (reg->encoding() >= 8) {
984 prefix(REX_B);
985 }
986 }
988 void Assembler::prefix(Address adr) {
989 if (adr.base_needs_rex()) {
990 if (adr.index_needs_rex()) {
991 prefix(REX_XB);
992 } else {
993 prefix(REX_B);
994 }
995 } else {
996 if (adr.index_needs_rex()) {
997 prefix(REX_X);
998 }
999 }
1000 }
1002 void Assembler::prefixq(Address adr) {
1003 if (adr.base_needs_rex()) {
1004 if (adr.index_needs_rex()) {
1005 prefix(REX_WXB);
1006 } else {
1007 prefix(REX_WB);
1008 }
1009 } else {
1010 if (adr.index_needs_rex()) {
1011 prefix(REX_WX);
1012 } else {
1013 prefix(REX_W);
1014 }
1015 }
1016 }
1019 void Assembler::prefix(Address adr, Register reg, bool byteinst) {
1020 if (reg->encoding() < 8) {
1021 if (adr.base_needs_rex()) {
1022 if (adr.index_needs_rex()) {
1023 prefix(REX_XB);
1024 } else {
1025 prefix(REX_B);
1026 }
1027 } else {
1028 if (adr.index_needs_rex()) {
1029 prefix(REX_X);
1030 } else if (reg->encoding() >= 4 ) {
1031 prefix(REX);
1032 }
1033 }
1034 } else {
1035 if (adr.base_needs_rex()) {
1036 if (adr.index_needs_rex()) {
1037 prefix(REX_RXB);
1038 } else {
1039 prefix(REX_RB);
1040 }
1041 } else {
1042 if (adr.index_needs_rex()) {
1043 prefix(REX_RX);
1044 } else {
1045 prefix(REX_R);
1046 }
1047 }
1048 }
1049 }
1051 void Assembler::prefixq(Address adr, Register src) {
1052 if (src->encoding() < 8) {
1053 if (adr.base_needs_rex()) {
1054 if (adr.index_needs_rex()) {
1055 prefix(REX_WXB);
1056 } else {
1057 prefix(REX_WB);
1058 }
1059 } else {
1060 if (adr.index_needs_rex()) {
1061 prefix(REX_WX);
1062 } else {
1063 prefix(REX_W);
1064 }
1065 }
1066 } else {
1067 if (adr.base_needs_rex()) {
1068 if (adr.index_needs_rex()) {
1069 prefix(REX_WRXB);
1070 } else {
1071 prefix(REX_WRB);
1072 }
1073 } else {
1074 if (adr.index_needs_rex()) {
1075 prefix(REX_WRX);
1076 } else {
1077 prefix(REX_WR);
1078 }
1079 }
1080 }
1081 }
1083 void Assembler::prefix(Address adr, XMMRegister reg) {
1084 if (reg->encoding() < 8) {
1085 if (adr.base_needs_rex()) {
1086 if (adr.index_needs_rex()) {
1087 prefix(REX_XB);
1088 } else {
1089 prefix(REX_B);
1090 }
1091 } else {
1092 if (adr.index_needs_rex()) {
1093 prefix(REX_X);
1094 }
1095 }
1096 } else {
1097 if (adr.base_needs_rex()) {
1098 if (adr.index_needs_rex()) {
1099 prefix(REX_RXB);
1100 } else {
1101 prefix(REX_RB);
1102 }
1103 } else {
1104 if (adr.index_needs_rex()) {
1105 prefix(REX_RX);
1106 } else {
1107 prefix(REX_R);
1108 }
1109 }
1110 }
1111 }
1113 void Assembler::emit_operand(Register reg, Address adr,
1114 int rip_relative_correction) {
1115 emit_operand(reg, adr._base, adr._index, adr._scale, adr._disp,
1116 adr._rspec,
1117 rip_relative_correction);
1118 }
1120 void Assembler::emit_operand(XMMRegister reg, Address adr,
1121 int rip_relative_correction) {
1122 emit_operand(reg, adr._base, adr._index, adr._scale, adr._disp,
1123 adr._rspec,
1124 rip_relative_correction);
1125 }
1127 void Assembler::emit_farith(int b1, int b2, int i) {
1128 assert(isByte(b1) && isByte(b2), "wrong opcode");
1129 assert(0 <= i && i < 8, "illegal stack offset");
1130 emit_byte(b1);
1131 emit_byte(b2 + i);
1132 }
1134 // pushad is invalid, use this instead.
1135 // NOTE: Kills flags!!
1136 void Assembler::pushaq() {
1137 // we have to store original rsp. ABI says that 128 bytes
1138 // below rsp are local scratch.
1139 movq(Address(rsp, -5 * wordSize), rsp);
1141 subq(rsp, 16 * wordSize);
1143 movq(Address(rsp, 15 * wordSize), rax);
1144 movq(Address(rsp, 14 * wordSize), rcx);
1145 movq(Address(rsp, 13 * wordSize), rdx);
1146 movq(Address(rsp, 12 * wordSize), rbx);
1147 // skip rsp
1148 movq(Address(rsp, 10 * wordSize), rbp);
1149 movq(Address(rsp, 9 * wordSize), rsi);
1150 movq(Address(rsp, 8 * wordSize), rdi);
1151 movq(Address(rsp, 7 * wordSize), r8);
1152 movq(Address(rsp, 6 * wordSize), r9);
1153 movq(Address(rsp, 5 * wordSize), r10);
1154 movq(Address(rsp, 4 * wordSize), r11);
1155 movq(Address(rsp, 3 * wordSize), r12);
1156 movq(Address(rsp, 2 * wordSize), r13);
1157 movq(Address(rsp, wordSize), r14);
1158 movq(Address(rsp, 0), r15);
1159 }
1161 // popad is invalid, use this instead
1162 // NOTE: Kills flags!!
1163 void Assembler::popaq() {
1164 movq(r15, Address(rsp, 0));
1165 movq(r14, Address(rsp, wordSize));
1166 movq(r13, Address(rsp, 2 * wordSize));
1167 movq(r12, Address(rsp, 3 * wordSize));
1168 movq(r11, Address(rsp, 4 * wordSize));
1169 movq(r10, Address(rsp, 5 * wordSize));
1170 movq(r9, Address(rsp, 6 * wordSize));
1171 movq(r8, Address(rsp, 7 * wordSize));
1172 movq(rdi, Address(rsp, 8 * wordSize));
1173 movq(rsi, Address(rsp, 9 * wordSize));
1174 movq(rbp, Address(rsp, 10 * wordSize));
1175 // skip rsp
1176 movq(rbx, Address(rsp, 12 * wordSize));
1177 movq(rdx, Address(rsp, 13 * wordSize));
1178 movq(rcx, Address(rsp, 14 * wordSize));
1179 movq(rax, Address(rsp, 15 * wordSize));
1181 addq(rsp, 16 * wordSize);
1182 }
1184 void Assembler::pushfq() {
1185 emit_byte(0x9C);
1186 }
1188 void Assembler::popfq() {
1189 emit_byte(0x9D);
1190 }
1192 void Assembler::pushq(int imm32) {
1193 emit_byte(0x68);
1194 emit_long(imm32);
1195 }
1197 void Assembler::pushq(Register src) {
1198 int encode = prefix_and_encode(src->encoding());
1200 emit_byte(0x50 | encode);
1201 }
1203 void Assembler::pushq(Address src) {
1204 InstructionMark im(this);
1205 prefix(src);
1206 emit_byte(0xFF);
1207 emit_operand(rsi, src);
1208 }
1210 void Assembler::popq(Register dst) {
1211 int encode = prefix_and_encode(dst->encoding());
1212 emit_byte(0x58 | encode);
1213 }
1215 void Assembler::popq(Address dst) {
1216 InstructionMark im(this);
1217 prefix(dst);
1218 emit_byte(0x8F);
1219 emit_operand(rax, dst);
1220 }
1222 void Assembler::prefix(Prefix p) {
1223 a_byte(p);
1224 }
1226 void Assembler::movb(Register dst, Address src) {
1227 InstructionMark im(this);
1228 prefix(src, dst, true);
1229 emit_byte(0x8A);
1230 emit_operand(dst, src);
1231 }
1233 void Assembler::movb(Address dst, int imm8) {
1234 InstructionMark im(this);
1235 prefix(dst);
1236 emit_byte(0xC6);
1237 emit_operand(rax, dst, 1);
1238 emit_byte(imm8);
1239 }
1241 void Assembler::movb(Address dst, Register src) {
1242 InstructionMark im(this);
1243 prefix(dst, src, true);
1244 emit_byte(0x88);
1245 emit_operand(src, dst);
1246 }
1248 void Assembler::movw(Address dst, int imm16) {
1249 InstructionMark im(this);
1250 emit_byte(0x66); // switch to 16-bit mode
1251 prefix(dst);
1252 emit_byte(0xC7);
1253 emit_operand(rax, dst, 2);
1254 emit_word(imm16);
1255 }
1257 void Assembler::movw(Register dst, Address src) {
1258 InstructionMark im(this);
1259 emit_byte(0x66);
1260 prefix(src, dst);
1261 emit_byte(0x8B);
1262 emit_operand(dst, src);
1263 }
1265 void Assembler::movw(Address dst, Register src) {
1266 InstructionMark im(this);
1267 emit_byte(0x66);
1268 prefix(dst, src);
1269 emit_byte(0x89);
1270 emit_operand(src, dst);
1271 }
1273 // Uses zero extension.
1274 void Assembler::movl(Register dst, int imm32) {
1275 int encode = prefix_and_encode(dst->encoding());
1276 emit_byte(0xB8 | encode);
1277 emit_long(imm32);
1278 }
1280 void Assembler::movl(Register dst, Register src) {
1281 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1282 emit_byte(0x8B);
1283 emit_byte(0xC0 | encode);
1284 }
1286 void Assembler::movl(Register dst, Address src) {
1287 InstructionMark im(this);
1288 prefix(src, dst);
1289 emit_byte(0x8B);
1290 emit_operand(dst, src);
1291 }
1293 void Assembler::movl(Address dst, int imm32) {
1294 InstructionMark im(this);
1295 prefix(dst);
1296 emit_byte(0xC7);
1297 emit_operand(rax, dst, 4);
1298 emit_long(imm32);
1299 }
1301 void Assembler::movl(Address dst, Register src) {
1302 InstructionMark im(this);
1303 prefix(dst, src);
1304 emit_byte(0x89);
1305 emit_operand(src, dst);
1306 }
1308 void Assembler::mov64(Register dst, intptr_t imm64) {
1309 InstructionMark im(this);
1310 int encode = prefixq_and_encode(dst->encoding());
1311 emit_byte(0xB8 | encode);
1312 emit_long64(imm64);
1313 }
1315 void Assembler::mov_literal64(Register dst, intptr_t imm64, RelocationHolder const& rspec) {
1316 InstructionMark im(this);
1317 int encode = prefixq_and_encode(dst->encoding());
1318 emit_byte(0xB8 | encode);
1319 emit_data64(imm64, rspec);
1320 }
1322 void Assembler::movq(Register dst, Register src) {
1323 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
1324 emit_byte(0x8B);
1325 emit_byte(0xC0 | encode);
1326 }
1328 void Assembler::movq(Register dst, Address src) {
1329 InstructionMark im(this);
1330 prefixq(src, dst);
1331 emit_byte(0x8B);
1332 emit_operand(dst, src);
1333 }
1335 void Assembler::mov64(Address dst, intptr_t imm32) {
1336 assert(is_simm32(imm32), "lost bits");
1337 InstructionMark im(this);
1338 prefixq(dst);
1339 emit_byte(0xC7);
1340 emit_operand(rax, dst, 4);
1341 emit_long(imm32);
1342 }
1344 void Assembler::movq(Address dst, Register src) {
1345 InstructionMark im(this);
1346 prefixq(dst, src);
1347 emit_byte(0x89);
1348 emit_operand(src, dst);
1349 }
1351 void Assembler::movsbl(Register dst, Address src) {
1352 InstructionMark im(this);
1353 prefix(src, dst);
1354 emit_byte(0x0F);
1355 emit_byte(0xBE);
1356 emit_operand(dst, src);
1357 }
1359 void Assembler::movsbl(Register dst, Register src) {
1360 int encode = prefix_and_encode(dst->encoding(), src->encoding(), true);
1361 emit_byte(0x0F);
1362 emit_byte(0xBE);
1363 emit_byte(0xC0 | encode);
1364 }
1366 void Assembler::movswl(Register dst, Address src) {
1367 InstructionMark im(this);
1368 prefix(src, dst);
1369 emit_byte(0x0F);
1370 emit_byte(0xBF);
1371 emit_operand(dst, src);
1372 }
1374 void Assembler::movswl(Register dst, Register src) {
1375 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1376 emit_byte(0x0F);
1377 emit_byte(0xBF);
1378 emit_byte(0xC0 | encode);
1379 }
1381 void Assembler::movslq(Register dst, Address src) {
1382 InstructionMark im(this);
1383 prefixq(src, dst);
1384 emit_byte(0x63);
1385 emit_operand(dst, src);
1386 }
1388 void Assembler::movslq(Register dst, Register src) {
1389 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
1390 emit_byte(0x63);
1391 emit_byte(0xC0 | encode);
1392 }
1394 void Assembler::movzbl(Register dst, Address src) {
1395 InstructionMark im(this);
1396 prefix(src, dst);
1397 emit_byte(0x0F);
1398 emit_byte(0xB6);
1399 emit_operand(dst, src);
1400 }
1402 void Assembler::movzbl(Register dst, Register src) {
1403 int encode = prefix_and_encode(dst->encoding(), src->encoding(), true);
1404 emit_byte(0x0F);
1405 emit_byte(0xB6);
1406 emit_byte(0xC0 | encode);
1407 }
1409 void Assembler::movzwl(Register dst, Address src) {
1410 InstructionMark im(this);
1411 prefix(src, dst);
1412 emit_byte(0x0F);
1413 emit_byte(0xB7);
1414 emit_operand(dst, src);
1415 }
1417 void Assembler::movzwl(Register dst, Register src) {
1418 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1419 emit_byte(0x0F);
1420 emit_byte(0xB7);
1421 emit_byte(0xC0 | encode);
1422 }
1424 void Assembler::movss(XMMRegister dst, XMMRegister src) {
1425 emit_byte(0xF3);
1426 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1427 emit_byte(0x0F);
1428 emit_byte(0x10);
1429 emit_byte(0xC0 | encode);
1430 }
1432 void Assembler::movss(XMMRegister dst, Address src) {
1433 InstructionMark im(this);
1434 emit_byte(0xF3);
1435 prefix(src, dst);
1436 emit_byte(0x0F);
1437 emit_byte(0x10);
1438 emit_operand(dst, src);
1439 }
1441 void Assembler::movss(Address dst, XMMRegister src) {
1442 InstructionMark im(this);
1443 emit_byte(0xF3);
1444 prefix(dst, src);
1445 emit_byte(0x0F);
1446 emit_byte(0x11);
1447 emit_operand(src, dst);
1448 }
1450 void Assembler::movsd(XMMRegister dst, XMMRegister src) {
1451 emit_byte(0xF2);
1452 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1453 emit_byte(0x0F);
1454 emit_byte(0x10);
1455 emit_byte(0xC0 | encode);
1456 }
1458 void Assembler::movsd(XMMRegister dst, Address src) {
1459 InstructionMark im(this);
1460 emit_byte(0xF2);
1461 prefix(src, dst);
1462 emit_byte(0x0F);
1463 emit_byte(0x10);
1464 emit_operand(dst, src);
1465 }
1467 void Assembler::movsd(Address dst, XMMRegister src) {
1468 InstructionMark im(this);
1469 emit_byte(0xF2);
1470 prefix(dst, src);
1471 emit_byte(0x0F);
1472 emit_byte(0x11);
1473 emit_operand(src, dst);
1474 }
1476 // New cpus require to use movsd and movss to avoid partial register stall
1477 // when loading from memory. But for old Opteron use movlpd instead of movsd.
1478 // The selection is done in MacroAssembler::movdbl() and movflt().
