Wed, 08 Dec 2010 17:50:49 -0800
Merge
1 /*
2 * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #include "precompiled.hpp"
26 #include "assembler_x86.inline.hpp"
27 #include "gc_interface/collectedHeap.inline.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "memory/cardTableModRefBS.hpp"
30 #include "memory/resourceArea.hpp"
31 #include "prims/methodHandles.hpp"
32 #include "runtime/biasedLocking.hpp"
33 #include "runtime/interfaceSupport.hpp"
34 #include "runtime/objectMonitor.hpp"
35 #include "runtime/os.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "runtime/stubRoutines.hpp"
38 #ifndef SERIALGC
39 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
40 #include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp"
41 #include "gc_implementation/g1/heapRegion.hpp"
42 #endif
44 // Implementation of AddressLiteral
46 AddressLiteral::AddressLiteral(address target, relocInfo::relocType rtype) {
47 _is_lval = false;
48 _target = target;
49 switch (rtype) {
50 case relocInfo::oop_type:
51 // Oops are a special case. Normally they would be their own section
52 // but in cases like icBuffer they are literals in the code stream that
53 // we don't have a section for. We use none so that we get a literal address
54 // which is always patchable.
55 break;
56 case relocInfo::external_word_type:
57 _rspec = external_word_Relocation::spec(target);
58 break;
59 case relocInfo::internal_word_type:
60 _rspec = internal_word_Relocation::spec(target);
61 break;
62 case relocInfo::opt_virtual_call_type:
63 _rspec = opt_virtual_call_Relocation::spec();
64 break;
65 case relocInfo::static_call_type:
66 _rspec = static_call_Relocation::spec();
67 break;
68 case relocInfo::runtime_call_type:
69 _rspec = runtime_call_Relocation::spec();
70 break;
71 case relocInfo::poll_type:
72 case relocInfo::poll_return_type:
73 _rspec = Relocation::spec_simple(rtype);
74 break;
75 case relocInfo::none:
76 break;
77 default:
78 ShouldNotReachHere();
79 break;
80 }
81 }
83 // Implementation of Address
85 #ifdef _LP64
87 Address Address::make_array(ArrayAddress adr) {
88 // Not implementable on 64bit machines
89 // Should have been handled higher up the call chain.
90 ShouldNotReachHere();
91 return Address();
92 }
94 // exceedingly dangerous constructor
95 Address::Address(int disp, address loc, relocInfo::relocType rtype) {
96 _base = noreg;
97 _index = noreg;
98 _scale = no_scale;
99 _disp = disp;
100 switch (rtype) {
101 case relocInfo::external_word_type:
102 _rspec = external_word_Relocation::spec(loc);
103 break;
104 case relocInfo::internal_word_type:
105 _rspec = internal_word_Relocation::spec(loc);
106 break;
107 case relocInfo::runtime_call_type:
108 // HMM
109 _rspec = runtime_call_Relocation::spec();
110 break;
111 case relocInfo::poll_type:
112 case relocInfo::poll_return_type:
113 _rspec = Relocation::spec_simple(rtype);
114 break;
115 case relocInfo::none:
116 break;
117 default:
118 ShouldNotReachHere();
119 }
120 }
121 #else // LP64
123 Address Address::make_array(ArrayAddress adr) {
124 AddressLiteral base = adr.base();
125 Address index = adr.index();
126 assert(index._disp == 0, "must not have disp"); // maybe it can?
127 Address array(index._base, index._index, index._scale, (intptr_t) base.target());
128 array._rspec = base._rspec;
129 return array;
130 }
132 // exceedingly dangerous constructor
133 Address::Address(address loc, RelocationHolder spec) {
134 _base = noreg;
135 _index = noreg;
136 _scale = no_scale;
137 _disp = (intptr_t) loc;
138 _rspec = spec;
139 }
141 #endif // _LP64
145 // Convert the raw encoding form into the form expected by the constructor for
146 // Address. An index of 4 (rsp) corresponds to having no index, so convert
147 // that to noreg for the Address constructor.
148 Address Address::make_raw(int base, int index, int scale, int disp, bool disp_is_oop) {
149 RelocationHolder rspec;
150 if (disp_is_oop) {
151 rspec = Relocation::spec_simple(relocInfo::oop_type);
152 }
153 bool valid_index = index != rsp->encoding();
154 if (valid_index) {
155 Address madr(as_Register(base), as_Register(index), (Address::ScaleFactor)scale, in_ByteSize(disp));
156 madr._rspec = rspec;
157 return madr;
158 } else {
159 Address madr(as_Register(base), noreg, Address::no_scale, in_ByteSize(disp));
160 madr._rspec = rspec;
161 return madr;
162 }
163 }
165 // Implementation of Assembler
167 int AbstractAssembler::code_fill_byte() {
168 return (u_char)'\xF4'; // hlt
169 }
171 // make this go away someday
172 void Assembler::emit_data(jint data, relocInfo::relocType rtype, int format) {
173 if (rtype == relocInfo::none)
174 emit_long(data);
175 else emit_data(data, Relocation::spec_simple(rtype), format);
176 }
178 void Assembler::emit_data(jint data, RelocationHolder const& rspec, int format) {
179 assert(imm_operand == 0, "default format must be immediate in this file");
180 assert(inst_mark() != NULL, "must be inside InstructionMark");
181 if (rspec.type() != relocInfo::none) {
182 #ifdef ASSERT
183 check_relocation(rspec, format);
184 #endif
185 // Do not use AbstractAssembler::relocate, which is not intended for
186 // embedded words. Instead, relocate to the enclosing instruction.