1479 void Assembler::movlpd(XMMRegister dst, Address src) {
1480 InstructionMark im(this);
1481 emit_byte(0x66);
1482 prefix(src, dst);
1483 emit_byte(0x0F);
1484 emit_byte(0x12);
1485 emit_operand(dst, src);
1486 }
1488 void Assembler::movapd(XMMRegister dst, XMMRegister src) {
1489 int dstenc = dst->encoding();
1490 int srcenc = src->encoding();
1491 emit_byte(0x66);
1492 if (dstenc < 8) {
1493 if (srcenc >= 8) {
1494 prefix(REX_B);
1495 srcenc -= 8;
1496 }
1497 } else {
1498 if (srcenc < 8) {
1499 prefix(REX_R);
1500 } else {
1501 prefix(REX_RB);
1502 srcenc -= 8;
1503 }
1504 dstenc -= 8;
1505 }
1506 emit_byte(0x0F);
1507 emit_byte(0x28);
1508 emit_byte(0xC0 | dstenc << 3 | srcenc);
1509 }
1511 void Assembler::movaps(XMMRegister dst, XMMRegister src) {
1512 int dstenc = dst->encoding();
1513 int srcenc = src->encoding();
1514 if (dstenc < 8) {
1515 if (srcenc >= 8) {
1516 prefix(REX_B);
1517 srcenc -= 8;
1518 }
1519 } else {
1520 if (srcenc < 8) {
1521 prefix(REX_R);
1522 } else {
1523 prefix(REX_RB);
1524 srcenc -= 8;
1525 }
1526 dstenc -= 8;
1527 }
1528 emit_byte(0x0F);
1529 emit_byte(0x28);
1530 emit_byte(0xC0 | dstenc << 3 | srcenc);
1531 }
1533 void Assembler::movdl(XMMRegister dst, Register src) {
1534 emit_byte(0x66);
1535 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1536 emit_byte(0x0F);
1537 emit_byte(0x6E);
1538 emit_byte(0xC0 | encode);
1539 }
1541 void Assembler::movdl(Register dst, XMMRegister src) {
1542 emit_byte(0x66);
1543 // swap src/dst to get correct prefix
1544 int encode = prefix_and_encode(src->encoding(), dst->encoding());
1545 emit_byte(0x0F);
1546 emit_byte(0x7E);
1547 emit_byte(0xC0 | encode);
1548 }
1550 void Assembler::movdq(XMMRegister dst, Register src) {
1551 emit_byte(0x66);
1552 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
1553 emit_byte(0x0F);
1554 emit_byte(0x6E);
1555 emit_byte(0xC0 | encode);
1556 }
1558 void Assembler::movdq(Register dst, XMMRegister src) {
1559 emit_byte(0x66);
1560 // swap src/dst to get correct prefix
1561 int encode = prefixq_and_encode(src->encoding(), dst->encoding());
1562 emit_byte(0x0F);
1563 emit_byte(0x7E);
1564 emit_byte(0xC0 | encode);
1565 }
1567 void Assembler::pxor(XMMRegister dst, Address src) {
1568 InstructionMark im(this);
1569 emit_byte(0x66);
1570 prefix(src, dst);
1571 emit_byte(0x0F);
1572 emit_byte(0xEF);
1573 emit_operand(dst, src);
1574 }
1576 void Assembler::pxor(XMMRegister dst, XMMRegister src) {
1577 InstructionMark im(this);
1578 emit_byte(0x66);
1579 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1580 emit_byte(0x0F);
1581 emit_byte(0xEF);
1582 emit_byte(0xC0 | encode);
1583 }
1585 void Assembler::movdqa(XMMRegister dst, Address src) {
1586 InstructionMark im(this);
1587 emit_byte(0x66);
1588 prefix(src, dst);
1589 emit_byte(0x0F);
1590 emit_byte(0x6F);
1591 emit_operand(dst, src);
1592 }
1594 void Assembler::movdqa(XMMRegister dst, XMMRegister src) {
1595 emit_byte(0x66);
1596 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
1597 emit_byte(0x0F);
1598 emit_byte(0x6F);
1599 emit_byte(0xC0 | encode);
1600 }
1602 void Assembler::movdqa(Address dst, XMMRegister src) {
1603 InstructionMark im(this);
1604 emit_byte(0x66);
1605 prefix(dst, src);
1606 emit_byte(0x0F);
1607 emit_byte(0x7F);
1608 emit_operand(src, dst);
1609 }
1611 void Assembler::movq(XMMRegister dst, Address src) {
1612 InstructionMark im(this);
1613 emit_byte(0xF3);
1614 prefix(src, dst);
1615 emit_byte(0x0F);
1616 emit_byte(0x7E);
1617 emit_operand(dst, src);
1618 }
1620 void Assembler::movq(Address dst, XMMRegister src) {
1621 InstructionMark im(this);
1622 emit_byte(0x66);
1623 prefix(dst, src);
1624 emit_byte(0x0F);
1625 emit_byte(0xD6);
1626 emit_operand(src, dst);
1627 }
1629 void Assembler::pshufd(XMMRegister dst, XMMRegister src, int mode) {
1630 assert(isByte(mode), "invalid value");
1631 emit_byte(0x66);
1632 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1633 emit_byte(0x0F);
1634 emit_byte(0x70);
1635 emit_byte(0xC0 | encode);
1636 emit_byte(mode & 0xFF);
1637 }
1639 void Assembler::pshufd(XMMRegister dst, Address src, int mode) {
1640 assert(isByte(mode), "invalid value");
1641 InstructionMark im(this);
1642 emit_byte(0x66);
1643 emit_byte(0x0F);
1644 emit_byte(0x70);
1645 emit_operand(dst, src);
1646 emit_byte(mode & 0xFF);
1647 }
1649 void Assembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) {
1650 assert(isByte(mode), "invalid value");
1651 emit_byte(0xF2);
1652 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1653 emit_byte(0x0F);
1654 emit_byte(0x70);
1655 emit_byte(0xC0 | encode);
1656 emit_byte(mode & 0xFF);
1657 }
1659 void Assembler::pshuflw(XMMRegister dst, Address src, int mode) {
1660 assert(isByte(mode), "invalid value");
1661 InstructionMark im(this);
1662 emit_byte(0xF2);
1663 emit_byte(0x0F);
1664 emit_byte(0x70);
1665 emit_operand(dst, src);
1666 emit_byte(mode & 0xFF);
1667 }
1669 void Assembler::cmovl(Condition cc, Register dst, Register src) {
1670 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1671 emit_byte(0x0F);
1672 emit_byte(0x40 | cc);
1673 emit_byte(0xC0 | encode);
1674 }
1676 void Assembler::cmovl(Condition cc, Register dst, Address src) {
1677 InstructionMark im(this);
1678 prefix(src, dst);
1679 emit_byte(0x0F);
1680 emit_byte(0x40 | cc);
1681 emit_operand(dst, src);
1682 }
1684 void Assembler::cmovq(Condition cc, Register dst, Register src) {
1685 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
1686 emit_byte(0x0F);
1687 emit_byte(0x40 | cc);
1688 emit_byte(0xC0 | encode);
1689 }
1691 void Assembler::cmovq(Condition cc, Register dst, Address src) {
1692 InstructionMark im(this);
1693 prefixq(src, dst);
1694 emit_byte(0x0F);
1695 emit_byte(0x40 | cc);
1696 emit_operand(dst, src);
1697 }
1699 void Assembler::prefetch_prefix(Address src) {
1700 prefix(src);
1701 emit_byte(0x0F);
1702 }
1704 void Assembler::prefetcht0(Address src) {
1705 InstructionMark im(this);
1706 prefetch_prefix(src);
1707 emit_byte(0x18);
1708 emit_operand(rcx, src); // 1, src
1709 }
1711 void Assembler::prefetcht1(Address src) {
1712 InstructionMark im(this);
1713 prefetch_prefix(src);
1714 emit_byte(0x18);
1715 emit_operand(rdx, src); // 2, src
1716 }
1718 void Assembler::prefetcht2(Address src) {
1719 InstructionMark im(this);
1720 prefetch_prefix(src);
1721 emit_byte(0x18);
1722 emit_operand(rbx, src); // 3, src
1723 }
1725 void Assembler::prefetchnta(Address src) {
1726 InstructionMark im(this);
1727 prefetch_prefix(src);
1728 emit_byte(0x18);
1729 emit_operand(rax, src); // 0, src
1730 }
1732 void Assembler::prefetchw(Address src) {
1733 InstructionMark im(this);
1734 prefetch_prefix(src);
1735 emit_byte(0x0D);
1736 emit_operand(rcx, src); // 1, src
1737 }
1739 void Assembler::adcl(Register dst, int imm32) {
1740 prefix(dst);
1741 emit_arith(0x81, 0xD0, dst, imm32);
1742 }
1744 void Assembler::adcl(Register dst, Address src) {
1745 InstructionMark im(this);
1746 prefix(src, dst);
1747 emit_byte(0x13);
1748 emit_operand(dst, src);
1749 }
1751 void Assembler::adcl(Register dst, Register src) {
1752 (void) prefix_and_encode(dst->encoding(), src->encoding());
1753 emit_arith(0x13, 0xC0, dst, src);
1754 }
1756 void Assembler::adcq(Register dst, int imm32) {
1757 (void) prefixq_and_encode(dst->encoding());
1758 emit_arith(0x81, 0xD0, dst, imm32);
1759 }
1761 void Assembler::adcq(Register dst, Address src) {
1762 InstructionMark im(this);
1763 prefixq(src, dst);
1764 emit_byte(0x13);
1765 emit_operand(dst, src);
1766 }
1768 void Assembler::adcq(Register dst, Register src) {
1769 (int) prefixq_and_encode(dst->encoding(), src->encoding());
1770 emit_arith(0x13, 0xC0, dst, src);
1771 }
1773 void Assembler::addl(Address dst, int imm32) {
1774 InstructionMark im(this);
1775 prefix(dst);
1776 emit_arith_operand(0x81, rax, dst,imm32);
1777 }
1779 void Assembler::addl(Address dst, Register src) {
1780 InstructionMark im(this);
1781 prefix(dst, src);
1782 emit_byte(0x01);
1783 emit_operand(src, dst);
1784 }
1786 void Assembler::addl(Register dst, int imm32) {
1787 prefix(dst);
1788 emit_arith(0x81, 0xC0, dst, imm32);
1789 }
1791 void Assembler::addl(Register dst, Address src) {
1792 InstructionMark im(this);
1793 prefix(src, dst);
1794 emit_byte(0x03);
1795 emit_operand(dst, src);
1796 }
1798 void Assembler::addl(Register dst, Register src) {
1799 (void) prefix_and_encode(dst->encoding(), src->encoding());
1800 emit_arith(0x03, 0xC0, dst, src);
1801 }
1803 void Assembler::addq(Address dst, int imm32) {
1804 InstructionMark im(this);
1805 prefixq(dst);
1806 emit_arith_operand(0x81, rax, dst,imm32);
1807 }
1809 void Assembler::addq(Address dst, Register src) {
1810 InstructionMark im(this);
1811 prefixq(dst, src);
1812 emit_byte(0x01);
1813 emit_operand(src, dst);
1814 }
1816 void Assembler::addq(Register dst, int imm32) {
1817 (void) prefixq_and_encode(dst->encoding());
1818 emit_arith(0x81, 0xC0, dst, imm32);
1819 }
1821 void Assembler::addq(Register dst, Address src) {
1822 InstructionMark im(this);
1823 prefixq(src, dst);
1824 emit_byte(0x03);
1825 emit_operand(dst, src);
1826 }
1828 void Assembler::addq(Register dst, Register src) {
1829 (void) prefixq_and_encode(dst->encoding(), src->encoding());
1830 emit_arith(0x03, 0xC0, dst, src);
1831 }
1833 void Assembler::andl(Register dst, int imm32) {
1834 prefix(dst);
1835 emit_arith(0x81, 0xE0, dst, imm32);
1836 }
1838 void Assembler::andl(Register dst, Address src) {
1839 InstructionMark im(this);
1840 prefix(src, dst);
1841 emit_byte(0x23);
1842 emit_operand(dst, src);
1843 }
1845 void Assembler::andl(Register dst, Register src) {
1846 (void) prefix_and_encode(dst->encoding(), src->encoding());
1847 emit_arith(0x23, 0xC0, dst, src);
1848 }
1850 void Assembler::andq(Register dst, int imm32) {
1851 (void) prefixq_and_encode(dst->encoding());
1852 emit_arith(0x81, 0xE0, dst, imm32);
1853 }
1855 void Assembler::andq(Register dst, Address src) {
1856 InstructionMark im(this);
1857 prefixq(src, dst);
1858 emit_byte(0x23);
1859 emit_operand(dst, src);
1860 }
1862 void Assembler::andq(Register dst, Register src) {
1863 (int) prefixq_and_encode(dst->encoding(), src->encoding());
1864 emit_arith(0x23, 0xC0, dst, src);
1865 }
1867 void Assembler::cmpb(Address dst, int imm8) {
1868 InstructionMark im(this);
1869 prefix(dst);
1870 emit_byte(0x80);
1871 emit_operand(rdi, dst, 1);
1872 emit_byte(imm8);
1873 }
1875 void Assembler::cmpl(Address dst, int imm32) {
1876 InstructionMark im(this);
1877 prefix(dst);
1878 emit_byte(0x81);
1879 emit_operand(rdi, dst, 4);
1880 emit_long(imm32);
1881 }
1883 void Assembler::cmpl(Register dst, int imm32) {
1884 prefix(dst);
1885 emit_arith(0x81, 0xF8, dst, imm32);
1886 }
1888 void Assembler::cmpl(Register dst, Register src) {
1889 (void) prefix_and_encode(dst->encoding(), src->encoding());
1890 emit_arith(0x3B, 0xC0, dst, src);
1891 }
1893 void Assembler::cmpl(Register dst, Address src) {
1894 InstructionMark im(this);
1895 prefix(src, dst);
1896 emit_byte(0x3B);
1897 emit_operand(dst, src);
1898 }
1900 void Assembler::cmpq(Address dst, int imm32) {
1901 InstructionMark im(this);
1902 prefixq(dst);
1903 emit_byte(0x81);
1904 emit_operand(rdi, dst, 4);
1905 emit_long(imm32);
1906 }
1908 void Assembler::cmpq(Register dst, int imm32) {
1909 (void) prefixq_and_encode(dst->encoding());
1910 emit_arith(0x81, 0xF8, dst, imm32);
1911 }
1913 void Assembler::cmpq(Address dst, Register src) {
1914 prefixq(dst, src);
1915 emit_byte(0x3B);
1916 emit_operand(src, dst);
1917 }
1919 void Assembler::cmpq(Register dst, Register src) {
1920 (void) prefixq_and_encode(dst->encoding(), src->encoding());
1921 emit_arith(0x3B, 0xC0, dst, src);
1922 }
1924 void Assembler::cmpq(Register dst, Address src) {
1925 InstructionMark im(this);
1926 prefixq(src, dst);
1927 emit_byte(0x3B);
1928 emit_operand(dst, src);
1929 }
1931 void Assembler::ucomiss(XMMRegister dst, XMMRegister src) {
1932 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1933 emit_byte(0x0F);
1934 emit_byte(0x2E);
1935 emit_byte(0xC0 | encode);
1936 }
1938 void Assembler::ucomisd(XMMRegister dst, XMMRegister src) {
1939 emit_byte(0x66);
1940 ucomiss(dst, src);
1941 }
1943 void Assembler::decl(Register dst) {
1944 // Don't use it directly. Use MacroAssembler::decrementl() instead.
1945 // Use two-byte form (one-byte from is a REX prefix in 64-bit mode)
1946 int encode = prefix_and_encode(dst->encoding());
1947 emit_byte(0xFF);
1948 emit_byte(0xC8 | encode);
1949 }
1951 void Assembler::decl(Address dst) {
1952 // Don't use it directly. Use MacroAssembler::decrementl() instead.
1953 InstructionMark im(this);
1954 prefix(dst);
1955 emit_byte(0xFF);
1956 emit_operand(rcx, dst);
1957 }
1959 void Assembler::decq(Register dst) {
1960 // Don't use it directly. Use MacroAssembler::decrementq() instead.
1961 // Use two-byte form (one-byte from is a REX prefix in 64-bit mode)
1962 int encode = prefixq_and_encode(dst->encoding());
1963 emit_byte(0xFF);
1964 emit_byte(0xC8 | encode);
1965 }
1967 void Assembler::decq(Address dst) {
1968 // Don't use it directly. Use MacroAssembler::decrementq() instead.
1969 InstructionMark im(this);
1970 prefixq(dst);
1971 emit_byte(0xFF);
1972 emit_operand(rcx, dst);
1973 }
1975 void Assembler::idivl(Register src) {
1976 int encode = prefix_and_encode(src->encoding());
1977 emit_byte(0xF7);
1978 emit_byte(0xF8 | encode);
1979 }
1981 void Assembler::idivq(Register src) {
1982 int encode = prefixq_and_encode(src->encoding());
1983 emit_byte(0xF7);
1984 emit_byte(0xF8 | encode);
1985 }
1987 void Assembler::cdql() {
1988 emit_byte(0x99);
1989 }
1991 void Assembler::cdqq() {
1992 prefix(REX_W);
1993 emit_byte(0x99);
1994 }
1996 void Assembler::imull(Register dst, Register src) {
1997 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1998 emit_byte(0x0F);
1999 emit_byte(0xAF);
2000 emit_byte(0xC0 | encode);
2001 }
2003 void Assembler::imull(Register dst, Register src, int value) {
2004 int encode = prefix_and_encode(dst->encoding(), src->encoding());
2005 if (is8bit(value)) {
2006 emit_byte(0x6B);
2007 emit_byte(0xC0 | encode);
2008 emit_byte(value);
2009 } else {
2010 emit_byte(0x69);
2011 emit_byte(0xC0 | encode);
2012 emit_long(value);
2013 }
2014 }
2016 void Assembler::imulq(Register dst, Register src) {
2017 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
2018 emit_byte(0x0F);
2019 emit_byte(0xAF);
2020 emit_byte(0xC0 | encode);
2021 }
2023 void Assembler::imulq(Register dst, Register src, int value) {
2024 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
2025 if (is8bit(value)) {
2026 emit_byte(0x6B);
2027 emit_byte(0xC0 | encode);
2028 emit_byte(value);
2029 } else {
2030 emit_byte(0x69);
2031 emit_byte(0xC0 | encode);
2032 emit_long(value);
2033 }
2034 }
2036 void Assembler::incl(Register dst) {
2037 // Don't use it directly. Use MacroAssembler::incrementl() instead.
2038 // Use two-byte form (one-byte from is a REX prefix in 64-bit mode)
2039 int encode = prefix_and_encode(dst->encoding());
2040 emit_byte(0xFF);
2041 emit_byte(0xC0 | encode);
2042 }
2044 void Assembler::incl(Address dst) {
2045 // Don't use it directly. Use MacroAssembler::incrementl() instead.
2046 InstructionMark im(this);
2047 prefix(dst);
2048 emit_byte(0xFF);
2049 emit_operand(rax, dst);
2050 }
2052 void Assembler::incq(Register dst) {
2053 // Don't use it directly. Use MacroAssembler::incrementq() instead.
2054 // Use two-byte form (one-byte from is a REX prefix in 64-bit mode)
2055 int encode = prefixq_and_encode(dst->encoding());
2056 emit_byte(0xFF);
2057 emit_byte(0xC0 | encode);
2058 }
2060 void Assembler::incq(Address dst) {
2061 // Don't use it directly. Use MacroAssembler::incrementq() instead.
2062 InstructionMark im(this);
2063 prefixq(dst);
2064 emit_byte(0xFF);
2065 emit_operand(rax, dst);
2066 }
2068 void Assembler::leal(Register dst, Address src) {
2069 InstructionMark im(this);
2070 emit_byte(0x67); // addr32
2071 prefix(src, dst);
2072 emit_byte(0x8D);
2073 emit_operand(dst, src);
2074 }
2076 void Assembler::leaq(Register dst, Address src) {
2077 InstructionMark im(this);
2078 prefixq(src, dst);
2079 emit_byte(0x8D);
2080 emit_operand(dst, src);
2081 }
2083 void Assembler::mull(Address src) {
2084 InstructionMark im(this);
2085 // was missing
2086 prefix(src);
2087 emit_byte(0xF7);
2088 emit_operand(rsp, src);
2089 }
2091 void Assembler::mull(Register src) {
2092 // was missing
2093 int encode = prefix_and_encode(src->encoding());
2094 emit_byte(0xF7);
2095 emit_byte(0xE0 | encode);
2096 }
2098 void Assembler::negl(Register dst) {
2099 int encode = prefix_and_encode(dst->encoding());
2100 emit_byte(0xF7);
2101 emit_byte(0xD8 | encode);
2102 }
2104 void Assembler::negq(Register dst) {
2105 int encode = prefixq_and_encode(dst->encoding());
2106 emit_byte(0xF7);
2107 emit_byte(0xD8 | encode);
2108 }
2110 void Assembler::notl(Register dst) {
2111 int encode = prefix_and_encode(dst->encoding());
2112 emit_byte(0xF7);
2113 emit_byte(0xD0 | encode);
2114 }
2116 void Assembler::notq(Register dst) {
2117 int encode = prefixq_and_encode(dst->encoding());
2118 emit_byte(0xF7);
2119 emit_byte(0xD0 | encode);
2120 }
2122 void Assembler::orl(Address dst, int imm32) {
2123 InstructionMark im(this);
2124 prefix(dst);
2125 emit_byte(0x81);
2126 emit_operand(rcx, dst, 4);
2127 emit_long(imm32);
2128 }
2130 void Assembler::orl(Register dst, int imm32) {
2131 prefix(dst);
2132 emit_arith(0x81, 0xC8, dst, imm32);
2133 }
2135 void Assembler::orl(Register dst, Address src) {
2136 InstructionMark im(this);
2137 prefix(src, dst);
2138 emit_byte(0x0B);
2139 emit_operand(dst, src);
2140 }
2142 void Assembler::orl(Register dst, Register src) {
2143 (void) prefix_and_encode(dst->encoding(), src->encoding());
2144 emit_arith(0x0B, 0xC0, dst, src);
2145 }
2147 void Assembler::orq(Address dst, int imm32) {
2148 InstructionMark im(this);
2149 prefixq(dst);
2150 emit_byte(0x81);
2151 emit_operand(rcx, dst, 4);
2152 emit_long(imm32);
2153 }
2155 void Assembler::orq(Register dst, int imm32) {
2156 (void) prefixq_and_encode(dst->encoding());
2157 emit_arith(0x81, 0xC8, dst, imm32);
2158 }
2160 void Assembler::orq(Register dst, Address src) {
2161 InstructionMark im(this);
2162 prefixq(src, dst);
2163 emit_byte(0x0B);
2164 emit_operand(dst, src);
2165 }
2167 void Assembler::orq(Register dst, Register src) {
2168 (void) prefixq_and_encode(dst->encoding(), src->encoding());
2169 emit_arith(0x0B, 0xC0, dst, src);
2170 }
2172 void Assembler::rcll(Register dst, int imm8) {
2173 assert(isShiftCount(imm8), "illegal shift count");
2174 int encode = prefix_and_encode(dst->encoding());
2175 if (imm8 == 1) {
2176 emit_byte(0xD1);
2177 emit_byte(0xD0 | encode);
2178 } else {
2179 emit_byte(0xC1);
2180 emit_byte(0xD0 | encode);
2181 emit_byte(imm8);
2182 }
2183 }
2185 void Assembler::rclq(Register dst, int imm8) {
2186 assert(isShiftCount(imm8 >> 1), "illegal shift count");
2187 int encode = prefixq_and_encode(dst->encoding());
2188 if (imm8 == 1) {
2189 emit_byte(0xD1);
2190 emit_byte(0xD0 | encode);
2191 } else {
2192 emit_byte(0xC1);
2193 emit_byte(0xD0 | encode);
2194 emit_byte(imm8);
2195 }
2196 }
2198 void Assembler::sarl(Register dst, int imm8) {
2199 int encode = prefix_and_encode(dst->encoding());
2200 assert(isShiftCount(imm8), "illegal shift count");
2201 if (imm8 == 1) {
2202 emit_byte(0xD1);
2203 emit_byte(0xF8 | encode);
2204 } else {
2205 emit_byte(0xC1);
2206 emit_byte(0xF8 | encode);
2207 emit_byte(imm8);
2208 }
2209 }
2211 void Assembler::sarl(Register dst) {
2212 int encode = prefix_and_encode(dst->encoding());
2213 emit_byte(0xD3);
2214 emit_byte(0xF8 | encode);
2215 }
2217 void Assembler::sarq(Register dst, int imm8) {
2218 assert(isShiftCount(imm8 >> 1), "illegal shift count");
2219 int encode = prefixq_and_encode(dst->encoding());
2220 if (imm8 == 1) {
2221 emit_byte(0xD1);
2222 emit_byte(0xF8 | encode);
2223 } else {
2224 emit_byte(0xC1);
2225 emit_byte(0xF8 | encode);
2226 emit_byte(imm8);
2227 }
2228 }
2230 void Assembler::sarq(Register dst) {
2231 int encode = prefixq_and_encode(dst->encoding());
2232 emit_byte(0xD3);
2233 emit_byte(0xF8 | encode);
2234 }
2236 void Assembler::sbbl(Address dst, int imm32) {
2237 InstructionMark im(this);
2238 prefix(dst);
2239 emit_arith_operand(0x81, rbx, dst, imm32);
2240 }
2242 void Assembler::sbbl(Register dst, int imm32) {
2243 prefix(dst);
2244 emit_arith(0x81, 0xD8, dst, imm32);
2245 }
2247 void Assembler::sbbl(Register dst, Address src) {
2248 InstructionMark im(this);
2249 prefix(src, dst);
2250 emit_byte(0x1B);
2251 emit_operand(dst, src);
2252 }
2254 void Assembler::sbbl(Register dst, Register src) {
2255 (void) prefix_and_encode(dst->encoding(), src->encoding());
2256 emit_arith(0x1B, 0xC0, dst, src);
2257 }
2259 void Assembler::sbbq(Address dst, int imm32) {
2260 InstructionMark im(this);
2261 prefixq(dst);
2262 emit_arith_operand(0x81, rbx, dst, imm32);
2263 }
2265 void Assembler::sbbq(Register dst, int imm32) {
2266 (void) prefixq_and_encode(dst->encoding());
2267 emit_arith(0x81, 0xD8, dst, imm32);
2268 }
2270 void Assembler::sbbq(Register dst, Address src) {
2271 InstructionMark im(this);
2272 prefixq(src, dst);
2273 emit_byte(0x1B);
2274 emit_operand(dst, src);
2275 }
2277 void Assembler::sbbq(Register dst, Register src) {
2278 (void) prefixq_and_encode(dst->encoding(), src->encoding());
2279 emit_arith(0x1B, 0xC0, dst, src);
2280 }
2282 void Assembler::shll(Register dst, int imm8) {
2283 assert(isShiftCount(imm8), "illegal shift count");
2284 int encode = prefix_and_encode(dst->encoding());
2285 if (imm8 == 1 ) {
2286 emit_byte(0xD1);
2287 emit_byte(0xE0 | encode);
2288 } else {
2289 emit_byte(0xC1);
2290 emit_byte(0xE0 | encode);
2291 emit_byte(imm8);
2292 }
2293 }
2295 void Assembler::shll(Register dst) {
2296 int encode = prefix_and_encode(dst->encoding());
2297 emit_byte(0xD3);
2298 emit_byte(0xE0 | encode);
2299 }
2301 void Assembler::shlq(Register dst, int imm8) {
2302 assert(isShiftCount(imm8 >> 1), "illegal shift count");
2303 int encode = prefixq_and_encode(dst->encoding());
2304 if (imm8 == 1) {
2305 emit_byte(0xD1);
2306 emit_byte(0xE0 | encode);
2307 } else {
2308 emit_byte(0xC1);
2309 emit_byte(0xE0 | encode);
2310 emit_byte(imm8);
2311 }
2312 }
2314 void Assembler::shlq(Register dst) {
2315 int encode = prefixq_and_encode(dst->encoding());
2316 emit_byte(0xD3);
2317 emit_byte(0xE0 | encode);
2318 }
2320 void Assembler::shrl(Register dst, int imm8) {
2321 assert(isShiftCount(imm8), "illegal shift count");
2322 int encode = prefix_and_encode(dst->encoding());
2323 emit_byte(0xC1);
2324 emit_byte(0xE8 | encode);
2325 emit_byte(imm8);
2326 }
2328 void Assembler::shrl(Register dst) {
2329 int encode = prefix_and_encode(dst->encoding());
2330 emit_byte(0xD3);
2331 emit_byte(0xE8 | encode);
2332 }
2334 void Assembler::shrq(Register dst, int imm8) {
2335 assert(isShiftCount(imm8 >> 1), "illegal shift count");
2336 int encode = prefixq_and_encode(dst->encoding());
2337 emit_byte(0xC1);
2338 emit_byte(0xE8 | encode);
2339 emit_byte(imm8);
2340 }
2342 void Assembler::shrq(Register dst) {
2343 int encode = prefixq_and_encode(dst->encoding());
2344 emit_byte(0xD3);
2345 emit_byte(0xE8 | encode);
2346 }
2348 void Assembler::subl(Address dst, int imm32) {
2349 InstructionMark im(this);
2350 prefix(dst);
2351 if (is8bit(imm32)) {
2352 emit_byte(0x83);
2353 emit_operand(rbp, dst, 1);
2354 emit_byte(imm32 & 0xFF);
2355 } else {
2356 emit_byte(0x81);
2357 emit_operand(rbp, dst, 4);
2358 emit_long(imm32);
2359 }
2360 }
2362 void Assembler::subl(Register dst, int imm32) {
2363 prefix(dst);
2364 emit_arith(0x81, 0xE8, dst, imm32);
2365 }
2367 void Assembler::subl(Address dst, Register src) {
2368 InstructionMark im(this);
2369 prefix(dst, src);
2370 emit_byte(0x29);
2371 emit_operand(src, dst);
2372 }
2374 void Assembler::subl(Register dst, Address src) {
2375 InstructionMark im(this);
2376 prefix(src, dst);
2377 emit_byte(0x2B);
2378 emit_operand(dst, src);
2379 }
2381 void Assembler::subl(Register dst, Register src) {
2382 (void) prefix_and_encode(dst->encoding(), src->encoding());
2383 emit_arith(0x2B, 0xC0, dst, src);
2384 }
2386 void Assembler::subq(Address dst, int imm32) {
2387 InstructionMark im(this);
2388 prefixq(dst);
2389 if (is8bit(imm32)) {
2390 emit_byte(0x83);
2391 emit_operand(rbp, dst, 1);
2392 emit_byte(imm32 & 0xFF);
2393 } else {
2394 emit_byte(0x81);
2395 emit_operand(rbp, dst, 4);
2396 emit_long(imm32);
2397 }
2398 }
2400 void Assembler::subq(Register dst, int imm32) {
2401 (void) prefixq_and_encode(dst->encoding());
2402 emit_arith(0x81, 0xE8, dst, imm32);
2403 }
2405 void Assembler::subq(Address dst, Register src) {
2406 InstructionMark im(this);
2407 prefixq(dst, src);
2408 emit_byte(0x29);
2409 emit_operand(src, dst);
2410 }
2412 void Assembler::subq(Register dst, Address src) {
2413 InstructionMark im(this);
2414 prefixq(src, dst);
2415 emit_byte(0x2B);
2416 emit_operand(dst, src);
2417 }
2419 void Assembler::subq(Register dst, Register src) {
2420 (void) prefixq_and_encode(dst->encoding(), src->encoding());
2421 emit_arith(0x2B, 0xC0, dst, src);
2422 }
2424 void Assembler::testb(Register dst, int imm8) {
2425 (void) prefix_and_encode(dst->encoding(), true);
2426 emit_arith_b(0xF6, 0xC0, dst, imm8);
2427 }
2429 void Assembler::testl(Register dst, int imm32) {
2430 // not using emit_arith because test
2431 // doesn't support sign-extension of
2432 // 8bit operands
2433 int encode = dst->encoding();
2434 if (encode == 0) {
2435 emit_byte(0xA9);
2436 } else {
2437 encode = prefix_and_encode(encode);
2438 emit_byte(0xF7);
2439 emit_byte(0xC0 | encode);
2440 }
2441 emit_long(imm32);
2442 }
2444 void Assembler::testl(Register dst, Register src) {
2445 (void) prefix_and_encode(dst->encoding(), src->encoding());
2446 emit_arith(0x85, 0xC0, dst, src);
2447 }
2449 void Assembler::testq(Register dst, int imm32) {
2450 // not using emit_arith because test
2451 // doesn't support sign-extension of
2452 // 8bit operands
2453 int encode = dst->encoding();
2454 if (encode == 0) {
2455 prefix(REX_W);
2456 emit_byte(0xA9);
2457 } else {
2458 encode = prefixq_and_encode(encode);
2459 emit_byte(0xF7);
2460 emit_byte(0xC0 | encode);
2461 }
2462 emit_long(imm32);
2463 }
2465 void Assembler::testq(Register dst, Register src) {
2466 (void) prefixq_and_encode(dst->encoding(), src->encoding());
2467 emit_arith(0x85, 0xC0, dst, src);
2468 }
2470 void Assembler::xaddl(Address dst, Register src) {
2471 InstructionMark im(this);
2472 prefix(dst, src);
2473 emit_byte(0x0F);
2474 emit_byte(0xC1);
2475 emit_operand(src, dst);
2476 }
2478 void Assembler::xaddq(Address dst, Register src) {
2479 InstructionMark im(this);
2480 prefixq(dst, src);
2481 emit_byte(0x0F);
2482 emit_byte(0xC1);
2483 emit_operand(src, dst);
2484 }
2486 void Assembler::xorl(Register dst, int imm32) {
2487 prefix(dst);
2488 emit_arith(0x81, 0xF0, dst, imm32);
2489 }
2491 void Assembler::xorl(Register dst, Register src) {
2492 (void) prefix_and_encode(dst->encoding(), src->encoding());
2493 emit_arith(0x33, 0xC0, dst, src);
2494 }
2496 void Assembler::xorl(Register dst, Address src) {
2497 InstructionMark im(this);
2498 prefix(src, dst);
2499 emit_byte(0x33);
2500 emit_operand(dst, src);
2501 }
2503 void Assembler::xorq(Register dst, int imm32) {
2504 (void) prefixq_and_encode(dst->encoding());
2505 emit_arith(0x81, 0xF0, dst, imm32);
2506 }
2508 void Assembler::xorq(Register dst, Register src) {
2509 (void) prefixq_and_encode(dst->encoding(), src->encoding());
2510 emit_arith(0x33, 0xC0, dst, src);
2511 }
2513 void Assembler::xorq(Register dst, Address src) {
2514 InstructionMark im(this);
2515 prefixq(src, dst);
2516 emit_byte(0x33);
2517 emit_operand(dst, src);
2518 }
2520 void Assembler::bswapl(Register reg) {
2521 int encode = prefix_and_encode(reg->encoding());
2522 emit_byte(0x0F);
2523 emit_byte(0xC8 | encode);
2524 }
2526 void Assembler::bswapq(Register reg) {
2527 int encode = prefixq_and_encode(reg->encoding());
2528 emit_byte(0x0F);
2529 emit_byte(0xC8 | encode);
2530 }
2532 void Assembler::lock() {
2533 emit_byte(0xF0);
2534 }
2536 void Assembler::xchgl(Register dst, Address src) {
2537 InstructionMark im(this);
2538 prefix(src, dst);
2539 emit_byte(0x87);
2540 emit_operand(dst, src);
2541 }
2543 void Assembler::xchgl(Register dst, Register src) {
2544 int encode = prefix_and_encode(dst->encoding(), src->encoding());
2545 emit_byte(0x87);
2546 emit_byte(0xc0 | encode);
2547 }
2549 void Assembler::xchgq(Register dst, Address src) {
2550 InstructionMark im(this);
2551 prefixq(src, dst);
2552 emit_byte(0x87);
2553 emit_operand(dst, src);
2554 }
2556 void Assembler::xchgq(Register dst, Register src) {
2557 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
2558 emit_byte(0x87);
2559 emit_byte(0xc0 | encode);
2560 }
2562 void Assembler::cmpxchgl(Register reg, Address adr) {
2563 InstructionMark im(this);
2564 prefix(adr, reg);
2565 emit_byte(0x0F);
2566 emit_byte(0xB1);
2567 emit_operand(reg, adr);
2568 }
2570 void Assembler::cmpxchgq(Register reg, Address adr) {
2571 InstructionMark im(this);
2572 prefixq(adr, reg);
2573 emit_byte(0x0F);
2574 emit_byte(0xB1);
2575 emit_operand(reg, adr);
2576 }
2578 void Assembler::hlt() {
2579 emit_byte(0xF4);
2580 }
2583 void Assembler::addr_nop_4() {
2584 // 4 bytes: NOP DWORD PTR [EAX+0]
2585 emit_byte(0x0F);
2586 emit_byte(0x1F);
2587 emit_byte(0x40); // emit_rm(cbuf, 0x1, EAX_enc, EAX_enc);
2588 emit_byte(0); // 8-bits offset (1 byte)
2589 }
2591 void Assembler::addr_nop_5() {
2592 // 5 bytes: NOP DWORD PTR [EAX+EAX*0+0] 8-bits offset
2593 emit_byte(0x0F);
2594 emit_byte(0x1F);
2595 emit_byte(0x44); // emit_rm(cbuf, 0x1, EAX_enc, 0x4);
2596 emit_byte(0x00); // emit_rm(cbuf, 0x0, EAX_enc, EAX_enc);
2597 emit_byte(0); // 8-bits offset (1 byte)
2598 }
2600 void Assembler::addr_nop_7() {
2601 // 7 bytes: NOP DWORD PTR [EAX+0] 32-bits offset
2602 emit_byte(0x0F);
2603 emit_byte(0x1F);
2604 emit_byte(0x80); // emit_rm(cbuf, 0x2, EAX_enc, EAX_enc);
2605 emit_long(0); // 32-bits offset (4 bytes)
2606 }
2608 void Assembler::addr_nop_8() {
2609 // 8 bytes: NOP DWORD PTR [EAX+EAX*0+0] 32-bits offset
2610 emit_byte(0x0F);
2611 emit_byte(0x1F);
2612 emit_byte(0x84); // emit_rm(cbuf, 0x2, EAX_enc, 0x4);
2613 emit_byte(0x00); // emit_rm(cbuf, 0x0, EAX_enc, EAX_enc);
2614 emit_long(0); // 32-bits offset (4 bytes)
2615 }
2617 void Assembler::nop(int i) {
2618 assert(i > 0, " ");
2619 if (UseAddressNop && VM_Version::is_intel()) {
2620 //
2621 // Using multi-bytes nops "0x0F 0x1F [address]" for Intel
2622 // 1: 0x90
2623 // 2: 0x66 0x90
2624 // 3: 0x66 0x66 0x90 (don't use "0x0F 0x1F 0x00" - need patching safe padding)
2625 // 4: 0x0F 0x1F 0x40 0x00
2626 // 5: 0x0F 0x1F 0x44 0x00 0x00
2627 // 6: 0x66 0x0F 0x1F 0x44 0x00 0x00
2628 // 7: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00
2629 // 8: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2630 // 9: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2631 // 10: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2632 // 11: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2634 // The rest coding is Intel specific - don't use consecutive address nops
2636 // 12: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90
2637 // 13: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90
2638 // 14: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90
2639 // 15: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90
2641 while(i >= 15) {
2642 // For Intel don't generate consecutive addess nops (mix with regular nops)
2643 i -= 15;
2644 emit_byte(0x66); // size prefix
2645 emit_byte(0x66); // size prefix
2646 emit_byte(0x66); // size prefix
2647 addr_nop_8();
2648 emit_byte(0x66); // size prefix
2649 emit_byte(0x66); // size prefix
2650 emit_byte(0x66); // size prefix
2651 emit_byte(0x90); // nop
2652 }
2653 switch (i) {
2654 case 14:
2655 emit_byte(0x66); // size prefix
2656 case 13:
2657 emit_byte(0x66); // size prefix
2658 case 12:
2659 addr_nop_8();
2660 emit_byte(0x66); // size prefix
2661 emit_byte(0x66); // size prefix
2662 emit_byte(0x66); // size prefix
2663 emit_byte(0x90); // nop
2664 break;
2665 case 11:
2666 emit_byte(0x66); // size prefix
2667 case 10:
2668 emit_byte(0x66); // size prefix
2669 case 9:
2670 emit_byte(0x66); // size prefix
2671 case 8:
2672 addr_nop_8();
2673 break;
2674 case 7:
2675 addr_nop_7();
2676 break;
2677 case 6:
2678 emit_byte(0x66); // size prefix
2679 case 5:
2680 addr_nop_5();
2681 break;
2682 case 4:
2683 addr_nop_4();
2684 break;
2685 case 3:
2686 // Don't use "0x0F 0x1F 0x00" - need patching safe padding
2687 emit_byte(0x66); // size prefix
2688 case 2:
2689 emit_byte(0x66); // size prefix
2690 case 1:
2691 emit_byte(0x90); // nop
2692 break;
2693 default:
2694 assert(i == 0, " ");
2695 }
2696 return;
2697 }
2698 if (UseAddressNop && VM_Version::is_amd()) {
2699 //
2700 // Using multi-bytes nops "0x0F 0x1F [address]" for AMD.