188 // hack. call32 is too wide for mask so use disp32
189 if (format == call32_operand)
190 code_section()->relocate(inst_mark(), rspec, disp32_operand);
191 else
192 code_section()->relocate(inst_mark(), rspec, format);
193 }
194 emit_long(data);
195 }
197 static int encode(Register r) {
198 int enc = r->encoding();
199 if (enc >= 8) {
200 enc -= 8;
201 }
202 return enc;
203 }
205 static int encode(XMMRegister r) {
206 int enc = r->encoding();
207 if (enc >= 8) {
208 enc -= 8;
209 }
210 return enc;
211 }
213 void Assembler::emit_arith_b(int op1, int op2, Register dst, int imm8) {
214 assert(dst->has_byte_register(), "must have byte register");
215 assert(isByte(op1) && isByte(op2), "wrong opcode");
216 assert(isByte(imm8), "not a byte");
217 assert((op1 & 0x01) == 0, "should be 8bit operation");
218 emit_byte(op1);
219 emit_byte(op2 | encode(dst));
220 emit_byte(imm8);
221 }
224 void Assembler::emit_arith(int op1, int op2, Register dst, int32_t imm32) {
225 assert(isByte(op1) && isByte(op2), "wrong opcode");
226 assert((op1 & 0x01) == 1, "should be 32bit operation");
227 assert((op1 & 0x02) == 0, "sign-extension bit should not be set");
228 if (is8bit(imm32)) {
229 emit_byte(op1 | 0x02); // set sign bit
230 emit_byte(op2 | encode(dst));
231 emit_byte(imm32 & 0xFF);
232 } else {
233 emit_byte(op1);
234 emit_byte(op2 | encode(dst));
235 emit_long(imm32);
236 }
237 }
239 // immediate-to-memory forms
240 void Assembler::emit_arith_operand(int op1, Register rm, Address adr, int32_t imm32) {
241 assert((op1 & 0x01) == 1, "should be 32bit operation");
242 assert((op1 & 0x02) == 0, "sign-extension bit should not be set");
243 if (is8bit(imm32)) {
244 emit_byte(op1 | 0x02); // set sign bit
245 emit_operand(rm, adr, 1);
246 emit_byte(imm32 & 0xFF);
247 } else {
248 emit_byte(op1);
249 emit_operand(rm, adr, 4);
250 emit_long(imm32);
251 }
252 }
254 void Assembler::emit_arith(int op1, int op2, Register dst, jobject obj) {
255 LP64_ONLY(ShouldNotReachHere());
256 assert(isByte(op1) && isByte(op2), "wrong opcode");
257 assert((op1 & 0x01) == 1, "should be 32bit operation");
258 assert((op1 & 0x02) == 0, "sign-extension bit should not be set");
259 InstructionMark im(this);
260 emit_byte(op1);
261 emit_byte(op2 | encode(dst));
262 emit_data((intptr_t)obj, relocInfo::oop_type, 0);
263 }
266 void Assembler::emit_arith(int op1, int op2, Register dst, Register src) {
267 assert(isByte(op1) && isByte(op2), "wrong opcode");
268 emit_byte(op1);
269 emit_byte(op2 | encode(dst) << 3 | encode(src));
270 }
273 void Assembler::emit_operand(Register reg, Register base, Register index,
274 Address::ScaleFactor scale, int disp,
275 RelocationHolder const& rspec,
276 int rip_relative_correction) {
277 relocInfo::relocType rtype = (relocInfo::relocType) rspec.type();
279 // Encode the registers as needed in the fields they are used in
281 int regenc = encode(reg) << 3;
282 int indexenc = index->is_valid() ? encode(index) << 3 : 0;
283 int baseenc = base->is_valid() ? encode(base) : 0;
285 if (base->is_valid()) {
286 if (index->is_valid()) {
287 assert(scale != Address::no_scale, "inconsistent address");
288 // [base + index*scale + disp]
289 if (disp == 0 && rtype == relocInfo::none &&
290 base != rbp LP64_ONLY(&& base != r13)) {
291 // [base + index*scale]
292 // [00 reg 100][ss index base]
293 assert(index != rsp, "illegal addressing mode");
294 emit_byte(0x04 | regenc);
295 emit_byte(scale << 6 | indexenc | baseenc);
296 } else if (is8bit(disp) && rtype == relocInfo::none) {
297 // [base + index*scale + imm8]
298 // [01 reg 100][ss index base] imm8
299 assert(index != rsp, "illegal addressing mode");
300 emit_byte(0x44 | regenc);
301 emit_byte(scale << 6 | indexenc | baseenc);
302 emit_byte(disp & 0xFF);
303 } else {
304 // [base + index*scale + disp32]
305 // [10 reg 100][ss index base] disp32
306 assert(index != rsp, "illegal addressing mode");
307 emit_byte(0x84 | regenc);
308 emit_byte(scale << 6 | indexenc | baseenc);
309 emit_data(disp, rspec, disp32_operand);
310 }
311 } else if (base == rsp LP64_ONLY(|| base == r12)) {
312 // [rsp + disp]
313 if (disp == 0 && rtype == relocInfo::none) {
314 // [rsp]
315 // [00 reg 100][00 100 100]
316 emit_byte(0x04 | regenc);
317 emit_byte(0x24);
318 } else if (is8bit(disp) && rtype == relocInfo::none) {
319 // [rsp + imm8]
320 // [01 reg 100][00 100 100] disp8
321 emit_byte(0x44 | regenc);
322 emit_byte(0x24);
323 emit_byte(disp & 0xFF);
324 } else {
325 // [rsp + imm32]
326 // [10 reg 100][00 100 100] disp32
327 emit_byte(0x84 | regenc);
328 emit_byte(0x24);
329 emit_data(disp, rspec, disp32_operand);
330 }
331 } else {
332 // [base + disp]
333 assert(base != rsp LP64_ONLY(&& base != r12), "illegal addressing mode");
334 if (disp == 0 && rtype == relocInfo::none &&
335 base != rbp LP64_ONLY(&& base != r13)) {
336 // [base]
337 // [00 reg base]
338 emit_byte(0x00 | regenc | baseenc);
339 } else if (is8bit(disp) && rtype == relocInfo::none) {
340 // [base + disp8]
341 // [01 reg base] disp8
342 emit_byte(0x40 | regenc | baseenc);
343 emit_byte(disp & 0xFF);
344 } else {
345 // [base + disp32]
346 // [10 reg base] disp32
347 emit_byte(0x80 | regenc | baseenc);
348 emit_data(disp, rspec, disp32_operand);
349 }
350 }
351 } else {
352 if (index->is_valid()) {
353 assert(scale != Address::no_scale, "inconsistent address");
354 // [index*scale + disp]
355 // [00 reg 100][ss index 101] disp32
356 assert(index != rsp, "illegal addressing mode");
357 emit_byte(0x04 | regenc);
358 emit_byte(scale << 6 | indexenc | 0x05);
359 emit_data(disp, rspec, disp32_operand);
360 } else if (rtype != relocInfo::none ) {
361 // [disp] (64bit) RIP-RELATIVE (32bit) abs
362 // [00 000 101] disp32
364 emit_byte(0x05 | regenc);
365 // Note that the RIP-rel. correction applies to the generated
366 // disp field, but _not_ to the target address in the rspec.
368 // disp was created by converting the target address minus the pc
369 // at the start of the instruction. That needs more correction here.