2701 // 1: 0x90
2702 // 2: 0x66 0x90
2703 // 3: 0x66 0x66 0x90 (don't use "0x0F 0x1F 0x00" - need patching safe padding)
2704 // 4: 0x0F 0x1F 0x40 0x00
2705 // 5: 0x0F 0x1F 0x44 0x00 0x00
2706 // 6: 0x66 0x0F 0x1F 0x44 0x00 0x00
2707 // 7: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00
2708 // 8: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2709 // 9: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2710 // 10: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2711 // 11: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2713 // The rest coding is AMD specific - use consecutive address nops
2715 // 12: 0x66 0x0F 0x1F 0x44 0x00 0x00 0x66 0x0F 0x1F 0x44 0x00 0x00
2716 // 13: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 0x66 0x0F 0x1F 0x44 0x00 0x00
2717 // 14: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00
2718 // 15: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00
2719 // 16: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2720 // Size prefixes (0x66) are added for larger sizes
2722 while(i >= 22) {
2723 i -= 11;
2724 emit_byte(0x66); // size prefix
2725 emit_byte(0x66); // size prefix
2726 emit_byte(0x66); // size prefix
2727 addr_nop_8();
2728 }
2729 // Generate first nop for size between 21-12
2730 switch (i) {
2731 case 21:
2732 i -= 1;
2733 emit_byte(0x66); // size prefix
2734 case 20:
2735 case 19:
2736 i -= 1;
2737 emit_byte(0x66); // size prefix
2738 case 18:
2739 case 17:
2740 i -= 1;
2741 emit_byte(0x66); // size prefix
2742 case 16:
2743 case 15:
2744 i -= 8;
2745 addr_nop_8();
2746 break;
2747 case 14:
2748 case 13:
2749 i -= 7;
2750 addr_nop_7();
2751 break;
2752 case 12:
2753 i -= 6;
2754 emit_byte(0x66); // size prefix
2755 addr_nop_5();
2756 break;
2757 default:
2758 assert(i < 12, " ");
2759 }
2761 // Generate second nop for size between 11-1
2762 switch (i) {
2763 case 11:
2764 emit_byte(0x66); // size prefix
2765 case 10:
2766 emit_byte(0x66); // size prefix
2767 case 9:
2768 emit_byte(0x66); // size prefix
2769 case 8:
2770 addr_nop_8();
2771 break;
2772 case 7:
2773 addr_nop_7();
2774 break;
2775 case 6:
2776 emit_byte(0x66); // size prefix
2777 case 5:
2778 addr_nop_5();
2779 break;
2780 case 4:
2781 addr_nop_4();
2782 break;
2783 case 3:
2784 // Don't use "0x0F 0x1F 0x00" - need patching safe padding
2785 emit_byte(0x66); // size prefix
2786 case 2:
2787 emit_byte(0x66); // size prefix
2788 case 1:
2789 emit_byte(0x90); // nop
2790 break;
2791 default:
2792 assert(i == 0, " ");
2793 }
2794 return;
2795 }
2797 // Using nops with size prefixes "0x66 0x90".
2798 // From AMD Optimization Guide:
2799 // 1: 0x90
2800 // 2: 0x66 0x90
2801 // 3: 0x66 0x66 0x90
2802 // 4: 0x66 0x66 0x66 0x90
2803 // 5: 0x66 0x66 0x90 0x66 0x90
2804 // 6: 0x66 0x66 0x90 0x66 0x66 0x90
2805 // 7: 0x66 0x66 0x66 0x90 0x66 0x66 0x90
2806 // 8: 0x66 0x66 0x66 0x90 0x66 0x66 0x66 0x90
2807 // 9: 0x66 0x66 0x90 0x66 0x66 0x90 0x66 0x66 0x90
2808 // 10: 0x66 0x66 0x66 0x90 0x66 0x66 0x90 0x66 0x66 0x90
2809 //
2810 while(i > 12) {
2811 i -= 4;
2812 emit_byte(0x66); // size prefix
2813 emit_byte(0x66);
2814 emit_byte(0x66);
2815 emit_byte(0x90); // nop
2816 }
2817 // 1 - 12 nops
2818 if(i > 8) {
2819 if(i > 9) {
2820 i -= 1;
2821 emit_byte(0x66);
2822 }
2823 i -= 3;
2824 emit_byte(0x66);
2825 emit_byte(0x66);
2826 emit_byte(0x90);
2827 }
2828 // 1 - 8 nops
2829 if(i > 4) {
2830 if(i > 6) {
2831 i -= 1;
2832 emit_byte(0x66);
2833 }
2834 i -= 3;
2835 emit_byte(0x66);
2836 emit_byte(0x66);
2837 emit_byte(0x90);
2838 }
2839 switch (i) {
2840 case 4:
2841 emit_byte(0x66);
2842 case 3:
2843 emit_byte(0x66);
2844 case 2:
2845 emit_byte(0x66);
2846 case 1:
2847 emit_byte(0x90);
2848 break;
2849 default:
2850 assert(i == 0, " ");
2851 }
2852 }
2854 void Assembler::ret(int imm16) {
2855 if (imm16 == 0) {
2856 emit_byte(0xC3);
2857 } else {
2858 emit_byte(0xC2);
2859 emit_word(imm16);
2860 }
2861 }
2863 // copies a single word from [esi] to [edi]
2864 void Assembler::smovl() {
2865 emit_byte(0xA5);
2866 }
2868 // copies data from [rsi] to [rdi] using rcx words (m32)
2869 void Assembler::rep_movl() {
2870 // REP
2871 emit_byte(0xF3);
2872 // MOVSL
2873 emit_byte(0xA5);
2874 }
2876 // copies data from [rsi] to [rdi] using rcx double words (m64)
2877 void Assembler::rep_movq() {
2878 // REP
2879 emit_byte(0xF3);
2880 // MOVSQ
2881 prefix(REX_W);
2882 emit_byte(0xA5);
2883 }
2885 // sets rcx double words (m64) with rax value at [rdi]
2886 void Assembler::rep_set() {
2887 // REP
2888 emit_byte(0xF3);
2889 // STOSQ
2890 prefix(REX_W);
2891 emit_byte(0xAB);
2892 }
2894 // scans rcx double words (m64) at [rdi] for occurance of rax
2895 void Assembler::repne_scanq() {
2896 // REPNE/REPNZ
2897 emit_byte(0xF2);
2898 // SCASQ
2899 prefix(REX_W);
2900 emit_byte(0xAF);
2901 }
2903 void Assembler::repne_scanl() {
2904 // REPNE/REPNZ
2905 emit_byte(0xF2);
2906 // SCASL
2907 emit_byte(0xAF);
2908 }
2911 void Assembler::setb(Condition cc, Register dst) {
2912 assert(0 <= cc && cc < 16, "illegal cc");
2913 int encode = prefix_and_encode(dst->encoding(), true);
2914 emit_byte(0x0F);
2915 emit_byte(0x90 | cc);
2916 emit_byte(0xC0 | encode);
2917 }
2919 void Assembler::clflush(Address adr) {
2920 prefix(adr);
2921 emit_byte(0x0F);
2922 emit_byte(0xAE);
2923 emit_operand(rdi, adr);
2924 }
2926 void Assembler::call(Label& L, relocInfo::relocType rtype) {
2927 if (L.is_bound()) {
2928 const int long_size = 5;
2929 int offs = (int)( target(L) - pc() );
2930 assert(offs <= 0, "assembler error");
2931 InstructionMark im(this);
2932 // 1110 1000 #32-bit disp
2933 emit_byte(0xE8);
2934 emit_data(offs - long_size, rtype, disp32_operand);
2935 } else {
2936 InstructionMark im(this);
2937 // 1110 1000 #32-bit disp
2938 L.add_patch_at(code(), locator());
2940 emit_byte(0xE8);
2941 emit_data(int(0), rtype, disp32_operand);
2942 }
2943 }
2945 void Assembler::call_literal(address entry, RelocationHolder const& rspec) {
2946 assert(entry != NULL, "call most probably wrong");
2947 InstructionMark im(this);
2948 emit_byte(0xE8);
2949 intptr_t disp = entry - (_code_pos + sizeof(int32_t));
2950 assert(is_simm32(disp), "must be 32bit offset (call2)");
2951 // Technically, should use call32_operand, but this format is
2952 // implied by the fact that we're emitting a call instruction.
2953 emit_data((int) disp, rspec, disp32_operand);
2954 }
2957 void Assembler::call(Register dst) {
2958 // This was originally using a 32bit register encoding
2959 // and surely we want 64bit!
2960 // this is a 32bit encoding but in 64bit mode the default
2961 // operand size is 64bit so there is no need for the
2962 // wide prefix. So prefix only happens if we use the
2963 // new registers. Much like push/pop.
2964 int encode = prefixq_and_encode(dst->encoding());
2965 emit_byte(0xFF);
2966 emit_byte(0xD0 | encode);
2967 }
2969 void Assembler::call(Address adr) {
2970 InstructionMark im(this);
2971 prefix(adr);
2972 emit_byte(0xFF);
2973 emit_operand(rdx, adr);
2974 }
2976 void Assembler::jmp(Register reg) {
2977 int encode = prefix_and_encode(reg->encoding());
2978 emit_byte(0xFF);
2979 emit_byte(0xE0 | encode);
2980 }
2982 void Assembler::jmp(Address adr) {
2983 InstructionMark im(this);
2984 prefix(adr);
2985 emit_byte(0xFF);
2986 emit_operand(rsp, adr);
2987 }
2989 void Assembler::jmp_literal(address dest, RelocationHolder const& rspec) {
2990 InstructionMark im(this);
2991 emit_byte(0xE9);
2992 assert(dest != NULL, "must have a target");
2993 intptr_t disp = dest - (_code_pos + sizeof(int32_t));
2994 assert(is_simm32(disp), "must be 32bit offset (jmp)");
2995 emit_data(disp, rspec.reloc(), call32_operand);
2996 }
2998 void Assembler::jmp(Label& L, relocInfo::relocType rtype) {
2999 if (L.is_bound()) {
3000 address entry = target(L);
3001 assert(entry != NULL, "jmp most probably wrong");
3002 InstructionMark im(this);
3003 const int short_size = 2;
3004 const int long_size = 5;
3005 intptr_t offs = entry - _code_pos;
3006 if (rtype == relocInfo::none && is8bit(offs - short_size)) {
3007 emit_byte(0xEB);
3008 emit_byte((offs - short_size) & 0xFF);
3009 } else {
3010 emit_byte(0xE9);
3011 emit_long(offs - long_size);
3012 }
3013 } else {
3014 // By default, forward jumps are always 32-bit displacements, since
3015 // we can't yet know where the label will be bound. If you're sure that
3016 // the forward jump will not run beyond 256 bytes, use jmpb to
3017 // force an 8-bit displacement.
3018 InstructionMark im(this);
3019 relocate(rtype);
3020 L.add_patch_at(code(), locator());
3021 emit_byte(0xE9);
3022 emit_long(0);
3023 }
3024 }
3026 void Assembler::jmpb(Label& L) {
3027 if (L.is_bound()) {
3028 const int short_size = 2;
3029 address entry = target(L);
3030 assert(is8bit((entry - _code_pos) + short_size),
3031 "Dispacement too large for a short jmp");
3032 assert(entry != NULL, "jmp most probably wrong");
3033 intptr_t offs = entry - _code_pos;
3034 emit_byte(0xEB);
3035 emit_byte((offs - short_size) & 0xFF);
3036 } else {
3037 InstructionMark im(this);
3038 L.add_patch_at(code(), locator());
3039 emit_byte(0xEB);
3040 emit_byte(0);
3041 }
3042 }
3044 void Assembler::jcc(Condition cc, Label& L, relocInfo::relocType rtype) {
3045 InstructionMark im(this);
3046 relocate(rtype);
3047 assert((0 <= cc) && (cc < 16), "illegal cc");
3048 if (L.is_bound()) {
3049 address dst = target(L);
3050 assert(dst != NULL, "jcc most probably wrong");
3052 const int short_size = 2;
3053 const int long_size = 6;
3054 intptr_t offs = (intptr_t)dst - (intptr_t)_code_pos;
3055 if (rtype == relocInfo::none && is8bit(offs - short_size)) {
3056 // 0111 tttn #8-bit disp
3057 emit_byte(0x70 | cc);
3058 emit_byte((offs - short_size) & 0xFF);
3059 } else {
3060 // 0000 1111 1000 tttn #32-bit disp
3061 assert(is_simm32(offs - long_size),
3062 "must be 32bit offset (call4)");
3063 emit_byte(0x0F);
3064 emit_byte(0x80 | cc);
3065 emit_long(offs - long_size);
3066 }
3067 } else {
3068 // Note: could eliminate cond. jumps to this jump if condition
3069 // is the same however, seems to be rather unlikely case.
3070 // Note: use jccb() if label to be bound is very close to get
3071 // an 8-bit displacement
3072 L.add_patch_at(code(), locator());
3073 emit_byte(0x0F);
3074 emit_byte(0x80 | cc);
3075 emit_long(0);
3076 }
3077 }
3079 void Assembler::jccb(Condition cc, Label& L) {
3080 if (L.is_bound()) {
3081 const int short_size = 2;
3082 const int long_size = 6;
3083 address entry = target(L);
3084 assert(is8bit((intptr_t)entry - ((intptr_t)_code_pos + short_size)),
3085 "Dispacement too large for a short jmp");
3086 intptr_t offs = (intptr_t)entry - (intptr_t)_code_pos;
3087 // 0111 tttn #8-bit disp
3088 emit_byte(0x70 | cc);
3089 emit_byte((offs - short_size) & 0xFF);
3090 } else {
3091 InstructionMark im(this);
3092 L.add_patch_at(code(), locator());
3093 emit_byte(0x70 | cc);
3094 emit_byte(0);
3095 }
3096 }
3098 // FP instructions
3100 void Assembler::fxsave(Address dst) {
3101 prefixq(dst);
3102 emit_byte(0x0F);
3103 emit_byte(0xAE);
3104 emit_operand(as_Register(0), dst);
3105 }
3107 void Assembler::fxrstor(Address src) {
3108 prefixq(src);
3109 emit_byte(0x0F);
3110 emit_byte(0xAE);
3111 emit_operand(as_Register(1), src);
3112 }
3114 void Assembler::ldmxcsr(Address src) {
3115 InstructionMark im(this);
3116 prefix(src);
3117 emit_byte(0x0F);
3118 emit_byte(0xAE);
3119 emit_operand(as_Register(2), src);
3120 }
3122 void Assembler::stmxcsr(Address dst) {
3123 InstructionMark im(this);
3124 prefix(dst);
3125 emit_byte(0x0F);
3126 emit_byte(0xAE);
3127 emit_operand(as_Register(3), dst);
3128 }
3130 void Assembler::addss(XMMRegister dst, XMMRegister src) {
3131 emit_byte(0xF3);
3132 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3133 emit_byte(0x0F);
3134 emit_byte(0x58);
3135 emit_byte(0xC0 | encode);
3136 }
3138 void Assembler::addss(XMMRegister dst, Address src) {
3139 InstructionMark im(this);
3140 emit_byte(0xF3);
3141 prefix(src, dst);
3142 emit_byte(0x0F);
3143 emit_byte(0x58);
3144 emit_operand(dst, src);
3145 }
3147 void Assembler::subss(XMMRegister dst, XMMRegister src) {
3148 emit_byte(0xF3);
3149 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3150 emit_byte(0x0F);
3151 emit_byte(0x5C);
3152 emit_byte(0xC0 | encode);
3153 }
3155 void Assembler::subss(XMMRegister dst, Address src) {
3156 InstructionMark im(this);
3157 emit_byte(0xF3);
3158 prefix(src, dst);
3159 emit_byte(0x0F);
3160 emit_byte(0x5C);
3161 emit_operand(dst, src);
3162 }
3164 void Assembler::mulss(XMMRegister dst, XMMRegister src) {
3165 emit_byte(0xF3);
3166 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3167 emit_byte(0x0F);
3168 emit_byte(0x59);
3169 emit_byte(0xC0 | encode);
3170 }
3172 void Assembler::mulss(XMMRegister dst, Address src) {
3173 InstructionMark im(this);
3174 emit_byte(0xF3);
3175 prefix(src, dst);
3176 emit_byte(0x0F);
3177 emit_byte(0x59);
3178 emit_operand(dst, src);
3179 }
3181 void Assembler::divss(XMMRegister dst, XMMRegister src) {
3182 emit_byte(0xF3);
3183 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3184 emit_byte(0x0F);
3185 emit_byte(0x5E);
3186 emit_byte(0xC0 | encode);
3187 }
3189 void Assembler::divss(XMMRegister dst, Address src) {
3190 InstructionMark im(this);
3191 emit_byte(0xF3);
3192 prefix(src, dst);
3193 emit_byte(0x0F);
3194 emit_byte(0x5E);
3195 emit_operand(dst, src);
3196 }
3198 void Assembler::addsd(XMMRegister dst, XMMRegister src) {
3199 emit_byte(0xF2);
3200 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3201 emit_byte(0x0F);
3202 emit_byte(0x58);
3203 emit_byte(0xC0 | encode);
3204 }
3206 void Assembler::addsd(XMMRegister dst, Address src) {
3207 InstructionMark im(this);
3208 emit_byte(0xF2);
3209 prefix(src, dst);
3210 emit_byte(0x0F);
3211 emit_byte(0x58);
3212 emit_operand(dst, src);
3213 }
3215 void Assembler::subsd(XMMRegister dst, XMMRegister src) {
3216 emit_byte(0xF2);
3217 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3218 emit_byte(0x0F);
3219 emit_byte(0x5C);
3220 emit_byte(0xC0 | encode);
3221 }
3223 void Assembler::subsd(XMMRegister dst, Address src) {
3224 InstructionMark im(this);
3225 emit_byte(0xF2);
3226 prefix(src, dst);
3227 emit_byte(0x0F);
3228 emit_byte(0x5C);
3229 emit_operand(dst, src);
3230 }
3232 void Assembler::mulsd(XMMRegister dst, XMMRegister src) {
3233 emit_byte(0xF2);
3234 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3235 emit_byte(0x0F);
3236 emit_byte(0x59);
3237 emit_byte(0xC0 | encode);
3238 }
3240 void Assembler::mulsd(XMMRegister dst, Address src) {
3241 InstructionMark im(this);
3242 emit_byte(0xF2);
3243 prefix(src, dst);
3244 emit_byte(0x0F);
3245 emit_byte(0x59);
3246 emit_operand(dst, src);
3247 }
3249 void Assembler::divsd(XMMRegister dst, XMMRegister src) {
3250 emit_byte(0xF2);
3251 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3252 emit_byte(0x0F);
3253 emit_byte(0x5E);
3254 emit_byte(0xC0 | encode);
3255 }
3257 void Assembler::divsd(XMMRegister dst, Address src) {
3258 InstructionMark im(this);
3259 emit_byte(0xF2);
3260 prefix(src, dst);
3261 emit_byte(0x0F);
3262 emit_byte(0x5E);
3263 emit_operand(dst, src);
3264 }
3266 void Assembler::sqrtsd(XMMRegister dst, XMMRegister src) {
3267 emit_byte(0xF2);
3268 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3269 emit_byte(0x0F);
3270 emit_byte(0x51);
3271 emit_byte(0xC0 | encode);
3272 }
3274 void Assembler::sqrtsd(XMMRegister dst, Address src) {
3275 InstructionMark im(this);
3276 emit_byte(0xF2);
3277 prefix(src, dst);
3278 emit_byte(0x0F);
3279 emit_byte(0x51);
3280 emit_operand(dst, src);
3281 }
3283 void Assembler::xorps(XMMRegister dst, XMMRegister src) {
3284 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3285 emit_byte(0x0F);
3286 emit_byte(0x57);
3287 emit_byte(0xC0 | encode);
3288 }
3290 void Assembler::xorps(XMMRegister dst, Address src) {
3291 InstructionMark im(this);
3292 prefix(src, dst);
3293 emit_byte(0x0F);
3294 emit_byte(0x57);
3295 emit_operand(dst, src);