370 // intptr_t disp = target - next_ip;
371 assert(inst_mark() != NULL, "must be inside InstructionMark");
372 address next_ip = pc() + sizeof(int32_t) + rip_relative_correction;
373 int64_t adjusted = disp;
374 // Do rip-rel adjustment for 64bit
375 LP64_ONLY(adjusted -= (next_ip - inst_mark()));
376 assert(is_simm32(adjusted),
377 "must be 32bit offset (RIP relative address)");
378 emit_data((int32_t) adjusted, rspec, disp32_operand);
380 } else {
381 // 32bit never did this, did everything as the rip-rel/disp code above
382 // [disp] ABSOLUTE
383 // [00 reg 100][00 100 101] disp32
384 emit_byte(0x04 | regenc);
385 emit_byte(0x25);
386 emit_data(disp, rspec, disp32_operand);
387 }
388 }
389 }
391 void Assembler::emit_operand(XMMRegister reg, Register base, Register index,
392 Address::ScaleFactor scale, int disp,
393 RelocationHolder const& rspec) {
394 emit_operand((Register)reg, base, index, scale, disp, rspec);
395 }
397 // Secret local extension to Assembler::WhichOperand:
398 #define end_pc_operand (_WhichOperand_limit)
400 address Assembler::locate_operand(address inst, WhichOperand which) {
401 // Decode the given instruction, and return the address of
402 // an embedded 32-bit operand word.
404 // If "which" is disp32_operand, selects the displacement portion
405 // of an effective address specifier.
406 // If "which" is imm64_operand, selects the trailing immediate constant.
407 // If "which" is call32_operand, selects the displacement of a call or jump.
408 // Caller is responsible for ensuring that there is such an operand,
409 // and that it is 32/64 bits wide.
411 // If "which" is end_pc_operand, find the end of the instruction.
413 address ip = inst;
414 bool is_64bit = false;
416 debug_only(bool has_disp32 = false);
417 int tail_size = 0; // other random bytes (#32, #16, etc.) at end of insn
419 again_after_prefix:
420 switch (0xFF & *ip++) {
422 // These convenience macros generate groups of "case" labels for the switch.
423 #define REP4(x) (x)+0: case (x)+1: case (x)+2: case (x)+3
424 #define REP8(x) (x)+0: case (x)+1: case (x)+2: case (x)+3: \
425 case (x)+4: case (x)+5: case (x)+6: case (x)+7
426 #define REP16(x) REP8((x)+0): \
427 case REP8((x)+8)
429 case CS_segment:
430 case SS_segment:
431 case DS_segment:
432 case ES_segment:
433 case FS_segment:
434 case GS_segment:
435 // Seems dubious
436 LP64_ONLY(assert(false, "shouldn't have that prefix"));
437 assert(ip == inst+1, "only one prefix allowed");
438 goto again_after_prefix;
440 case 0x67:
441 case REX:
442 case REX_B:
443 case REX_X:
444 case REX_XB:
445 case REX_R:
446 case REX_RB:
447 case REX_RX:
448 case REX_RXB:
449 NOT_LP64(assert(false, "64bit prefixes"));
450 goto again_after_prefix;
452 case REX_W:
453 case REX_WB:
454 case REX_WX:
455 case REX_WXB:
456 case REX_WR:
457 case REX_WRB:
458 case REX_WRX:
459 case REX_WRXB:
460 NOT_LP64(assert(false, "64bit prefixes"));
461 is_64bit = true;
462 goto again_after_prefix;
464 case 0xFF: // pushq a; decl a; incl a; call a; jmp a
465 case 0x88: // movb a, r
466 case 0x89: // movl a, r
467 case 0x8A: // movb r, a
468 case 0x8B: // movl r, a
469 case 0x8F: // popl a
470 debug_only(has_disp32 = true);
471 break;
473 case 0x68: // pushq #32
474 if (which == end_pc_operand) {
475 return ip + 4;
476 }
477 assert(which == imm_operand && !is_64bit, "pushl has no disp32 or 64bit immediate");
478 return ip; // not produced by emit_operand
480 case 0x66: // movw ... (size prefix)
481 again_after_size_prefix2:
482 switch (0xFF & *ip++) {
483 case REX:
484 case REX_B:
485 case REX_X:
486 case REX_XB:
487 case REX_R:
488 case REX_RB:
489 case REX_RX:
490 case REX_RXB:
491 case REX_W:
492 case REX_WB:
493 case REX_WX:
494 case REX_WXB:
495 case REX_WR:
496 case REX_WRB:
497 case REX_WRX:
498 case REX_WRXB:
499 NOT_LP64(assert(false, "64bit prefix found"));
500 goto again_after_size_prefix2;
501 case 0x8B: // movw r, a
502 case 0x89: // movw a, r
503 debug_only(has_disp32 = true);
504 break;
505 case 0xC7: // movw a, #16
506 debug_only(has_disp32 = true);
507 tail_size = 2; // the imm16
508 break;
509 case 0x0F: // several SSE/SSE2 variants
510 ip--; // reparse the 0x0F
511 goto again_after_prefix;
512 default:
513 ShouldNotReachHere();
514 }
515 break;
517 case REP8(0xB8): // movl/q r, #32/#64(oop?)
518 if (which == end_pc_operand) return ip + (is_64bit ? 8 : 4);
519 // these asserts are somewhat nonsensical
520 #ifndef _LP64
521 assert(which == imm_operand || which == disp32_operand, "");
522 #else
523 assert((which == call32_operand || which == imm_operand) && is_64bit ||
524 which == narrow_oop_operand && !is_64bit, "");
525 #endif // _LP64
526 return ip;
528 case 0x69: // imul r, a, #32
529 case 0xC7: // movl a, #32(oop?)
530 tail_size = 4;
531 debug_only(has_disp32 = true); // has both kinds of operands!
532 break;
534 case 0x0F: // movx..., etc.
535 switch (0xFF & *ip++) {
536 case 0x12: // movlps
537 case 0x28: // movaps
538 case 0x2E: // ucomiss
539 case 0x2F: // comiss
540 case 0x54: // andps
541 case 0x55: // andnps
542 case 0x56: // orps
543 case 0x57: // xorps
544 case 0x6E: // movd
545 case 0x7E: // movd
546 case 0xAE: // ldmxcsr a
547 // 64bit side says it these have both operands but that doesn't
548 // appear to be true
549 debug_only(has_disp32 = true);
550 break;
552 case 0xAD: // shrd r, a, %cl
553 case 0xAF: // imul r, a
554 case 0xBE: // movsbl r, a (movsxb)
555 case 0xBF: // movswl r, a (movsxw)
556 case 0xB6: // movzbl r, a (movzxb)
557 case 0xB7: // movzwl r, a (movzxw)
558 case REP16(0x40): // cmovl cc, r, a
559 case 0xB0: // cmpxchgb
560 case 0xB1: // cmpxchg
561 case 0xC1: // xaddl
562 case 0xC7: // cmpxchg8
563 case REP16(0x90): // setcc a
564 debug_only(has_disp32 = true);
565 // fall out of the switch to decode the address
566 break;
568 case 0xAC: // shrd r, a, #8
569 debug_only(has_disp32 = true);
570 tail_size = 1; // the imm8
571 break;
573 case REP16(0x80): // jcc rdisp32
574 if (which == end_pc_operand) return ip + 4;
575 assert(which == call32_operand, "jcc has no disp32 or imm");
576 return ip;
577 default:
578 ShouldNotReachHere();
579 }
580 break;
582 case 0x81: // addl a, #32; addl r, #32
583 // also: orl, adcl, sbbl, andl, subl, xorl, cmpl
584 // on 32bit in the case of cmpl, the imm might be an oop
585 tail_size = 4;
586 debug_only(has_disp32 = true); // has both kinds of operands!