3296 }
3298 void Assembler::xorpd(XMMRegister dst, XMMRegister src) {
3299 emit_byte(0x66);
3300 xorps(dst, src);
3301 }
3303 void Assembler::xorpd(XMMRegister dst, Address src) {
3304 InstructionMark im(this);
3305 emit_byte(0x66);
3306 prefix(src, dst);
3307 emit_byte(0x0F);
3308 emit_byte(0x57);
3309 emit_operand(dst, src);
3310 }
3312 void Assembler::cvtsi2ssl(XMMRegister dst, Register src) {
3313 emit_byte(0xF3);
3314 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3315 emit_byte(0x0F);
3316 emit_byte(0x2A);
3317 emit_byte(0xC0 | encode);
3318 }
3320 void Assembler::cvtsi2ssq(XMMRegister dst, Register src) {
3321 emit_byte(0xF3);
3322 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3323 emit_byte(0x0F);
3324 emit_byte(0x2A);
3325 emit_byte(0xC0 | encode);
3326 }
3328 void Assembler::cvtsi2sdl(XMMRegister dst, Register src) {
3329 emit_byte(0xF2);
3330 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3331 emit_byte(0x0F);
3332 emit_byte(0x2A);
3333 emit_byte(0xC0 | encode);
3334 }
3336 void Assembler::cvtsi2sdq(XMMRegister dst, Register src) {
3337 emit_byte(0xF2);
3338 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3339 emit_byte(0x0F);
3340 emit_byte(0x2A);
3341 emit_byte(0xC0 | encode);
3342 }
3344 void Assembler::cvttss2sil(Register dst, XMMRegister src) {
3345 emit_byte(0xF3);
3346 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3347 emit_byte(0x0F);
3348 emit_byte(0x2C);
3349 emit_byte(0xC0 | encode);
3350 }
3352 void Assembler::cvttss2siq(Register dst, XMMRegister src) {
3353 emit_byte(0xF3);
3354 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3355 emit_byte(0x0F);
3356 emit_byte(0x2C);
3357 emit_byte(0xC0 | encode);
3358 }
3360 void Assembler::cvttsd2sil(Register dst, XMMRegister src) {
3361 emit_byte(0xF2);
3362 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3363 emit_byte(0x0F);
3364 emit_byte(0x2C);
3365 emit_byte(0xC0 | encode);
3366 }
3368 void Assembler::cvttsd2siq(Register dst, XMMRegister src) {
3369 emit_byte(0xF2);
3370 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3371 emit_byte(0x0F);
3372 emit_byte(0x2C);
3373 emit_byte(0xC0 | encode);
3374 }
3376 void Assembler::cvtss2sd(XMMRegister dst, XMMRegister src) {
3377 emit_byte(0xF3);
3378 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3379 emit_byte(0x0F);
3380 emit_byte(0x5A);
3381 emit_byte(0xC0 | encode);
3382 }
3384 void Assembler::cvtdq2pd(XMMRegister dst, XMMRegister src) {
3385 emit_byte(0xF3);
3386 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3387 emit_byte(0x0F);
3388 emit_byte(0xE6);
3389 emit_byte(0xC0 | encode);
3390 }
3392 void Assembler::cvtdq2ps(XMMRegister dst, XMMRegister src) {
3393 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3394 emit_byte(0x0F);
3395 emit_byte(0x5B);
3396 emit_byte(0xC0 | encode);
3397 }
3399 void Assembler::cvtsd2ss(XMMRegister dst, XMMRegister src) {
3400 emit_byte(0xF2);
3401 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3402 emit_byte(0x0F);
3403 emit_byte(0x5A);
3404 emit_byte(0xC0 | encode);
3405 }
3407 void Assembler::punpcklbw(XMMRegister dst, XMMRegister src) {
3408 emit_byte(0x66);
3409 int encode = prefix_and_encode(dst->encoding(), src->encoding());
3410 emit_byte(0x0F);
3411 emit_byte(0x60);
3412 emit_byte(0xC0 | encode);
3413 }
3415 // Implementation of MacroAssembler
3417 // On 32 bit it returns a vanilla displacement on 64 bit is a rip relative displacement
3418 Address MacroAssembler::as_Address(AddressLiteral adr) {
3419 assert(!adr.is_lval(), "must be rval");
3420 assert(reachable(adr), "must be");
3421 return Address((int)(intptr_t)(adr.target() - pc()), adr.target(), adr.reloc());
3422 }
3424 Address MacroAssembler::as_Address(ArrayAddress adr) {
3425 #ifdef _LP64
3426 AddressLiteral base = adr.base();
3427 lea(rscratch1, base);
3428 Address index = adr.index();
3429 assert(index._disp == 0, "must not have disp"); // maybe it can?
3430 Address array(rscratch1, index._index, index._scale, index._disp);
3431 return array;
3432 #else
3433 return Address::make_array(adr);
3434 #endif // _LP64
3436 }
3438 void MacroAssembler::fat_nop() {
3439 // A 5 byte nop that is safe for patching (see patch_verified_entry)
3440 // Recommened sequence from 'Software Optimization Guide for the AMD
3441 // Hammer Processor'
3442 emit_byte(0x66);
3443 emit_byte(0x66);
3444 emit_byte(0x90);
3445 emit_byte(0x66);
3446 emit_byte(0x90);
3447 }
3449 static Assembler::Condition reverse[] = {
3450 Assembler::noOverflow /* overflow = 0x0 */ ,
3451 Assembler::overflow /* noOverflow = 0x1 */ ,
3452 Assembler::aboveEqual /* carrySet = 0x2, below = 0x2 */ ,
3453 Assembler::below /* aboveEqual = 0x3, carryClear = 0x3 */ ,
3454 Assembler::notZero /* zero = 0x4, equal = 0x4 */ ,
3455 Assembler::zero /* notZero = 0x5, notEqual = 0x5 */ ,
3456 Assembler::above /* belowEqual = 0x6 */ ,
3457 Assembler::belowEqual /* above = 0x7 */ ,
3458 Assembler::positive /* negative = 0x8 */ ,
3459 Assembler::negative /* positive = 0x9 */ ,
3460 Assembler::noParity /* parity = 0xa */ ,
3461 Assembler::parity /* noParity = 0xb */ ,
3462 Assembler::greaterEqual /* less = 0xc */ ,
3463 Assembler::less /* greaterEqual = 0xd */ ,
3464 Assembler::greater /* lessEqual = 0xe */ ,
3465 Assembler::lessEqual /* greater = 0xf, */
3467 };
3469 // 32bit can do a case table jump in one instruction but we no longer allow the base
3470 // to be installed in the Address class
3471 void MacroAssembler::jump(ArrayAddress entry) {
3472 #ifdef _LP64
3473 lea(rscratch1, entry.base());
3474 Address dispatch = entry.index();
3475 assert(dispatch._base == noreg, "must be");
3476 dispatch._base = rscratch1;
3477 jmp(dispatch);
3478 #else
3479 jmp(as_Address(entry));
3480 #endif // _LP64
3481 }
3483 void MacroAssembler::jump(AddressLiteral dst) {
3484 if (reachable(dst)) {
3485 jmp_literal(dst.target(), dst.rspec());
3486 } else {
3487 lea(rscratch1, dst);
3488 jmp(rscratch1);
3489 }
3490 }
3492 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst) {
3493 if (reachable(dst)) {
3494 InstructionMark im(this);
3495 relocate(dst.reloc());
3496 const int short_size = 2;
3497 const int long_size = 6;
3498 int offs = (intptr_t)dst.target() - ((intptr_t)_code_pos);
3499 if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) {
3500 // 0111 tttn #8-bit disp
3501 emit_byte(0x70 | cc);
3502 emit_byte((offs - short_size) & 0xFF);
3503 } else {
3504 // 0000 1111 1000 tttn #32-bit disp
3505 emit_byte(0x0F);
3506 emit_byte(0x80 | cc);
3507 emit_long(offs - long_size);
3508 }
3509 } else {
3510 #ifdef ASSERT
3511 warning("reversing conditional branch");
3512 #endif /* ASSERT */
3513 Label skip;
3514 jccb(reverse[cc], skip);
3515 lea(rscratch1, dst);
3516 Assembler::jmp(rscratch1);
3517 bind(skip);
3518 }
3519 }
3521 // Wouldn't need if AddressLiteral version had new name
3522 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) {
3523 Assembler::call(L, rtype);
3524 }
3526 // Wouldn't need if AddressLiteral version had new name
3527 void MacroAssembler::call(Register entry) {
3528 Assembler::call(entry);
3529 }
3531 void MacroAssembler::call(AddressLiteral entry) {
3532 if (reachable(entry)) {
3533 Assembler::call_literal(entry.target(), entry.rspec());
3534 } else {
3535 lea(rscratch1, entry);
3536 Assembler::call(rscratch1);
3537 }
3538 }
3540 void MacroAssembler::cmp8(AddressLiteral src1, int8_t src2) {
3541 if (reachable(src1)) {
3542 cmpb(as_Address(src1), src2);
3543 } else {
3544 lea(rscratch1, src1);
3545 cmpb(Address(rscratch1, 0), src2);
3546 }
3547 }
3549 void MacroAssembler::cmp32(AddressLiteral src1, int32_t src2) {
3550 if (reachable(src1)) {
3551 cmpl(as_Address(src1), src2);
3552 } else {
3553 lea(rscratch1, src1);
3554 cmpl(Address(rscratch1, 0), src2);
3555 }
3556 }
3558 void MacroAssembler::cmp32(Register src1, AddressLiteral src2) {
3559 if (reachable(src2)) {
3560 cmpl(src1, as_Address(src2));
3561 } else {
3562 lea(rscratch1, src2);
3563 cmpl(src1, Address(rscratch1, 0));
3564 }
3565 }
3567 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2) {
3568 #ifdef _LP64
3569 if (src2.is_lval()) {
3570 movptr(rscratch1, src2);
3571 Assembler::cmpq(src1, rscratch1);
3572 } else if (reachable(src2)) {
3573 cmpq(src1, as_Address(src2));
3574 } else {
3575 lea(rscratch1, src2);
3576 Assembler::cmpq(src1, Address(rscratch1, 0));
3577 }
3578 #else
3579 if (src2.is_lval()) {
3580 cmp_literal32(src1, (int32_t) src2.target(), src2.rspec());
3581 } else {
3582 cmpl(src1, as_Address(src2));
3583 }
3584 #endif // _LP64
3585 }
3587 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2) {
3588 assert(src2.is_lval(), "not a mem-mem compare");
3589 #ifdef _LP64
3590 // moves src2's literal address
3591 movptr(rscratch1, src2);
3592 Assembler::cmpq(src1, rscratch1);
3593 #else
3594 cmp_literal32(src1, (int32_t) src2.target(), src2.rspec());
3595 #endif // _LP64
3596 }
3598 void MacroAssembler::cmp64(Register src1, AddressLiteral src2) {
3599 assert(!src2.is_lval(), "should use cmpptr");
3601 if (reachable(src2)) {
3602 #ifdef _LP64
3603 cmpq(src1, as_Address(src2));
3604 #else
3605 ShouldNotReachHere();
3606 #endif // _LP64
3607 } else {
3608 lea(rscratch1, src2);
3609 Assembler::cmpq(src1, Address(rscratch1, 0));
3610 }
3611 }
3613 void MacroAssembler::cmpxchgptr(Register reg, AddressLiteral adr) {
3614 if (reachable(adr)) {
3615 #ifdef _LP64
3616 cmpxchgq(reg, as_Address(adr));
3617 #else
3618 cmpxchgl(reg, as_Address(adr));
3619 #endif // _LP64
3620 } else {
3621 lea(rscratch1, adr);
3622 cmpxchgq(reg, Address(rscratch1, 0));
3623 }
3624 }
3626 void MacroAssembler::incrementl(AddressLiteral dst) {
3627 if (reachable(dst)) {
3628 incrementl(as_Address(dst));
3629 } else {
3630 lea(rscratch1, dst);
3631 incrementl(Address(rscratch1, 0));
3632 }
3633 }
3635 void MacroAssembler::incrementl(ArrayAddress dst) {
3636 incrementl(as_Address(dst));
3637 }
3639 void MacroAssembler::lea(Register dst, Address src) {
3640 #ifdef _LP64
3641 leaq(dst, src);
3642 #else
3643 leal(dst, src);
3644 #endif // _LP64
3645 }
3647 void MacroAssembler::lea(Register dst, AddressLiteral src) {
3648 #ifdef _LP64
3649 mov_literal64(dst, (intptr_t)src.target(), src.rspec());
3650 #else
3651 mov_literal32(dst, (intptr_t)src.target(), src.rspec());
3652 #endif // _LP64
3653 }
3655 void MacroAssembler::mov32(AddressLiteral dst, Register src) {
3656 if (reachable(dst)) {
3657 movl(as_Address(dst), src);
3658 } else {
3659 lea(rscratch1, dst);
3660 movl(Address(rscratch1, 0), src);
3661 }
3662 }
3664 void MacroAssembler::mov32(Register dst, AddressLiteral src) {
3665 if (reachable(src)) {
3666 movl(dst, as_Address(src));
3667 } else {
3668 lea(rscratch1, src);
3669 movl(dst, Address(rscratch1, 0));
3670 }
3671 }
3673 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src) {
3674 if (reachable(src)) {
3675 if (UseXmmLoadAndClearUpper) {
3676 movsd (dst, as_Address(src));
3677 } else {
3678 movlpd(dst, as_Address(src));
3679 }
3680 } else {
3681 lea(rscratch1, src);
3682 if (UseXmmLoadAndClearUpper) {
3683 movsd (dst, Address(rscratch1, 0));
3684 } else {
3685 movlpd(dst, Address(rscratch1, 0));
3686 }
3687 }
3688 }
3690 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src) {
3691 if (reachable(src)) {
3692 movss(dst, as_Address(src));
3693 } else {
3694 lea(rscratch1, src);
3695 movss(dst, Address(rscratch1, 0));
3696 }
3697 }
3699 void MacroAssembler::movoop(Register dst, jobject obj) {
3700 mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate());
3701 }
3703 void MacroAssembler::movoop(Address dst, jobject obj) {
3704 mov_literal64(rscratch1, (intptr_t)obj, oop_Relocation::spec_for_immediate());
3705 movq(dst, rscratch1);
3706 }
3708 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
3709 #ifdef _LP64
3710 if (src.is_lval()) {
3711 mov_literal64(dst, (intptr_t)src.target(), src.rspec());
3712 } else {
3713 if (reachable(src)) {
3714 movq(dst, as_Address(src));
3715 } else {
3716 lea(rscratch1, src);
3717 movq(dst, Address(rscratch1,0));
3718 }
3719 }
3720 #else
3721 if (src.is_lval()) {
3722 mov_literal32(dst, (intptr_t)src.target(), src.rspec());
3723 } else {
3724 movl(dst, as_Address(src));
3725 }
3726 #endif // LP64
3727 }
3729 void MacroAssembler::movptr(ArrayAddress dst, Register src) {
3730 #ifdef _LP64
3731 movq(as_Address(dst), src);
3732 #else
3733 movl(as_Address(dst), src);
3734 #endif // _LP64
3735 }
3737 void MacroAssembler::pushoop(jobject obj) {
3738 #ifdef _LP64
3739 movoop(rscratch1, obj);
3740 pushq(rscratch1);
3741 #else
3742 push_literal32((int32_t)obj, oop_Relocation::spec_for_immediate());
3743 #endif // _LP64
3744 }
3746 void MacroAssembler::pushptr(AddressLiteral src) {
3747 #ifdef _LP64
3748 lea(rscratch1, src);
3749 if (src.is_lval()) {
3750 pushq(rscratch1);
3751 } else {
3752 pushq(Address(rscratch1, 0));
3753 }
3754 #else
3755 if (src.is_lval()) {
3756 push_literal((int32_t)src.target(), src.rspec());
3757 else {
3758 pushl(as_Address(src));
3759 }
3760 #endif // _LP64
3761 }
3763 void MacroAssembler::ldmxcsr(AddressLiteral src) {
3764 if (reachable(src)) {
3765 Assembler::ldmxcsr(as_Address(src));
3766 } else {
3767 lea(rscratch1, src);
3768 Assembler::ldmxcsr(Address(rscratch1, 0));
3769 }
3770 }
3772 void MacroAssembler::movlpd(XMMRegister dst, AddressLiteral src) {
3773 if (reachable(src)) {
3774 movlpd(dst, as_Address(src));
3775 } else {
3776 lea(rscratch1, src);
3777 movlpd(dst, Address(rscratch1, 0));
3778 }
3779 }
3781 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src) {
3782 if (reachable(src)) {
3783 movss(dst, as_Address(src));
3784 } else {
3785 lea(rscratch1, src);
3786 movss(dst, Address(rscratch1, 0));
3787 }
3788 }
3789 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src) {
3790 if (reachable(src)) {
3791 xorpd(dst, as_Address(src));
3792 } else {
3793 lea(rscratch1, src);
3794 xorpd(dst, Address(rscratch1, 0));
3795 }
3796 }
3798 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src) {
3799 if (reachable(src)) {
3800 xorps(dst, as_Address(src));
3801 } else {
3802 lea(rscratch1, src);
3803 xorps(dst, Address(rscratch1, 0));
3804 }
3805 }
3807 void MacroAssembler::null_check(Register reg, int offset) {
3808 if (needs_explicit_null_check(offset)) {
3809 // provoke OS NULL exception if reg = NULL by
3810 // accessing M[reg] w/o changing any (non-CC) registers
3811 cmpq(rax, Address(reg, 0));
3812 // Note: should probably use testl(rax, Address(reg, 0));
3813 // may be shorter code (however, this version of
3814 // testl needs to be implemented first)
3815 } else {
3816 // nothing to do, (later) access of M[reg + offset]
3817 // will provoke OS NULL exception if reg = NULL
3818 }
3819 }
3821 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
3822 int off = offset();
3823 movzbl(dst, src);
3824 return off;
3825 }
3827 int MacroAssembler::load_unsigned_word(Register dst, Address src) {
3828 int off = offset();
3829 movzwl(dst, src);
3830 return off;
3831 }
3833 int MacroAssembler::load_signed_byte(Register dst, Address src) {
3834 int off = offset();
3835 movsbl(dst, src);
3836 return off;
3837 }
3839 int MacroAssembler::load_signed_word(Register dst, Address src) {
3840 int off = offset();
3841 movswl(dst, src);
3842 return off;
3843 }
3845 void MacroAssembler::incrementl(Register reg, int value) {
3846 if (value == min_jint) { addl(reg, value); return; }
3847 if (value < 0) { decrementl(reg, -value); return; }
3848 if (value == 0) { ; return; }
3849 if (value == 1 && UseIncDec) { incl(reg) ; return; }
3850 /* else */ { addl(reg, value) ; return; }
3851 }
3853 void MacroAssembler::decrementl(Register reg, int value) {
3854 if (value == min_jint) { subl(reg, value); return; }
3855 if (value < 0) { incrementl(reg, -value); return; }
3856 if (value == 0) { ; return; }
3857 if (value == 1 && UseIncDec) { decl(reg) ; return; }
3858 /* else */ { subl(reg, value) ; return; }
3859 }
3861 void MacroAssembler::incrementq(Register reg, int value) {
3862 if (value == min_jint) { addq(reg, value); return; }
3863 if (value < 0) { decrementq(reg, -value); return; }
3864 if (value == 0) { ; return; }
3865 if (value == 1 && UseIncDec) { incq(reg) ; return; }
3866 /* else */ { addq(reg, value) ; return; }
3867 }
3869 void MacroAssembler::decrementq(Register reg, int value) {
3870 if (value == min_jint) { subq(reg, value); return; }
3871 if (value < 0) { incrementq(reg, -value); return; }
3872 if (value == 0) { ; return; }
3873 if (value == 1 && UseIncDec) { decq(reg) ; return; }
3874 /* else */ { subq(reg, value) ; return; }
3875 }
3877 void MacroAssembler::incrementl(Address dst, int value) {
3878 if (value == min_jint) { addl(dst, value); return; }
3879 if (value < 0) { decrementl(dst, -value); return; }
3880 if (value == 0) { ; return; }
3881 if (value == 1 && UseIncDec) { incl(dst) ; return; }
3882 /* else */ { addl(dst, value) ; return; }
3883 }
3885 void MacroAssembler::decrementl(Address dst, int value) {
3886 if (value == min_jint) { subl(dst, value); return; }
3887 if (value < 0) { incrementl(dst, -value); return; }
3888 if (value == 0) { ; return; }
3889 if (value == 1 && UseIncDec) { decl(dst) ; return; }
3890 /* else */ { subl(dst, value) ; return; }
3891 }
3893 void MacroAssembler::incrementq(Address dst, int value) {
3894 if (value == min_jint) { addq(dst, value); return; }
3895 if (value < 0) { decrementq(dst, -value); return; }
3896 if (value == 0) { ; return; }
3897 if (value == 1 && UseIncDec) { incq(dst) ; return; }
3898 /* else */ { addq(dst, value) ; return; }
3899 }
3901 void MacroAssembler::decrementq(Address dst, int value) {
3902 if (value == min_jint) { subq(dst, value); return; }
3903 if (value < 0) { incrementq(dst, -value); return; }
3904 if (value == 0) { ; return; }
3905 if (value == 1 && UseIncDec) { decq(dst) ; return; }
3906 /* else */ { subq(dst, value) ; return; }
3907 }
3909 void MacroAssembler::align(int modulus) {
3910 if (offset() % modulus != 0) {
3911 nop(modulus - (offset() % modulus));
3912 }
3913 }
3915 void MacroAssembler::enter() {
3916 pushq(rbp);
3917 movq(rbp, rsp);
3918 }
3920 void MacroAssembler::leave() {
3921 emit_byte(0xC9); // LEAVE
3922 }
3924 // C++ bool manipulation
3926 void MacroAssembler::movbool(Register dst, Address src) {
3927 if(sizeof(bool) == 1)
3928 movb(dst, src);
3929 else if(sizeof(bool) == 2)
3930 movw(dst, src);
3931 else if(sizeof(bool) == 4)
3932 movl(dst, src);
3933 else {
3934 // unsupported
3935 ShouldNotReachHere();
3936 }
3937 }
3939 void MacroAssembler::movbool(Address dst, bool boolconst) {
3940 if(sizeof(bool) == 1)
3941 movb(dst, (int) boolconst);
3942 else if(sizeof(bool) == 2)
3943 movw(dst, (int) boolconst);
3944 else if(sizeof(bool) == 4)
3945 movl(dst, (int) boolconst);
3946 else {
3947 // unsupported
3948 ShouldNotReachHere();
3949 }
3950 }
3952 void MacroAssembler::movbool(Address dst, Register src) {
3953 if(sizeof(bool) == 1)
3954 movb(dst, src);
3955 else if(sizeof(bool) == 2)
3956 movw(dst, src);
3957 else if(sizeof(bool) == 4)
3958 movl(dst, src);
3959 else {
3960 // unsupported
3961 ShouldNotReachHere();
3962 }
3963 }
3965 void MacroAssembler::testbool(Register dst) {
3966 if(sizeof(bool) == 1)
3967 testb(dst, (int) 0xff);
3968 else if(sizeof(bool) == 2) {
3969 // need testw impl
3970 ShouldNotReachHere();
3971 } else if(sizeof(bool) == 4)
3972 testl(dst, dst);
3973 else {
3974 // unsupported
3975 ShouldNotReachHere();
3976 }
3977 }
3979 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
3980 Register last_java_fp,
3981 address last_java_pc) {
3982 // determine last_java_sp register
3983 if (!last_java_sp->is_valid()) {
3984 last_java_sp = rsp;
3985 }
3987 // last_java_fp is optional
3988 if (last_java_fp->is_valid()) {
3989 movq(Address(r15_thread, JavaThread::last_Java_fp_offset()),
3990 last_java_fp);
3991 }
3993 // last_java_pc is optional
3994 if (last_java_pc != NULL) {
3995 Address java_pc(r15_thread,
3996 JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset());
3997 lea(rscratch1, InternalAddress(last_java_pc));
3998 movq(java_pc, rscratch1);
3999 }
4001 movq(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
4002 }
4004 void MacroAssembler::reset_last_Java_frame(bool clear_fp,
4005 bool clear_pc) {
4006 // we must set sp to zero to clear frame
4007 movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
4008 // must clear fp, so that compiled frames are not confused; it is
4009 // possible that we need it only for debugging
4010 if (clear_fp) {
4011 movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
4012 }
4014 if (clear_pc) {
4015 movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
4016 }
4017 }
4020 // Implementation of call_VM versions
4022 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) {
4023 Label L, E;
4025 #ifdef _WIN64
4026 // Windows always allocates space for it's register args
4027 assert(num_args <= 4, "only register arguments supported");
4028 subq(rsp, frame::arg_reg_save_area_bytes);
4029 #endif
4031 // Align stack if necessary
4032 testl(rsp, 15);
4033 jcc(Assembler::zero, L);
4035 subq(rsp, 8);
4036 {
4037 call(RuntimeAddress(entry_point));
4038 }
4039 addq(rsp, 8);
4040 jmp(E);
4042 bind(L);
4043 {
4044 call(RuntimeAddress(entry_point));
4045 }
4047 bind(E);
4049 #ifdef _WIN64
4050 // restore stack pointer
4051 addq(rsp, frame::arg_reg_save_area_bytes);
4052 #endif
4054 }
4057 void MacroAssembler::call_VM_base(Register oop_result,
4058 Register java_thread,
4059 Register last_java_sp,
4060 address entry_point,
4061 int num_args,
4062 bool check_exceptions) {
4063 // determine last_java_sp register
4064 if (!last_java_sp->is_valid()) {
4065 last_java_sp = rsp;
4066 }
4068 // debugging support
4069 assert(num_args >= 0, "cannot have negative number of arguments");
4070 assert(r15_thread != oop_result,
4071 "cannot use the same register for java_thread & oop_result");
4072 assert(r15_thread != last_java_sp,
4073 "cannot use the same register for java_thread & last_java_sp");
4075 // set last Java frame before call
4077 // This sets last_Java_fp which is only needed from interpreted frames
4078 // and should really be done only from the interp_masm version before
4079 // calling the underlying call_VM. That doesn't happen yet so we set
4080 // last_Java_fp here even though some callers don't need it and
4081 // also clear it below.
4082 set_last_Java_frame(last_java_sp, rbp, NULL);
4084 {
4085 Label L, E;
4087 // Align stack if necessary
4088 #ifdef _WIN64
4089 assert(num_args <= 4, "only register arguments supported");
4090 // Windows always allocates space for it's register args
4091 subq(rsp, frame::arg_reg_save_area_bytes);
4092 #endif
4093 testl(rsp, 15);
4094 jcc(Assembler::zero, L);
4096 subq(rsp, 8);
4097 {
4098 call(RuntimeAddress(entry_point));
4099 }
4100 addq(rsp, 8);
4101 jmp(E);
4104 bind(L);
4105 {
4106 call(RuntimeAddress(entry_point));
4107 }
4109 bind(E);
4111 #ifdef _WIN64
4112 // restore stack pointer
4113 addq(rsp, frame::arg_reg_save_area_bytes);
4114 #endif
4115 }
4117 #ifdef ASSERT
4118 pushq(rax);
4119 {
4120 Label L;
4121 get_thread(rax);
4122 cmpq(r15_thread, rax);
4123 jcc(Assembler::equal, L);
4124 stop("MacroAssembler::call_VM_base: register not callee saved?");
4125 bind(L);
4126 }
4127 popq(rax);
4128 #endif
4130 // reset last Java frame
4131 // This really shouldn't have to clear fp set note above at the
4132 // call to set_last_Java_frame
4133 reset_last_Java_frame(true, false);
4135 check_and_handle_popframe(noreg);
4136 check_and_handle_earlyret(noreg);
4138 if (check_exceptions) {
4139 cmpq(Address(r15_thread, Thread::pending_exception_offset()), (int) NULL);
4140 // This used to conditionally jump to forward_exception however it is
4141 // possible if we relocate that the branch will not reach. So we must jump
4142 // around so we can always reach
4143 Label ok;
4144 jcc(Assembler::equal, ok);
4145 jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
4146 bind(ok);
4147 }
4149 // get oop result if there is one and reset the value in the thread
4150 if (oop_result->is_valid()) {
4151 movq(oop_result, Address(r15_thread, JavaThread::vm_result_offset()));
4152 movptr(Address(r15_thread, JavaThread::vm_result_offset()), NULL_WORD);
4153 verify_oop(oop_result, "broken oop in call_VM_base");
4154 }
4155 }
4157 void MacroAssembler::check_and_handle_popframe(Register java_thread) {}
4158 void MacroAssembler::check_and_handle_earlyret(Register java_thread) {}
4160 void MacroAssembler::call_VM_helper(Register oop_result,
4161 address entry_point,
4162 int num_args,
4163 bool check_exceptions) {
4164 // Java thread becomes first argument of C function
4165 movq(c_rarg0, r15_thread);
4167 // We've pushed one address, correct last_Java_sp
4168 leaq(rax, Address(rsp, wordSize));
4170 call_VM_base(oop_result, noreg, rax, entry_point, num_args,
4171 check_exceptions);
4172 }
4175 void MacroAssembler::call_VM(Register oop_result,
4176 address entry_point,
4177 bool check_exceptions) {
4178 Label C, E;
4179 Assembler::call(C, relocInfo::none);
4180 jmp(E);
4182 bind(C);
4183 call_VM_helper(oop_result, entry_point, 0, check_exceptions);
4184 ret(0);
4186 bind(E);
4187 }
4190 void MacroAssembler::call_VM(Register oop_result,
4191 address entry_point,
4192 Register arg_1,
4193 bool check_exceptions) {
4194 assert(rax != arg_1, "smashed argument");
4195 assert(c_rarg0 != arg_1, "smashed argument");
4197 Label C, E;
4198 Assembler::call(C, relocInfo::none);
4199 jmp(E);
4201 bind(C);
4202 // c_rarg0 is reserved for thread
4203 if (c_rarg1 != arg_1) {
4204 movq(c_rarg1, arg_1);
4205 }
4206 call_VM_helper(oop_result, entry_point, 1, check_exceptions);
4207 ret(0);
4209 bind(E);
4210 }
4212 void MacroAssembler::call_VM(Register oop_result,
4213 address entry_point,
4214 Register arg_1,
4215 Register arg_2,
4216 bool check_exceptions) {
4217 assert(rax != arg_1, "smashed argument");
4218 assert(rax != arg_2, "smashed argument");
4219 assert(c_rarg0 != arg_1, "smashed argument");
4220 assert(c_rarg0 != arg_2, "smashed argument");
4221 assert(c_rarg1 != arg_2, "smashed argument");
4222 assert(c_rarg2 != arg_1, "smashed argument");
4224 Label C, E;
4225 Assembler::call(C, relocInfo::none);
4226 jmp(E);
4228 bind(C);
4229 // c_rarg0 is reserved for thread
4230 if (c_rarg1 != arg_1) {
4231 movq(c_rarg1, arg_1);
4232 }
4233 if (c_rarg2 != arg_2) {
4234 movq(c_rarg2, arg_2);
4235 }
4236 call_VM_helper(oop_result, entry_point, 2, check_exceptions);
4237 ret(0);
4239 bind(E);
4240 }
4243 void MacroAssembler::call_VM(Register oop_result,
4244 address entry_point,
4245 Register arg_1,
4246 Register arg_2,
4247 Register arg_3,
4248 bool check_exceptions) {
4249 assert(rax != arg_1, "smashed argument");
4250 assert(rax != arg_2, "smashed argument");
4251 assert(rax != arg_3, "smashed argument");
4252 assert(c_rarg0 != arg_1, "smashed argument");
4253 assert(c_rarg0 != arg_2, "smashed argument");
4254 assert(c_rarg0 != arg_3, "smashed argument");
4255 assert(c_rarg1 != arg_2, "smashed argument");
4256 assert(c_rarg1 != arg_3, "smashed argument");
4257 assert(c_rarg2 != arg_1, "smashed argument");
4258 assert(c_rarg2 != arg_3, "smashed argument");
4259 assert(c_rarg3 != arg_1, "smashed argument");
4260 assert(c_rarg3 != arg_2, "smashed argument");
4262 Label C, E;
4263 Assembler::call(C, relocInfo::none);
4264 jmp(E);
4266 bind(C);
4267 // c_rarg0 is reserved for thread
4268 if (c_rarg1 != arg_1) {
4269 movq(c_rarg1, arg_1);
4270 }
4271 if (c_rarg2 != arg_2) {
4272 movq(c_rarg2, arg_2);
4273 }
4274 if (c_rarg3 != arg_3) {
4275 movq(c_rarg3, arg_3);
4276 }
4277 call_VM_helper(oop_result, entry_point, 3, check_exceptions);
4278 ret(0);
4280 bind(E);
4281 }
4283 void MacroAssembler::call_VM(Register oop_result,
4284 Register last_java_sp,
4285 address entry_point,
4286 int num_args,
4287 bool check_exceptions) {
4288 call_VM_base(oop_result, noreg, last_java_sp, entry_point, num_args,
4289 check_exceptions);
4290 }
4292 void MacroAssembler::call_VM(Register oop_result,
4293 Register last_java_sp,
4294 address entry_point,
4295 Register arg_1,
4296 bool check_exceptions) {
4297 assert(c_rarg0 != arg_1, "smashed argument");
4298 assert(c_rarg1 != last_java_sp, "smashed argument");
4299 // c_rarg0 is reserved for thread
4300 if (c_rarg1 != arg_1) {
4301 movq(c_rarg1, arg_1);
4302 }
4303 call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
4304 }
4306 void MacroAssembler::call_VM(Register oop_result,
4307 Register last_java_sp,
4308 address entry_point,
4309 Register arg_1,
4310 Register arg_2,
4311 bool check_exceptions) {
4312 assert(c_rarg0 != arg_1, "smashed argument");
4313 assert(c_rarg0 != arg_2, "smashed argument");
4314 assert(c_rarg1 != arg_2, "smashed argument");
4315 assert(c_rarg1 != last_java_sp, "smashed argument");
4316 assert(c_rarg2 != arg_1, "smashed argument");
4317 assert(c_rarg2 != last_java_sp, "smashed argument");
4318 // c_rarg0 is reserved for thread
4319 if (c_rarg1 != arg_1) {
4320 movq(c_rarg1, arg_1);
4321 }
4322 if (c_rarg2 != arg_2) {
4323 movq(c_rarg2, arg_2);
4324 }
4325 call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
4326 }
4329 void MacroAssembler::call_VM(Register oop_result,
4330 Register last_java_sp,
4331 address entry_point,
4332 Register arg_1,
4333 Register arg_2,
4334 Register arg_3,
4335 bool check_exceptions) {
4336 assert(c_rarg0 != arg_1, "smashed argument");
4337 assert(c_rarg0 != arg_2, "smashed argument");
4338 assert(c_rarg0 != arg_3, "smashed argument");
4339 assert(c_rarg1 != arg_2, "smashed argument");
4340 assert(c_rarg1 != arg_3, "smashed argument");
4341 assert(c_rarg1 != last_java_sp, "smashed argument");
4342 assert(c_rarg2 != arg_1, "smashed argument");
4343 assert(c_rarg2 != arg_3, "smashed argument");
4344 assert(c_rarg2 != last_java_sp, "smashed argument");
4345 assert(c_rarg3 != arg_1, "smashed argument");
4346 assert(c_rarg3 != arg_2, "smashed argument");
4347 assert(c_rarg3 != last_java_sp, "smashed argument");
4348 // c_rarg0 is reserved for thread
4349 if (c_rarg1 != arg_1) {
4350 movq(c_rarg1, arg_1);
4351 }
4352 if (c_rarg2 != arg_2) {
4353 movq(c_rarg2, arg_2);
4354 }
4355 if (c_rarg3 != arg_3) {
4356 movq(c_rarg2, arg_3);
4357 }
4358 call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
4359 }
4361 void MacroAssembler::call_VM_leaf(address entry_point, int num_args) {
4362 call_VM_leaf_base(entry_point, num_args);
4363 }
4365 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_1) {
4366 if (c_rarg0 != arg_1) {
4367 movq(c_rarg0, arg_1);
4368 }
4369 call_VM_leaf(entry_point, 1);
4370 }
4372 void MacroAssembler::call_VM_leaf(address entry_point,
4373 Register arg_1,
4374 Register arg_2) {
4375 assert(c_rarg0 != arg_2, "smashed argument");
4376 assert(c_rarg1 != arg_1, "smashed argument");
4377 if (c_rarg0 != arg_1) {
4378 movq(c_rarg0, arg_1);
4379 }
4380 if (c_rarg1 != arg_2) {
4381 movq(c_rarg1, arg_2);
4382 }
4383 call_VM_leaf(entry_point, 2);
4384 }
4386 void MacroAssembler::call_VM_leaf(address entry_point,
4387 Register arg_1,
4388 Register arg_2,
4389 Register arg_3) {
4390 assert(c_rarg0 != arg_2, "smashed argument");
4391 assert(c_rarg0 != arg_3, "smashed argument");
4392 assert(c_rarg1 != arg_1, "smashed argument");
4393 assert(c_rarg1 != arg_3, "smashed argument");
4394 assert(c_rarg2 != arg_1, "smashed argument");
4395 assert(c_rarg2 != arg_2, "smashed argument");
4396 if (c_rarg0 != arg_1) {
4397 movq(c_rarg0, arg_1);
4398 }
4399 if (c_rarg1 != arg_2) {
4400 movq(c_rarg1, arg_2);
4401 }
4402 if (c_rarg2 != arg_3) {
4403 movq(c_rarg2, arg_3);
4404 }
4405 call_VM_leaf(entry_point, 3);
4406 }
4409 // Calls to C land
4410 //
4411 // When entering C land, the rbp & rsp of the last Java frame have to
4412 // be recorded in the (thread-local) JavaThread object. When leaving C
4413 // land, the last Java fp has to be reset to 0. This is required to
4414 // allow proper stack traversal.