587 break;
589 case 0x83: // addl a, #8; addl r, #8
590 // also: orl, adcl, sbbl, andl, subl, xorl, cmpl
591 debug_only(has_disp32 = true); // has both kinds of operands!
592 tail_size = 1;
593 break;
595 case 0x9B:
596 switch (0xFF & *ip++) {
597 case 0xD9: // fnstcw a
598 debug_only(has_disp32 = true);
599 break;
600 default:
601 ShouldNotReachHere();
602 }
603 break;
605 case REP4(0x00): // addb a, r; addl a, r; addb r, a; addl r, a
606 case REP4(0x10): // adc...
607 case REP4(0x20): // and...
608 case REP4(0x30): // xor...
609 case REP4(0x08): // or...
610 case REP4(0x18): // sbb...
611 case REP4(0x28): // sub...
612 case 0xF7: // mull a
613 case 0x8D: // lea r, a
614 case 0x87: // xchg r, a
615 case REP4(0x38): // cmp...
616 case 0x85: // test r, a
617 debug_only(has_disp32 = true); // has both kinds of operands!
618 break;
620 case 0xC1: // sal a, #8; sar a, #8; shl a, #8; shr a, #8
621 case 0xC6: // movb a, #8
622 case 0x80: // cmpb a, #8
623 case 0x6B: // imul r, a, #8
624 debug_only(has_disp32 = true); // has both kinds of operands!
625 tail_size = 1; // the imm8
626 break;
628 case 0xE8: // call rdisp32
629 case 0xE9: // jmp rdisp32
630 if (which == end_pc_operand) return ip + 4;
631 assert(which == call32_operand, "call has no disp32 or imm");
632 return ip;
634 case 0xD1: // sal a, 1; sar a, 1; shl a, 1; shr a, 1
635 case 0xD3: // sal a, %cl; sar a, %cl; shl a, %cl; shr a, %cl
636 case 0xD9: // fld_s a; fst_s a; fstp_s a; fldcw a
637 case 0xDD: // fld_d a; fst_d a; fstp_d a
638 case 0xDB: // fild_s a; fistp_s a; fld_x a; fstp_x a
639 case 0xDF: // fild_d a; fistp_d a
640 case 0xD8: // fadd_s a; fsubr_s a; fmul_s a; fdivr_s a; fcomp_s a
641 case 0xDC: // fadd_d a; fsubr_d a; fmul_d a; fdivr_d a; fcomp_d a
642 case 0xDE: // faddp_d a; fsubrp_d a; fmulp_d a; fdivrp_d a; fcompp_d a
643 debug_only(has_disp32 = true);
644 break;
646 case 0xF0: // Lock
647 assert(os::is_MP(), "only on MP");
648 goto again_after_prefix;
650 case 0xF3: // For SSE
651 case 0xF2: // For SSE2
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 NOT_LP64(assert(false, "found 64bit prefix"));
670 ip++;
671 default:
672 ip++;
673 }
674 debug_only(has_disp32 = true); // has both kinds of operands!
675 break;
677 default:
678 ShouldNotReachHere();
680 #undef REP8
681 #undef REP16
682 }
684 assert(which != call32_operand, "instruction is not a call, jmp, or jcc");
685 #ifdef _LP64
686 assert(which != imm_operand, "instruction is not a movq reg, imm64");
687 #else
688 // assert(which != imm_operand || has_imm32, "instruction has no imm32 field");
689 assert(which != imm_operand || has_disp32, "instruction has no imm32 field");
690 #endif // LP64
691 assert(which != disp32_operand || has_disp32, "instruction has no disp32 field");
693 // parse the output of emit_operand
694 int op2 = 0xFF & *ip++;
695 int base = op2 & 0x07;
696 int op3 = -1;
697 const int b100 = 4;
698 const int b101 = 5;
699 if (base == b100 && (op2 >> 6) != 3) {
700 op3 = 0xFF & *ip++;
701 base = op3 & 0x07; // refetch the base
702 }
703 // now ip points at the disp (if any)
705 switch (op2 >> 6) {
706 case 0:
707 // [00 reg 100][ss index base]
708 // [00 reg 100][00 100 esp]
709 // [00 reg base]
710 // [00 reg 100][ss index 101][disp32]
711 // [00 reg 101] [disp32]
713 if (base == b101) {
714 if (which == disp32_operand)
715 return ip; // caller wants the disp32
716 ip += 4; // skip the disp32
717 }
718 break;
720 case 1:
721 // [01 reg 100][ss index base][disp8]
722 // [01 reg 100][00 100 esp][disp8]
723 // [01 reg base] [disp8]
724 ip += 1; // skip the disp8
725 break;
727 case 2:
728 // [10 reg 100][ss index base][disp32]
729 // [10 reg 100][00 100 esp][disp32]
730 // [10 reg base] [disp32]
731 if (which == disp32_operand)
732 return ip; // caller wants the disp32
733 ip += 4; // skip the disp32
734 break;
736 case 3:
737 // [11 reg base] (not a memory addressing mode)
738 break;
739 }
741 if (which == end_pc_operand) {
742 return ip + tail_size;
743 }
745 #ifdef _LP64
746 assert(which == narrow_oop_operand && !is_64bit, "instruction is not a movl adr, imm32");
747 #else
748 assert(which == imm_operand, "instruction has only an imm field");
749 #endif // LP64
750 return ip;
751 }
753 address Assembler::locate_next_instruction(address inst) {
754 // Secretly share code with locate_operand:
755 return locate_operand(inst, end_pc_operand);
756 }
759 #ifdef ASSERT
760 void Assembler::check_relocation(RelocationHolder const& rspec, int format) {
761 address inst = inst_mark();
762 assert(inst != NULL && inst < pc(), "must point to beginning of instruction");
763 address opnd;
765 Relocation* r = rspec.reloc();
766 if (r->type() == relocInfo::none) {
767 return;
768 } else if (r->is_call() || format == call32_operand) {
769 // assert(format == imm32_operand, "cannot specify a nonzero format");
770 opnd = locate_operand(inst, call32_operand);
771 } else if (r->is_data()) {
772 assert(format == imm_operand || format == disp32_operand
773 LP64_ONLY(|| format == narrow_oop_operand), "format ok");
774 opnd = locate_operand(inst, (WhichOperand)format);
775 } else {
776 assert(format == imm_operand, "cannot specify a format");
777 return;
778 }
779 assert(opnd == pc(), "must put operand where relocs can find it");
780 }
781 #endif // ASSERT