4415 void MacroAssembler::store_check(Register obj) {
4416 // Does a store check for the oop in register obj. The content of
4417 // register obj is destroyed afterwards.
4418 store_check_part_1(obj);
4419 store_check_part_2(obj);
4420 }
4422 void MacroAssembler::store_check(Register obj, Address dst) {
4423 store_check(obj);
4424 }
4426 // split the store check operation so that other instructions can be
4427 // scheduled inbetween
4428 void MacroAssembler::store_check_part_1(Register obj) {
4429 BarrierSet* bs = Universe::heap()->barrier_set();
4430 assert(bs->kind() == BarrierSet::CardTableModRef, "Wrong barrier set kind");
4431 shrq(obj, CardTableModRefBS::card_shift);
4432 }
4434 void MacroAssembler::store_check_part_2(Register obj) {
4435 BarrierSet* bs = Universe::heap()->barrier_set();
4436 assert(bs->kind() == BarrierSet::CardTableModRef, "Wrong barrier set kind");
4437 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
4438 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
4440 // The calculation for byte_map_base is as follows:
4441 // byte_map_base = _byte_map - (uintptr_t(low_bound) >> card_shift);
4442 // So this essentially converts an address to a displacement and
4443 // it will never need to be relocated. On 64bit however the value may be too
4444 // large for a 32bit displacement
4446 intptr_t disp = (intptr_t) ct->byte_map_base;
4447 if (is_simm32(disp)) {
4448 Address cardtable(noreg, obj, Address::times_1, disp);
4449 movb(cardtable, 0);
4450 } else {
4451 // By doing it as an ExternalAddress disp could be converted to a rip-relative
4452 // displacement and done in a single instruction given favorable mapping and
4453 // a smarter version of as_Address. Worst case it is two instructions which
4454 // is no worse off then loading disp into a register and doing as a simple
4455 // Address() as above.
4456 // We can't do as ExternalAddress as the only style since if disp == 0 we'll
4457 // assert since NULL isn't acceptable in a reloci (see 6644928). In any case
4458 // in some cases we'll get a single instruction version.
4460 ExternalAddress cardtable((address)disp);
4461 Address index(noreg, obj, Address::times_1);
4462 movb(as_Address(ArrayAddress(cardtable, index)), 0);
4463 }
4465 }
4467 void MacroAssembler::c2bool(Register x) {
4468 // implements x == 0 ? 0 : 1
4469 // note: must only look at least-significant byte of x
4470 // since C-style booleans are stored in one byte
4471 // only! (was bug)
4472 andl(x, 0xFF);
4473 setb(Assembler::notZero, x);
4474 }
4476 int MacroAssembler::corrected_idivl(Register reg) {
4477 // Full implementation of Java idiv and irem; checks for special
4478 // case as described in JVM spec., p.243 & p.271. The function
4479 // returns the (pc) offset of the idivl instruction - may be needed
4480 // for implicit exceptions.
4481 //
4482 // normal case special case
4483 //
4484 // input : eax: dividend min_int
4485 // reg: divisor (may not be eax/edx) -1
4486 //
4487 // output: eax: quotient (= eax idiv reg) min_int
4488 // edx: remainder (= eax irem reg) 0
4489 assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register");
4490 const int min_int = 0x80000000;
4491 Label normal_case, special_case;
4493 // check for special case
4494 cmpl(rax, min_int);
4495 jcc(Assembler::notEqual, normal_case);
4496 xorl(rdx, rdx); // prepare edx for possible special case (where
4497 // remainder = 0)
4498 cmpl(reg, -1);
4499 jcc(Assembler::equal, special_case);
4501 // handle normal case
4502 bind(normal_case);
4503 cdql();
4504 int idivl_offset = offset();
4505 idivl(reg);
4507 // normal and special case exit
4508 bind(special_case);
4510 return idivl_offset;
4511 }
4513 int MacroAssembler::corrected_idivq(Register reg) {
4514 // Full implementation of Java ldiv and lrem; checks for special
4515 // case as described in JVM spec., p.243 & p.271. The function
4516 // returns the (pc) offset of the idivl instruction - may be needed
4517 // for implicit exceptions.
4518 //
4519 // normal case special case
4520 //
4521 // input : rax: dividend min_long
4522 // reg: divisor (may not be eax/edx) -1
4523 //
4524 // output: rax: quotient (= rax idiv reg) min_long
4525 // rdx: remainder (= rax irem reg) 0
4526 assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register");
4527 static const int64_t min_long = 0x8000000000000000;
4528 Label normal_case, special_case;
4530 // check for special case
4531 cmp64(rax, ExternalAddress((address) &min_long));
4532 jcc(Assembler::notEqual, normal_case);
4533 xorl(rdx, rdx); // prepare rdx for possible special case (where
4534 // remainder = 0)
4535 cmpq(reg, -1);
4536 jcc(Assembler::equal, special_case);
4538 // handle normal case
4539 bind(normal_case);
4540 cdqq();
4541 int idivq_offset = offset();
4542 idivq(reg);
4544 // normal and special case exit
4545 bind(special_case);
4547 return idivq_offset;
4548 }
4550 void MacroAssembler::push_IU_state() {
4551 pushfq(); // Push flags first because pushaq kills them
4552 subq(rsp, 8); // Make sure rsp stays 16-byte aligned
4553 pushaq();
4554 }
4556 void MacroAssembler::pop_IU_state() {
4557 popaq();
4558 addq(rsp, 8);
4559 popfq();
4560 }
4562 void MacroAssembler::push_FPU_state() {
4563 subq(rsp, FPUStateSizeInWords * wordSize);
4564 fxsave(Address(rsp, 0));
4565 }
4567 void MacroAssembler::pop_FPU_state() {
4568 fxrstor(Address(rsp, 0));
4569 addq(rsp, FPUStateSizeInWords * wordSize);
4570 }
4572 // Save Integer and Float state
4573 // Warning: Stack must be 16 byte aligned
4574 void MacroAssembler::push_CPU_state() {
4575 push_IU_state();
4576 push_FPU_state();
4577 }
4579 void MacroAssembler::pop_CPU_state() {
4580 pop_FPU_state();
4581 pop_IU_state();
4582 }
4584 void MacroAssembler::sign_extend_short(Register reg) {
4585 movswl(reg, reg);
4586 }
4588 void MacroAssembler::sign_extend_byte(Register reg) {
4589 movsbl(reg, reg);
4590 }
4592 void MacroAssembler::division_with_shift(Register reg, int shift_value) {
4593 assert (shift_value > 0, "illegal shift value");
4594 Label _is_positive;
4595 testl (reg, reg);
4596 jcc (Assembler::positive, _is_positive);
4597 int offset = (1 << shift_value) - 1 ;
4599 if (offset == 1) {
4600 incrementl(reg);
4601 } else {
4602 addl(reg, offset);
4603 }
4605 bind (_is_positive);
4606 sarl(reg, shift_value);
4607 }
4609 void MacroAssembler::round_to_l(Register reg, int modulus) {
4610 addl(reg, modulus - 1);
4611 andl(reg, -modulus);
4612 }
4614 void MacroAssembler::round_to_q(Register reg, int modulus) {
4615 addq(reg, modulus - 1);
4616 andq(reg, -modulus);
4617 }
4619 void MacroAssembler::verify_oop(Register reg, const char* s) {
4620 if (!VerifyOops) {
4621 return;
4622 }
4624 // Pass register number to verify_oop_subroutine
4625 char* b = new char[strlen(s) + 50];
4626 sprintf(b, "verify_oop: %s: %s", reg->name(), s);
4628 pushq(rax); // save rax, restored by receiver
4630 // pass args on stack, only touch rax
4631 pushq(reg);
4632 // avoid using pushptr, as it modifies scratch registers
4633 // and our contract is not to modify anything
4634 ExternalAddress buffer((address)b);
4635 movptr(rax, buffer.addr());
4636 pushq(rax);
4638 // call indirectly to solve generation ordering problem
4639 movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
4640 call(rax); // no alignment requirement
4641 // everything popped by receiver
4642 }
4644 void MacroAssembler::verify_oop_addr(Address addr, const char* s) {
4645 if (!VerifyOops) return;
4646 // Pass register number to verify_oop_subroutine
4647 char* b = new char[strlen(s) + 50];
4648 sprintf(b, "verify_oop_addr: %s", s);
4649 pushq(rax); // save rax
4650 movq(addr, rax);
4651 pushq(rax); // pass register argument
4654 // avoid using pushptr, as it modifies scratch registers
4655 // and our contract is not to modify anything
4656 ExternalAddress buffer((address)b);
4657 movptr(rax, buffer.addr());
4658 pushq(rax);
4660 // call indirectly to solve generation ordering problem
4661 movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
4662 call(rax); // no alignment requirement
4663 // everything popped by receiver
4664 }
4667 void MacroAssembler::stop(const char* msg) {
4668 address rip = pc();
4669 pushaq(); // get regs on stack
4670 lea(c_rarg0, ExternalAddress((address) msg));
4671 lea(c_rarg1, InternalAddress(rip));
4672 movq(c_rarg2, rsp); // pass pointer to regs array
4673 andq(rsp, -16); // align stack as required by ABI
4674 call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug)));
4675 hlt();
4676 }
4678 void MacroAssembler::warn(const char* msg) {
4679 pushq(r12);
4680 movq(r12, rsp);
4681 andq(rsp, -16); // align stack as required by push_CPU_state and call
4683 push_CPU_state(); // keeps alignment at 16 bytes
4684 lea(c_rarg0, ExternalAddress((address) msg));
4685 call_VM_leaf(CAST_FROM_FN_PTR(address, warning), c_rarg0);
4686 pop_CPU_state();
4688 movq(rsp, r12);
4689 popq(r12);
4690 }
4692 #ifndef PRODUCT
4693 extern "C" void findpc(intptr_t x);
4694 #endif
4696 void MacroAssembler::debug(char* msg, int64_t pc, int64_t regs[]) {
4697 // In order to get locks to work, we need to fake a in_VM state
4698 if (ShowMessageBoxOnError ) {
4699 JavaThread* thread = JavaThread::current();
4700 JavaThreadState saved_state = thread->thread_state();
4701 thread->set_thread_state(_thread_in_vm);
4702 #ifndef PRODUCT
4703 if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
4704 ttyLocker ttyl;
4705 BytecodeCounter::print();
4706 }
4707 #endif
4708 // To see where a verify_oop failed, get $ebx+40/X for this frame.
4709 // XXX correct this offset for amd64
4710 // This is the value of eip which points to where verify_oop will return.
4711 if (os::message_box(msg, "Execution stopped, print registers?")) {
4712 ttyLocker ttyl;
4713 tty->print_cr("rip = 0x%016lx", pc);
4714 #ifndef PRODUCT
4715 tty->cr();
4716 findpc(pc);
4717 tty->cr();
4718 #endif
4719 tty->print_cr("rax = 0x%016lx", regs[15]);
4720 tty->print_cr("rbx = 0x%016lx", regs[12]);
4721 tty->print_cr("rcx = 0x%016lx", regs[14]);
4722 tty->print_cr("rdx = 0x%016lx", regs[13]);
4723 tty->print_cr("rdi = 0x%016lx", regs[8]);
4724 tty->print_cr("rsi = 0x%016lx", regs[9]);
4725 tty->print_cr("rbp = 0x%016lx", regs[10]);
4726 tty->print_cr("rsp = 0x%016lx", regs[11]);
4727 tty->print_cr("r8 = 0x%016lx", regs[7]);
4728 tty->print_cr("r9 = 0x%016lx", regs[6]);
4729 tty->print_cr("r10 = 0x%016lx", regs[5]);
4730 tty->print_cr("r11 = 0x%016lx", regs[4]);
4731 tty->print_cr("r12 = 0x%016lx", regs[3]);
4732 tty->print_cr("r13 = 0x%016lx", regs[2]);
4733 tty->print_cr("r14 = 0x%016lx", regs[1]);
4734 tty->print_cr("r15 = 0x%016lx", regs[0]);
4735 BREAKPOINT;
4736 }
4737 ThreadStateTransition::transition(thread, _thread_in_vm, saved_state);
4738 } else {
4739 ttyLocker ttyl;
4740 ::tty->print_cr("=============== DEBUG MESSAGE: %s ================\n",
4741 msg);
4742 }
4743 }
4745 void MacroAssembler::os_breakpoint() {
4746 // instead of directly emitting a breakpoint, call os:breakpoint for
4747 // better debugability
4748 // This shouldn't need alignment, it's an empty function
4749 call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
4750 }
4752 // Write serialization page so VM thread can do a pseudo remote membar.
4753 // We use the current thread pointer to calculate a thread specific
4754 // offset to write to within the page. This minimizes bus traffic
4755 // due to cache line collision.
4756 void MacroAssembler::serialize_memory(Register thread,
4757 Register tmp) {
4759 movl(tmp, thread);
4760 shrl(tmp, os::get_serialize_page_shift_count());
4761 andl(tmp, (os::vm_page_size() - sizeof(int)));
4763 Address index(noreg, tmp, Address::times_1);
4764 ExternalAddress page(os::get_memory_serialize_page());
4766 movptr(ArrayAddress(page, index), tmp);
4767 }
4769 void MacroAssembler::verify_tlab() {
4770 #ifdef ASSERT
4771 if (UseTLAB) {
4772 Label next, ok;
4773 Register t1 = rsi;
4775 pushq(t1);
4777 movq(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
4778 cmpq(t1, Address(r15_thread, in_bytes(JavaThread::tlab_start_offset())));
4779 jcc(Assembler::aboveEqual, next);
4780 stop("assert(top >= start)");
4781 should_not_reach_here();
4783 bind(next);
4784 movq(t1, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
4785 cmpq(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
4786 jcc(Assembler::aboveEqual, ok);
4787 stop("assert(top <= end)");
4788 should_not_reach_here();
4790 bind(ok);
4792 popq(t1);
4793 }
4794 #endif
4795 }
4797 // Defines obj, preserves var_size_in_bytes
4798 void MacroAssembler::eden_allocate(Register obj,
4799 Register var_size_in_bytes,
4800 int con_size_in_bytes,
4801 Register t1,
4802 Label& slow_case) {
4803 assert(obj == rax, "obj must be in rax for cmpxchg");
4804 assert_different_registers(obj, var_size_in_bytes, t1);
4805 Register end = t1;
4806 Label retry;
4807 bind(retry);
4808 ExternalAddress heap_top((address) Universe::heap()->top_addr());
4809 movptr(obj, heap_top);
4810 if (var_size_in_bytes == noreg) {
4811 leaq(end, Address(obj, con_size_in_bytes));
4812 } else {
4813 leaq(end, Address(obj, var_size_in_bytes, Address::times_1));
4814 }
4815 // if end < obj then we wrapped around => object too long => slow case
4816 cmpq(end, obj);
4817 jcc(Assembler::below, slow_case);
4818 cmpptr(end, ExternalAddress((address) Universe::heap()->end_addr()));
4820 jcc(Assembler::above, slow_case);
4821 // Compare obj with the top addr, and if still equal, store the new
4822 // top addr in end at the address of the top addr pointer. Sets ZF
4823 // if was equal, and clears it otherwise. Use lock prefix for
4824 // atomicity on MPs.
4825 if (os::is_MP()) {
4826 lock();
4827 }
4828 cmpxchgptr(end, heap_top);
4829 // if someone beat us on the allocation, try again, otherwise continue
4830 jcc(Assembler::notEqual, retry);
4831 }
4833 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
4834 void MacroAssembler::tlab_allocate(Register obj,
4835 Register var_size_in_bytes,
4836 int con_size_in_bytes,
4837 Register t1,
4838 Register t2,
4839 Label& slow_case) {
4840 assert_different_registers(obj, t1, t2);
4841 assert_different_registers(obj, var_size_in_bytes, t1);
4842 Register end = t2;
4844 verify_tlab();
4846 movq(obj, Address(r15_thread, JavaThread::tlab_top_offset()));
4847 if (var_size_in_bytes == noreg) {
4848 leaq(end, Address(obj, con_size_in_bytes));
4849 } else {
4850 leaq(end, Address(obj, var_size_in_bytes, Address::times_1));
4851 }
4852 cmpq(end, Address(r15_thread, JavaThread::tlab_end_offset()));
4853 jcc(Assembler::above, slow_case);
4855 // update the tlab top pointer
4856 movq(Address(r15_thread, JavaThread::tlab_top_offset()), end);
4858 // recover var_size_in_bytes if necessary
4859 if (var_size_in_bytes == end) {
4860 subq(var_size_in_bytes, obj);
4861 }
4862 verify_tlab();
4863 }
4865 // Preserves rbx and rdx.
4866 void MacroAssembler::tlab_refill(Label& retry,
4867 Label& try_eden,
4868 Label& slow_case) {
4869 Register top = rax;
4870 Register t1 = rcx;
4871 Register t2 = rsi;
4872 Register t3 = r10;
4873 Register thread_reg = r15_thread;
4874 assert_different_registers(top, thread_reg, t1, t2, t3,
4875 /* preserve: */ rbx, rdx);
4876 Label do_refill, discard_tlab;
4878 if (CMSIncrementalMode || !Universe::heap()->supports_inline_contig_alloc()) {
4879 // No allocation in the shared eden.
4880 jmp(slow_case);
4881 }
4883 movq(top, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())));
4884 movq(t1, Address(thread_reg, in_bytes(JavaThread::tlab_end_offset())));
4886 // calculate amount of free space
4887 subq(t1, top);
4888 shrq(t1, LogHeapWordSize);
4890 // Retain tlab and allocate object in shared space if
4891 // the amount free in the tlab is too large to discard.