783 void Assembler::emit_operand32(Register reg, Address adr) {
784 assert(reg->encoding() < 8, "no extended registers");
785 assert(!adr.base_needs_rex() && !adr.index_needs_rex(), "no extended registers");
786 emit_operand(reg, adr._base, adr._index, adr._scale, adr._disp,
787 adr._rspec);
788 }
790 void Assembler::emit_operand(Register reg, Address adr,
791 int rip_relative_correction) {
792 emit_operand(reg, adr._base, adr._index, adr._scale, adr._disp,
793 adr._rspec,
794 rip_relative_correction);
795 }
797 void Assembler::emit_operand(XMMRegister reg, Address adr) {
798 emit_operand(reg, adr._base, adr._index, adr._scale, adr._disp,
799 adr._rspec);
800 }
802 // MMX operations
803 void Assembler::emit_operand(MMXRegister reg, Address adr) {
804 assert(!adr.base_needs_rex() && !adr.index_needs_rex(), "no extended registers");
805 emit_operand((Register)reg, adr._base, adr._index, adr._scale, adr._disp, adr._rspec);
806 }
808 // work around gcc (3.2.1-7a) bug
809 void Assembler::emit_operand(Address adr, MMXRegister reg) {
810 assert(!adr.base_needs_rex() && !adr.index_needs_rex(), "no extended registers");
811 emit_operand((Register)reg, adr._base, adr._index, adr._scale, adr._disp, adr._rspec);
812 }
815 void Assembler::emit_farith(int b1, int b2, int i) {
816 assert(isByte(b1) && isByte(b2), "wrong opcode");
817 assert(0 <= i && i < 8, "illegal stack offset");
818 emit_byte(b1);
819 emit_byte(b2 + i);
820 }
823 // Now the Assembler instruction (identical for 32/64 bits)
825 void Assembler::adcl(Register dst, int32_t imm32) {
826 prefix(dst);
827 emit_arith(0x81, 0xD0, dst, imm32);
828 }
830 void Assembler::adcl(Register dst, Address src) {
831 InstructionMark im(this);
832 prefix(src, dst);
833 emit_byte(0x13);
834 emit_operand(dst, src);
835 }
837 void Assembler::adcl(Register dst, Register src) {
838 (void) prefix_and_encode(dst->encoding(), src->encoding());
839 emit_arith(0x13, 0xC0, dst, src);
840 }
842 void Assembler::addl(Address dst, int32_t imm32) {
843 InstructionMark im(this);
844 prefix(dst);
845 emit_arith_operand(0x81, rax, dst, imm32);
846 }
848 void Assembler::addl(Address dst, Register src) {
849 InstructionMark im(this);
850 prefix(dst, src);
851 emit_byte(0x01);
852 emit_operand(src, dst);
853 }
855 void Assembler::addl(Register dst, int32_t imm32) {
856 prefix(dst);
857 emit_arith(0x81, 0xC0, dst, imm32);
858 }
860 void Assembler::addl(Register dst, Address src) {
861 InstructionMark im(this);
862 prefix(src, dst);
863 emit_byte(0x03);
864 emit_operand(dst, src);
865 }
867 void Assembler::addl(Register dst, Register src) {
868 (void) prefix_and_encode(dst->encoding(), src->encoding());
869 emit_arith(0x03, 0xC0, dst, src);
870 }
872 void Assembler::addr_nop_4() {
873 // 4 bytes: NOP DWORD PTR [EAX+0]
874 emit_byte(0x0F);
875 emit_byte(0x1F);
876 emit_byte(0x40); // emit_rm(cbuf, 0x1, EAX_enc, EAX_enc);
877 emit_byte(0); // 8-bits offset (1 byte)
878 }
880 void Assembler::addr_nop_5() {
881 // 5 bytes: NOP DWORD PTR [EAX+EAX*0+0] 8-bits offset
882 emit_byte(0x0F);
883 emit_byte(0x1F);
884 emit_byte(0x44); // emit_rm(cbuf, 0x1, EAX_enc, 0x4);
885 emit_byte(0x00); // emit_rm(cbuf, 0x0, EAX_enc, EAX_enc);
886 emit_byte(0); // 8-bits offset (1 byte)
887 }
889 void Assembler::addr_nop_7() {
890 // 7 bytes: NOP DWORD PTR [EAX+0] 32-bits offset
891 emit_byte(0x0F);
892 emit_byte(0x1F);
893 emit_byte(0x80); // emit_rm(cbuf, 0x2, EAX_enc, EAX_enc);
894 emit_long(0); // 32-bits offset (4 bytes)
895 }
897 void Assembler::addr_nop_8() {
898 // 8 bytes: NOP DWORD PTR [EAX+EAX*0+0] 32-bits offset
899 emit_byte(0x0F);
900 emit_byte(0x1F);
901 emit_byte(0x84); // emit_rm(cbuf, 0x2, EAX_enc, 0x4);
902 emit_byte(0x00); // emit_rm(cbuf, 0x0, EAX_enc, EAX_enc);
903 emit_long(0); // 32-bits offset (4 bytes)
904 }
906 void Assembler::addsd(XMMRegister dst, XMMRegister src) {
907 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
908 emit_byte(0xF2);
909 int encode = prefix_and_encode(dst->encoding(), src->encoding());
910 emit_byte(0x0F);
911 emit_byte(0x58);
912 emit_byte(0xC0 | encode);
913 }
915 void Assembler::addsd(XMMRegister dst, Address src) {
916 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
917 InstructionMark im(this);
918 emit_byte(0xF2);
919 prefix(src, dst);
920 emit_byte(0x0F);
921 emit_byte(0x58);
922 emit_operand(dst, src);
923 }
925 void Assembler::addss(XMMRegister dst, XMMRegister src) {
926 NOT_LP64(assert(VM_Version::supports_sse(), ""));
927 emit_byte(0xF3);
928 int encode = prefix_and_encode(dst->encoding(), src->encoding());
929 emit_byte(0x0F);
930 emit_byte(0x58);
931 emit_byte(0xC0 | encode);
932 }
934 void Assembler::addss(XMMRegister dst, Address src) {
935 NOT_LP64(assert(VM_Version::supports_sse(), ""));
936 InstructionMark im(this);
937 emit_byte(0xF3);
938 prefix(src, dst);
939 emit_byte(0x0F);
940 emit_byte(0x58);
941 emit_operand(dst, src);
942 }
944 void Assembler::andl(Register dst, int32_t imm32) {
945 prefix(dst);
946 emit_arith(0x81, 0xE0, dst, imm32);
947 }
949 void Assembler::andl(Register dst, Address src) {
950 InstructionMark im(this);
951 prefix(src, dst);
952 emit_byte(0x23);
953 emit_operand(dst, src);
954 }
956 void Assembler::andl(Register dst, Register src) {
957 (void) prefix_and_encode(dst->encoding(), src->encoding());
958 emit_arith(0x23, 0xC0, dst, src);
959 }
961 void Assembler::andpd(XMMRegister dst, Address src) {
962 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
963 InstructionMark im(this);
964 emit_byte(0x66);
965 prefix(src, dst);
966 emit_byte(0x0F);
967 emit_byte(0x54);
968 emit_operand(dst, src);
969 }
971 void Assembler::bsfl(Register dst, Register src) {
972 int encode = prefix_and_encode(dst->encoding(), src->encoding());
973 emit_byte(0x0F);
974 emit_byte(0xBC);
975 emit_byte(0xC0 | encode);
976 }
978 void Assembler::bsrl(Register dst, Register src) {
979 assert(!VM_Version::supports_lzcnt(), "encoding is treated as LZCNT");
980 int encode = prefix_and_encode(dst->encoding(), src->encoding());
981 emit_byte(0x0F);
982 emit_byte(0xBD);
983 emit_byte(0xC0 | encode);
984 }
986 void Assembler::bswapl(Register reg) { // bswap
987 int encode = prefix_and_encode(reg->encoding());
988 emit_byte(0x0F);
989 emit_byte(0xC8 | encode);
990 }
992 void Assembler::call(Label& L, relocInfo::relocType rtype) {
993 // suspect disp32 is always good
994 int operand = LP64_ONLY(disp32_operand) NOT_LP64(imm_operand);
996 if (L.is_bound()) {
997 const int long_size = 5;
998 int offs = (int)( target(L) - pc() );
999 assert(offs <= 0, "assembler error");
1000 InstructionMark im(this);
1001 // 1110 1000 #32-bit disp
1002 emit_byte(0xE8);
1003 emit_data(offs - long_size, rtype, operand);
1004 } else {
1005 InstructionMark im(this);
1006 // 1110 1000 #32-bit disp
1007 L.add_patch_at(code(), locator());
1009 emit_byte(0xE8);
1010 emit_data(int(0), rtype, operand);
1011 }
1012 }
1014 void Assembler::call(Register dst) {
1015 // This was originally using a 32bit register encoding
1016 // and surely we want 64bit!
1017 // this is a 32bit encoding but in 64bit mode the default
1018 // operand size is 64bit so there is no need for the
1019 // wide prefix. So prefix only happens if we use the
1020 // new registers. Much like push/pop.
1021 int x = offset();
1022 // this may be true but dbx disassembles it as if it
1023 // were 32bits...
1024 // int encode = prefix_and_encode(dst->encoding());
1025 // if (offset() != x) assert(dst->encoding() >= 8, "what?");
1026 int encode = prefixq_and_encode(dst->encoding());
1028 emit_byte(0xFF);
1029 emit_byte(0xD0 | encode);
1030 }
1033 void Assembler::call(Address adr) {
1034 InstructionMark im(this);
1035 prefix(adr);
1036 emit_byte(0xFF);
1037 emit_operand(rdx, adr);
1038 }
1040 void Assembler::call_literal(address entry, RelocationHolder const& rspec) {
1041 assert(entry != NULL, "call most probably wrong");
1042 InstructionMark im(this);
1043 emit_byte(0xE8);
1044 intptr_t disp = entry - (_code_pos + sizeof(int32_t));
1045 assert(is_simm32(disp), "must be 32bit offset (call2)");
1046 // Technically, should use call32_operand, but this format is
1047 // implied by the fact that we're emitting a call instruction.
1049 int operand = LP64_ONLY(disp32_operand) NOT_LP64(call32_operand);
1050 emit_data((int) disp, rspec, operand);
1051 }
1053 void Assembler::cdql() {
1054 emit_byte(0x99);
1055 }
1057 void Assembler::cmovl(Condition cc, Register dst, Register src) {
1058 NOT_LP64(guarantee(VM_Version::supports_cmov(), "illegal instruction"));
1059 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1060 emit_byte(0x0F);
1061 emit_byte(0x40 | cc);
1062 emit_byte(0xC0 | encode);
1063 }
1066 void Assembler::cmovl(Condition cc, Register dst, Address src) {
1067 NOT_LP64(guarantee(VM_Version::supports_cmov(), "illegal instruction"));
1068 prefix(src, dst);
1069 emit_byte(0x0F);
1070 emit_byte(0x40 | cc);
1071 emit_operand(dst, src);
1072 }
1074 void Assembler::cmpb(Address dst, int imm8) {
1075 InstructionMark im(this);
1076 prefix(dst);
1077 emit_byte(0x80);
1078 emit_operand(rdi, dst, 1);
1079 emit_byte(imm8);
1080 }
1082 void Assembler::cmpl(Address dst, int32_t imm32) {
1083 InstructionMark im(this);
1084 prefix(dst);
1085 emit_byte(0x81);
1086 emit_operand(rdi, dst, 4);
1087 emit_long(imm32);
1088 }
1090 void Assembler::cmpl(Register dst, int32_t imm32) {
1091 prefix(dst);
1092 emit_arith(0x81, 0xF8, dst, imm32);
1093 }
1095 void Assembler::cmpl(Register dst, Register src) {
1096 (void) prefix_and_encode(dst->encoding(), src->encoding());
1097 emit_arith(0x3B, 0xC0, dst, src);
1098 }
1101 void Assembler::cmpl(Register dst, Address src) {
1102 InstructionMark im(this);
1103 prefix(src, dst);
1104 emit_byte(0x3B);
1105 emit_operand(dst, src);
1106 }
1108 void Assembler::cmpw(Address dst, int imm16) {
1109 InstructionMark im(this);
1110 assert(!dst.base_needs_rex() && !dst.index_needs_rex(), "no extended registers");
1111 emit_byte(0x66);
1112 emit_byte(0x81);
1113 emit_operand(rdi, dst, 2);
1114 emit_word(imm16);
1115 }
1117 // The 32-bit cmpxchg compares the value at adr with the contents of rax,
1118 // and stores reg into adr if so; otherwise, the value at adr is loaded into rax,.
1119 // The ZF is set if the compared values were equal, and cleared otherwise.
1120 void Assembler::cmpxchgl(Register reg, Address adr) { // cmpxchg
1121 if (Atomics & 2) {
1122 // caveat: no instructionmark, so this isn't relocatable.
1123 // Emit a synthetic, non-atomic, CAS equivalent.
1124 // Beware. The synthetic form sets all ICCs, not just ZF.