4892 cmpq(t1, Address(thread_reg, // size_t
4893 in_bytes(JavaThread::tlab_refill_waste_limit_offset())));
4894 jcc(Assembler::lessEqual, discard_tlab);
4896 // Retain
4897 mov64(t2, ThreadLocalAllocBuffer::refill_waste_limit_increment());
4898 addq(Address(thread_reg, // size_t
4899 in_bytes(JavaThread::tlab_refill_waste_limit_offset())),
4900 t2);
4901 if (TLABStats) {
4902 // increment number of slow_allocations
4903 addl(Address(thread_reg, // unsigned int
4904 in_bytes(JavaThread::tlab_slow_allocations_offset())),
4905 1);
4906 }
4907 jmp(try_eden);
4909 bind(discard_tlab);
4910 if (TLABStats) {
4911 // increment number of refills
4912 addl(Address(thread_reg, // unsigned int
4913 in_bytes(JavaThread::tlab_number_of_refills_offset())),
4914 1);
4915 // accumulate wastage -- t1 is amount free in tlab
4916 addl(Address(thread_reg, // unsigned int
4917 in_bytes(JavaThread::tlab_fast_refill_waste_offset())),
4918 t1);
4919 }
4921 // if tlab is currently allocated (top or end != null) then
4922 // fill [top, end + alignment_reserve) with array object
4923 testq(top, top);
4924 jcc(Assembler::zero, do_refill);
4926 // set up the mark word
4927 mov64(t3, (int64_t) markOopDesc::prototype()->copy_set_hash(0x2));
4928 movq(Address(top, oopDesc::mark_offset_in_bytes()), t3);
4929 // set the length to the remaining space
4930 subq(t1, typeArrayOopDesc::header_size(T_INT));
4931 addq(t1, (int)ThreadLocalAllocBuffer::alignment_reserve());
4932 shlq(t1, log2_intptr(HeapWordSize / sizeof(jint)));
4933 movq(Address(top, arrayOopDesc::length_offset_in_bytes()), t1);
4934 // set klass to intArrayKlass
4935 movptr(t1, ExternalAddress((address) Universe::intArrayKlassObj_addr()));
4936 store_klass(top, t1);
4938 // refill the tlab with an eden allocation
4939 bind(do_refill);
4940 movq(t1, Address(thread_reg, in_bytes(JavaThread::tlab_size_offset())));
4941 shlq(t1, LogHeapWordSize);
4942 // add object_size ??
4943 eden_allocate(top, t1, 0, t2, slow_case);
4945 // Check that t1 was preserved in eden_allocate.
4946 #ifdef ASSERT
4947 if (UseTLAB) {
4948 Label ok;
4949 Register tsize = rsi;
4950 assert_different_registers(tsize, thread_reg, t1);
4951 pushq(tsize);
4952 movq(tsize, Address(thread_reg, in_bytes(JavaThread::tlab_size_offset())));
4953 shlq(tsize, LogHeapWordSize);
4954 cmpq(t1, tsize);
4955 jcc(Assembler::equal, ok);
4956 stop("assert(t1 != tlab size)");
4957 should_not_reach_here();
4959 bind(ok);
4960 popq(tsize);
4961 }
4962 #endif
4963 movq(Address(thread_reg, in_bytes(JavaThread::tlab_start_offset())), top);
4964 movq(Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())), top);
4965 addq(top, t1);
4966 subq(top, (int)ThreadLocalAllocBuffer::alignment_reserve_in_bytes());
4967 movq(Address(thread_reg, in_bytes(JavaThread::tlab_end_offset())), top);
4968 verify_tlab();
4969 jmp(retry);
4970 }
4973 int MacroAssembler::biased_locking_enter(Register lock_reg, Register obj_reg, Register swap_reg, Register tmp_reg,
4974 bool swap_reg_contains_mark,
4975 Label& done, Label* slow_case,
4976 BiasedLockingCounters* counters) {
4977 assert(UseBiasedLocking, "why call this otherwise?");
4978 assert(swap_reg == rax, "swap_reg must be rax for cmpxchgq");
4979 assert(tmp_reg != noreg, "tmp_reg must be supplied");
4980 assert_different_registers(lock_reg, obj_reg, swap_reg, tmp_reg);
4981 assert(markOopDesc::age_shift == markOopDesc::lock_bits + markOopDesc::biased_lock_bits, "biased locking makes assumptions about bit layout");
4982 Address mark_addr (obj_reg, oopDesc::mark_offset_in_bytes());
4983 Address saved_mark_addr(lock_reg, 0);
4985 if (PrintBiasedLockingStatistics && counters == NULL)
4986 counters = BiasedLocking::counters();
4988 // Biased locking
4989 // See whether the lock is currently biased toward our thread and
4990 // whether the epoch is still valid
4991 // Note that the runtime guarantees sufficient alignment of JavaThread
4992 // pointers to allow age to be placed into low bits
4993 // First check to see whether biasing is even enabled for this object
4994 Label cas_label;
4995 int null_check_offset = -1;
4996 if (!swap_reg_contains_mark) {
4997 null_check_offset = offset();
4998 movq(swap_reg, mark_addr);
4999 }
5000 movq(tmp_reg, swap_reg);
5001 andq(tmp_reg, markOopDesc::biased_lock_mask_in_place);
5002 cmpq(tmp_reg, markOopDesc::biased_lock_pattern);
5003 jcc(Assembler::notEqual, cas_label);
5004 // The bias pattern is present in the object's header. Need to check
5005 // whether the bias owner and the epoch are both still current.
5006 load_klass(tmp_reg, obj_reg);
5007 movq(tmp_reg, Address(tmp_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
5008 orq(tmp_reg, r15_thread);
5009 xorq(tmp_reg, swap_reg);
5010 andq(tmp_reg, ~((int) markOopDesc::age_mask_in_place));
5011 if (counters != NULL) {
5012 cond_inc32(Assembler::zero,
5013 ExternalAddress((address) counters->anonymously_biased_lock_entry_count_addr()));
5014 }
5015 jcc(Assembler::equal, done);
5017 Label try_revoke_bias;
5018 Label try_rebias;
5020 // At this point we know that the header has the bias pattern and
5021 // that we are not the bias owner in the current epoch. We need to
5022 // figure out more details about the state of the header in order to
5023 // know what operations can be legally performed on the object's
5024 // header.
5026 // If the low three bits in the xor result aren't clear, that means
5027 // the prototype header is no longer biased and we have to revoke
5028 // the bias on this object.
5029 testq(tmp_reg, markOopDesc::biased_lock_mask_in_place);
5030 jcc(Assembler::notZero, try_revoke_bias);
5032 // Biasing is still enabled for this data type. See whether the
5033 // epoch of the current bias is still valid, meaning that the epoch
5034 // bits of the mark word are equal to the epoch bits of the
5035 // prototype header. (Note that the prototype header's epoch bits
5036 // only change at a safepoint.) If not, attempt to rebias the object
5037 // toward the current thread. Note that we must be absolutely sure
5038 // that the current epoch is invalid in order to do this because
5039 // otherwise the manipulations it performs on the mark word are
5040 // illegal.
5041 testq(tmp_reg, markOopDesc::epoch_mask_in_place);
5042 jcc(Assembler::notZero, try_rebias);
5044 // The epoch of the current bias is still valid but we know nothing
5045 // about the owner; it might be set or it might be clear. Try to
5046 // acquire the bias of the object using an atomic operation. If this
5047 // fails we will go in to the runtime to revoke the object's bias.
5048 // Note that we first construct the presumed unbiased header so we
5049 // don't accidentally blow away another thread's valid bias.
5050 andq(swap_reg,
5051 markOopDesc::biased_lock_mask_in_place | markOopDesc::age_mask_in_place | markOopDesc::epoch_mask_in_place);
5052 movq(tmp_reg, swap_reg);
5053 orq(tmp_reg, r15_thread);
5054 if (os::is_MP()) {
5055 lock();
5056 }
5057 cmpxchgq(tmp_reg, Address(obj_reg, 0));
5058 // If the biasing toward our thread failed, this means that
5059 // another thread succeeded in biasing it toward itself and we
5060 // need to revoke that bias. The revocation will occur in the
5061 // interpreter runtime in the slow case.
5062 if (counters != NULL) {
5063 cond_inc32(Assembler::zero,
5064 ExternalAddress((address) counters->anonymously_biased_lock_entry_count_addr()));
5065 }
5066 if (slow_case != NULL) {
5067 jcc(Assembler::notZero, *slow_case);
5068 }
5069 jmp(done);
5071 bind(try_rebias);
5072 // At this point we know the epoch has expired, meaning that the
5073 // current "bias owner", if any, is actually invalid. Under these
5074 // circumstances _only_, we are allowed to use the current header's
5075 // value as the comparison value when doing the cas to acquire the
5076 // bias in the current epoch. In other words, we allow transfer of
5077 // the bias from one thread to another directly in this situation.
5078 //
5079 // FIXME: due to a lack of registers we currently blow away the age
5080 // bits in this situation. Should attempt to preserve them.
5081 load_klass(tmp_reg, obj_reg);
5082 movq(tmp_reg, Address(tmp_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
5083 orq(tmp_reg, r15_thread);
5084 if (os::is_MP()) {
5085 lock();
5086 }
5087 cmpxchgq(tmp_reg, Address(obj_reg, 0));
5088 // If the biasing toward our thread failed, then another thread
5089 // succeeded in biasing it toward itself and we need to revoke that
5090 // bias. The revocation will occur in the runtime in the slow case.
5091 if (counters != NULL) {
5092 cond_inc32(Assembler::zero,
5093 ExternalAddress((address) counters->rebiased_lock_entry_count_addr()));
5094 }
5095 if (slow_case != NULL) {
5096 jcc(Assembler::notZero, *slow_case);
5097 }
5098 jmp(done);
5100 bind(try_revoke_bias);
5101 // The prototype mark in the klass doesn't have the bias bit set any
5102 // more, indicating that objects of this data type are not supposed
5103 // to be biased any more. We are going to try to reset the mark of
5104 // this object to the prototype value and fall through to the
5105 // CAS-based locking scheme. Note that if our CAS fails, it means
5106 // that another thread raced us for the privilege of revoking the
5107 // bias of this particular object, so it's okay to continue in the
5108 // normal locking code.
5109 //
5110 // FIXME: due to a lack of registers we currently blow away the age
5111 // bits in this situation. Should attempt to preserve them.
5112 load_klass(tmp_reg, obj_reg);
5113 movq(tmp_reg, Address(tmp_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
5114 if (os::is_MP()) {
5115 lock();
5116 }
5117 cmpxchgq(tmp_reg, Address(obj_reg, 0));
5118 // Fall through to the normal CAS-based lock, because no matter what
5119 // the result of the above CAS, some thread must have succeeded in
5120 // removing the bias bit from the object's header.
5121 if (counters != NULL) {
5122 cond_inc32(Assembler::zero,
5123 ExternalAddress((address) counters->revoked_lock_entry_count_addr()));
5124 }
5126 bind(cas_label);
5128 return null_check_offset;
5129 }
5132 void MacroAssembler::biased_locking_exit(Register obj_reg, Register temp_reg, Label& done) {
5133 assert(UseBiasedLocking, "why call this otherwise?");
5135 // Check for biased locking unlock case, which is a no-op
5136 // Note: we do not have to check the thread ID for two reasons.
5137 // First, the interpreter checks for IllegalMonitorStateException at
5138 // a higher level. Second, if the bias was revoked while we held the
5139 // lock, the object could not be rebiased toward another thread, so
5140 // the bias bit would be clear.
5141 movq(temp_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
5142 andq(temp_reg, markOopDesc::biased_lock_mask_in_place);
5143 cmpq(temp_reg, markOopDesc::biased_lock_pattern);
5144 jcc(Assembler::equal, done);
5145 }
5148 void MacroAssembler::load_klass(Register dst, Register src) {
5149 if (UseCompressedOops) {
5150 movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5151 decode_heap_oop_not_null(dst);
5152 } else {
5153 movq(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5154 }
5155 }
5157 void MacroAssembler::store_klass(Register dst, Register src) {
5158 if (UseCompressedOops) {
5159 encode_heap_oop_not_null(src);
5160 // zero the entire klass field first as the gap needs to be zeroed too.
5161 movptr(Address(dst, oopDesc::klass_offset_in_bytes()), NULL_WORD);
5162 movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
5163 } else {
5164 movq(Address(dst, oopDesc::klass_offset_in_bytes()), src);
5165 }
5166 }
5168 void MacroAssembler::load_heap_oop(Register dst, Address src) {
5169 if (UseCompressedOops) {
5170 movl(dst, src);
5171 decode_heap_oop(dst);
5172 } else {
5173 movq(dst, src);
5174 }
5175 }
5177 void MacroAssembler::store_heap_oop(Address dst, Register src) {
5178 if (UseCompressedOops) {
5179 assert(!dst.uses(src), "not enough registers");
5180 encode_heap_oop(src);
5181 movl(dst, src);
5182 } else {
5183 movq(dst, src);
5184 }
5185 }
5187 // Algorithm must match oop.inline.hpp encode_heap_oop.
5188 void MacroAssembler::encode_heap_oop(Register r) {
5189 assert (UseCompressedOops, "should be compressed");
5190 #ifdef ASSERT
5191 Label ok;
5192 pushq(rscratch1); // cmpptr trashes rscratch1
5193 cmpptr(r12_heapbase, ExternalAddress((address)Universe::heap_base_addr()));
5194 jcc(Assembler::equal, ok);
5195 stop("MacroAssembler::encode_heap_oop: heap base corrupted?");
5196 bind(ok);
5197 popq(rscratch1);
5198 #endif
5199 verify_oop(r, "broken oop in encode_heap_oop");
5200 testq(r, r);
5201 cmovq(Assembler::equal, r, r12_heapbase);
5202 subq(r, r12_heapbase);
5203 shrq(r, LogMinObjAlignmentInBytes);
5204 }
5206 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5207 assert (UseCompressedOops, "should be compressed");
5208 #ifdef ASSERT
5209 Label ok;
5210 testq(r, r);
5211 jcc(Assembler::notEqual, ok);
5212 stop("null oop passed to encode_heap_oop_not_null");
5213 bind(ok);
5214 #endif
5215 verify_oop(r, "broken oop in encode_heap_oop_not_null");
5216 subq(r, r12_heapbase);
5217 shrq(r, LogMinObjAlignmentInBytes);
5218 }
5220 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5221 assert (UseCompressedOops, "should be compressed");
5222 #ifdef ASSERT
5223 Label ok;
5224 testq(src, src);
5225 jcc(Assembler::notEqual, ok);
5226 stop("null oop passed to encode_heap_oop_not_null2");
5227 bind(ok);
5228 #endif
5229 verify_oop(src, "broken oop in encode_heap_oop_not_null2");
5230 if (dst != src) {
5231 movq(dst, src);
5232 }
5233 subq(dst, r12_heapbase);
5234 shrq(dst, LogMinObjAlignmentInBytes);
5235 }
5237 void MacroAssembler::decode_heap_oop(Register r) {
5238 assert (UseCompressedOops, "should be compressed");
5239 #ifdef ASSERT
5240 Label ok;
5241 pushq(rscratch1);
5242 cmpptr(r12_heapbase,
5243 ExternalAddress((address)Universe::heap_base_addr()));
5244 jcc(Assembler::equal, ok);
5245 stop("MacroAssembler::decode_heap_oop: heap base corrupted?");
5246 bind(ok);
5247 popq(rscratch1);
5248 #endif
5250 Label done;
5251 shlq(r, LogMinObjAlignmentInBytes);
5252 jccb(Assembler::equal, done);
5253 addq(r, r12_heapbase);
5254 #if 0
5255 // alternate decoding probably a wash.
5256 testq(r, r);
5257 jccb(Assembler::equal, done);
5258 leaq(r, Address(r12_heapbase, r, Address::times_8, 0));
5259 #endif
5260 bind(done);
5261 verify_oop(r, "broken oop in decode_heap_oop");
5262 }
5264 void MacroAssembler::decode_heap_oop_not_null(Register r) {
5265 assert (UseCompressedOops, "should only be used for compressed headers");
5266 // Cannot assert, unverified entry point counts instructions (see .ad file)
5267 // vtableStubs also counts instructions in pd_code_size_limit.
5268 assert(Address::times_8 == LogMinObjAlignmentInBytes, "decode alg wrong");
5269 leaq(r, Address(r12_heapbase, r, Address::times_8, 0));
5270 }
5272 void MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5273 assert (UseCompressedOops, "should only be used for compressed headers");
5274 // Cannot assert, unverified entry point counts instructions (see .ad file)
5275 // vtableStubs also counts instructions in pd_code_size_limit.
5276 assert(Address::times_8 == LogMinObjAlignmentInBytes, "decode alg wrong");
5277 leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
5278 }
5280 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
5281 switch (cond) {
5282 // Note some conditions are synonyms for others
5283 case Assembler::zero: return Assembler::notZero;
5284 case Assembler::notZero: return Assembler::zero;
5285 case Assembler::less: return Assembler::greaterEqual;
5286 case Assembler::lessEqual: return Assembler::greater;
5287 case Assembler::greater: return Assembler::lessEqual;
5288 case Assembler::greaterEqual: return Assembler::less;
5289 case Assembler::below: return Assembler::aboveEqual;
5290 case Assembler::belowEqual: return Assembler::above;
5291 case Assembler::above: return Assembler::belowEqual;
5292 case Assembler::aboveEqual: return Assembler::below;
5293 case Assembler::overflow: return Assembler::noOverflow;
5294 case Assembler::noOverflow: return Assembler::overflow;
5295 case Assembler::negative: return Assembler::positive;
5296 case Assembler::positive: return Assembler::negative;
5297 case Assembler::parity: return Assembler::noParity;
5298 case Assembler::noParity: return Assembler::parity;
5299 }
5300 ShouldNotReachHere(); return Assembler::overflow;
5301 }
5304 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr) {
5305 Condition negated_cond = negate_condition(cond);
5306 Label L;
5307 jcc(negated_cond, L);
5308 atomic_incl(counter_addr);
5309 bind(L);
5310 }
5312 void MacroAssembler::atomic_incl(AddressLiteral counter_addr) {
5313 pushfq();
5314 if (os::is_MP())
5315 lock();
5316 incrementl(counter_addr);
5317 popfq();
5318 }
5320 SkipIfEqual::SkipIfEqual(
5321 MacroAssembler* masm, const bool* flag_addr, bool value) {
5322 _masm = masm;
5323 _masm->cmp8(ExternalAddress((address)flag_addr), value);
5324 _masm->jcc(Assembler::equal, _label);
5325 }
5327 SkipIfEqual::~SkipIfEqual() {
5328 _masm->bind(_label);
5329 }
5331 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5332 movq(tmp, rsp);
5333 // Bang stack for total size given plus shadow page size.
5334 // Bang one page at a time because large size can bang beyond yellow and
5335 // red zones.
5336 Label loop;
5337 bind(loop);
5338 movl(Address(tmp, (-os::vm_page_size())), size );
5339 subq(tmp, os::vm_page_size());
5340 subl(size, os::vm_page_size());
5341 jcc(Assembler::greater, loop);
5343 // Bang down shadow pages too.
5344 // The -1 because we already subtracted 1 page.
5345 for (int i = 0; i< StackShadowPages-1; i++) {
5346 movq(Address(tmp, (-i*os::vm_page_size())), size );
5347 }
5348 }
5350 void MacroAssembler::reinit_heapbase() {
5351 if (UseCompressedOops) {
5352 movptr(r12_heapbase, ExternalAddress((address)Universe::heap_base_addr()));
5353 }
5354 }