1125 // cmpxchg r,[m] is equivalent to rax, = CAS (m, rax, r)
1126 cmpl(rax, adr);
1127 movl(rax, adr);
1128 if (reg != rax) {
1129 Label L ;
1130 jcc(Assembler::notEqual, L);
1131 movl(adr, reg);
1132 bind(L);
1133 }
1134 } else {
1135 InstructionMark im(this);
1136 prefix(adr, reg);
1137 emit_byte(0x0F);
1138 emit_byte(0xB1);
1139 emit_operand(reg, adr);
1140 }
1141 }
1143 void Assembler::comisd(XMMRegister dst, Address src) {
1144 // NOTE: dbx seems to decode this as comiss even though the
1145 // 0x66 is there. Strangly ucomisd comes out correct
1146 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1147 emit_byte(0x66);
1148 comiss(dst, src);
1149 }
1151 void Assembler::comiss(XMMRegister dst, Address src) {
1152 NOT_LP64(assert(VM_Version::supports_sse(), ""));
1154 InstructionMark im(this);
1155 prefix(src, dst);
1156 emit_byte(0x0F);
1157 emit_byte(0x2F);
1158 emit_operand(dst, src);
1159 }
1161 void Assembler::cvtdq2pd(XMMRegister dst, XMMRegister src) {
1162 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1163 emit_byte(0xF3);
1164 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1165 emit_byte(0x0F);
1166 emit_byte(0xE6);
1167 emit_byte(0xC0 | encode);
1168 }
1170 void Assembler::cvtdq2ps(XMMRegister dst, XMMRegister src) {
1171 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1172 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1173 emit_byte(0x0F);
1174 emit_byte(0x5B);
1175 emit_byte(0xC0 | encode);
1176 }
1178 void Assembler::cvtsd2ss(XMMRegister dst, XMMRegister src) {
1179 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1180 emit_byte(0xF2);
1181 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1182 emit_byte(0x0F);
1183 emit_byte(0x5A);
1184 emit_byte(0xC0 | encode);
1185 }
1187 void Assembler::cvtsi2sdl(XMMRegister dst, Register src) {
1188 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1189 emit_byte(0xF2);
1190 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1191 emit_byte(0x0F);
1192 emit_byte(0x2A);
1193 emit_byte(0xC0 | encode);
1194 }
1196 void Assembler::cvtsi2ssl(XMMRegister dst, Register src) {
1197 NOT_LP64(assert(VM_Version::supports_sse(), ""));
1198 emit_byte(0xF3);
1199 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1200 emit_byte(0x0F);
1201 emit_byte(0x2A);
1202 emit_byte(0xC0 | encode);
1203 }
1205 void Assembler::cvtss2sd(XMMRegister dst, XMMRegister src) {
1206 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1207 emit_byte(0xF3);
1208 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1209 emit_byte(0x0F);
1210 emit_byte(0x5A);
1211 emit_byte(0xC0 | encode);
1212 }
1214 void Assembler::cvttsd2sil(Register dst, XMMRegister src) {
1215 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1216 emit_byte(0xF2);
1217 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1218 emit_byte(0x0F);
1219 emit_byte(0x2C);
1220 emit_byte(0xC0 | encode);
1221 }
1223 void Assembler::cvttss2sil(Register dst, XMMRegister src) {
1224 NOT_LP64(assert(VM_Version::supports_sse(), ""));
1225 emit_byte(0xF3);
1226 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1227 emit_byte(0x0F);
1228 emit_byte(0x2C);
1229 emit_byte(0xC0 | encode);
1230 }
1232 void Assembler::decl(Address dst) {
1233 // Don't use it directly. Use MacroAssembler::decrement() instead.
1234 InstructionMark im(this);
1235 prefix(dst);
1236 emit_byte(0xFF);
1237 emit_operand(rcx, dst);
1238 }
1240 void Assembler::divsd(XMMRegister dst, Address src) {
1241 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1242 InstructionMark im(this);
1243 emit_byte(0xF2);
1244 prefix(src, dst);
1245 emit_byte(0x0F);
1246 emit_byte(0x5E);
1247 emit_operand(dst, src);
1248 }
1250 void Assembler::divsd(XMMRegister dst, XMMRegister src) {
1251 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1252 emit_byte(0xF2);
1253 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1254 emit_byte(0x0F);
1255 emit_byte(0x5E);
1256 emit_byte(0xC0 | encode);
1257 }
1259 void Assembler::divss(XMMRegister dst, Address src) {
1260 NOT_LP64(assert(VM_Version::supports_sse(), ""));
1261 InstructionMark im(this);
1262 emit_byte(0xF3);
1263 prefix(src, dst);
1264 emit_byte(0x0F);
1265 emit_byte(0x5E);
1266 emit_operand(dst, src);
1267 }
1269 void Assembler::divss(XMMRegister dst, XMMRegister src) {
1270 NOT_LP64(assert(VM_Version::supports_sse(), ""));
1271 emit_byte(0xF3);
1272 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1273 emit_byte(0x0F);
1274 emit_byte(0x5E);
1275 emit_byte(0xC0 | encode);
1276 }
1278 void Assembler::emms() {
1279 NOT_LP64(assert(VM_Version::supports_mmx(), ""));
1280 emit_byte(0x0F);
1281 emit_byte(0x77);
1282 }
1284 void Assembler::hlt() {
1285 emit_byte(0xF4);
1286 }
1288 void Assembler::idivl(Register src) {
1289 int encode = prefix_and_encode(src->encoding());
1290 emit_byte(0xF7);
1291 emit_byte(0xF8 | encode);
1292 }
1294 void Assembler::divl(Register src) { // Unsigned
1295 int encode = prefix_and_encode(src->encoding());
1296 emit_byte(0xF7);
1297 emit_byte(0xF0 | encode);
1298 }
1300 void Assembler::imull(Register dst, Register src) {
1301 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1302 emit_byte(0x0F);
1303 emit_byte(0xAF);
1304 emit_byte(0xC0 | encode);
1305 }
1308 void Assembler::imull(Register dst, Register src, int value) {
1309 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1310 if (is8bit(value)) {
1311 emit_byte(0x6B);
1312 emit_byte(0xC0 | encode);
1313 emit_byte(value & 0xFF);
1314 } else {
1315 emit_byte(0x69);
1316 emit_byte(0xC0 | encode);
1317 emit_long(value);
1318 }
1319 }
1321 void Assembler::incl(Address dst) {
1322 // Don't use it directly. Use MacroAssembler::increment() instead.
1323 InstructionMark im(this);
1324 prefix(dst);
1325 emit_byte(0xFF);
1326 emit_operand(rax, dst);
1327 }
1329 void Assembler::jcc(Condition cc, Label& L, relocInfo::relocType rtype) {
1330 InstructionMark im(this);
1331 relocate(rtype);
1332 assert((0 <= cc) && (cc < 16), "illegal cc");
1333 if (L.is_bound()) {
1334 address dst = target(L);
1335 assert(dst != NULL, "jcc most probably wrong");
1337 const int short_size = 2;
1338 const int long_size = 6;
1339 intptr_t offs = (intptr_t)dst - (intptr_t)_code_pos;
1340 if (rtype == relocInfo::none && is8bit(offs - short_size)) {
1341 // 0111 tttn #8-bit disp
1342 emit_byte(0x70 | cc);
1343 emit_byte((offs - short_size) & 0xFF);
1344 } else {
1345 // 0000 1111 1000 tttn #32-bit disp
1346 assert(is_simm32(offs - long_size),
1347 "must be 32bit offset (call4)");
1348 emit_byte(0x0F);
1349 emit_byte(0x80 | cc);
1350 emit_long(offs - long_size);
1351 }
1352 } else {
1353 // Note: could eliminate cond. jumps to this jump if condition
1354 // is the same however, seems to be rather unlikely case.
1355 // Note: use jccb() if label to be bound is very close to get
1356 // an 8-bit displacement
1357 L.add_patch_at(code(), locator());
1358 emit_byte(0x0F);
1359 emit_byte(0x80 | cc);
1360 emit_long(0);
1361 }
1362 }
1364 void Assembler::jccb(Condition cc, Label& L) {
1365 if (L.is_bound()) {
1366 const int short_size = 2;
1367 address entry = target(L);
1368 assert(is8bit((intptr_t)entry - ((intptr_t)_code_pos + short_size)),
1369 "Dispacement too large for a short jmp");
1370 intptr_t offs = (intptr_t)entry - (intptr_t)_code_pos;
1371 // 0111 tttn #8-bit disp
1372 emit_byte(0x70 | cc);
1373 emit_byte((offs - short_size) & 0xFF);
1374 } else {
1375 InstructionMark im(this);
1376 L.add_patch_at(code(), locator());
1377 emit_byte(0x70 | cc);
1378 emit_byte(0);
1379 }
1380 }
1382 void Assembler::jmp(Address adr) {
1383 InstructionMark im(this);
1384 prefix(adr);
1385 emit_byte(0xFF);
1386 emit_operand(rsp, adr);
1387 }
1389 void Assembler::jmp(Label& L, relocInfo::relocType rtype) {
1390 if (L.is_bound()) {
1391 address entry = target(L);
1392 assert(entry != NULL, "jmp most probably wrong");
1393 InstructionMark im(this);
1394 const int short_size = 2;
1395 const int long_size = 5;
1396 intptr_t offs = entry - _code_pos;
1397 if (rtype == relocInfo::none && is8bit(offs - short_size)) {
1398 emit_byte(0xEB);
1399 emit_byte((offs - short_size) & 0xFF);
1400 } else {
1401 emit_byte(0xE9);
1402 emit_long(offs - long_size);
1403 }
1404 } else {
1405 // By default, forward jumps are always 32-bit displacements, since
1406 // we can't yet know where the label will be bound. If you're sure that
1407 // the forward jump will not run beyond 256 bytes, use jmpb to
1408 // force an 8-bit displacement.
1409 InstructionMark im(this);
1410 relocate(rtype);
1411 L.add_patch_at(code(), locator());
1412 emit_byte(0xE9);
1413 emit_long(0);
1414 }
1415 }
1417 void Assembler::jmp(Register entry) {
1418 int encode = prefix_and_encode(entry->encoding());
1419 emit_byte(0xFF);
1420 emit_byte(0xE0 | encode);
1421 }
1423 void Assembler::jmp_literal(address dest, RelocationHolder const& rspec) {
1424 InstructionMark im(this);
1425 emit_byte(0xE9);
1426 assert(dest != NULL, "must have a target");
1427 intptr_t disp = dest - (_code_pos + sizeof(int32_t));
1428 assert(is_simm32(disp), "must be 32bit offset (jmp)");
1429 emit_data(disp, rspec.reloc(), call32_operand);
1430 }
1432 void Assembler::jmpb(Label& L) {
1433 if (L.is_bound()) {
1434 const int short_size = 2;
1435 address entry = target(L);
1436 assert(is8bit((entry - _code_pos) + short_size),
1437 "Dispacement too large for a short jmp");
1438 assert(entry != NULL, "jmp most probably wrong");
1439 intptr_t offs = entry - _code_pos;
1440 emit_byte(0xEB);
1441 emit_byte((offs - short_size) & 0xFF);
1442 } else {
1443 InstructionMark im(this);
1444 L.add_patch_at(code(), locator());
1445 emit_byte(0xEB);
1446 emit_byte(0);
1447 }
1448 }
1450 void Assembler::ldmxcsr( Address src) {
1451 NOT_LP64(assert(VM_Version::supports_sse(), ""));
1452 InstructionMark im(this);
1453 prefix(src);
1454 emit_byte(0x0F);
1455 emit_byte(0xAE);
1456 emit_operand(as_Register(2), src);
1457 }
1459 void Assembler::leal(Register dst, Address src) {
1460 InstructionMark im(this);
1461 #ifdef _LP64
1462 emit_byte(0x67); // addr32
1463 prefix(src, dst);
1464 #endif // LP64
1465 emit_byte(0x8D);
1466 emit_operand(dst, src);
1467 }
1469 void Assembler::lock() {
1470 if (Atomics & 1) {
1471 // Emit either nothing, a NOP, or a NOP: prefix
1472 emit_byte(0x90) ;
1473 } else {
1474 emit_byte(0xF0);
1475 }
1476 }
1478 void Assembler::lzcntl(Register dst, Register src) {
1479 assert(VM_Version::supports_lzcnt(), "encoding is treated as BSR");
1480 emit_byte(0xF3);
1481 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1482 emit_byte(0x0F);
1483 emit_byte(0xBD);
1484 emit_byte(0xC0 | encode);
1485 }
1487 // Emit mfence instruction
1488 void Assembler::mfence() {
1489 NOT_LP64(assert(VM_Version::supports_sse2(), "unsupported");)
1490 emit_byte( 0x0F );
1491 emit_byte( 0xAE );
1492 emit_byte( 0xF0 );
1493 }
1495 void Assembler::mov(Register dst, Register src) {
1496 LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src));
1497 }
1499 void Assembler::movapd(XMMRegister dst, XMMRegister src) {
1500 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1501 int dstenc = dst->encoding();
1502 int srcenc = src->encoding();
1503 emit_byte(0x66);
1504 if (dstenc < 8) {
1505 if (srcenc >= 8) {
1506 prefix(REX_B);
1507 srcenc -= 8;
1508 }
1509 } else {
1510 if (srcenc < 8) {
1511 prefix(REX_R);
1512 } else {
1513 prefix(REX_RB);
1514 srcenc -= 8;
1515 }
1516 dstenc -= 8;
1517 }
1518 emit_byte(0x0F);
1519 emit_byte(0x28);
1520 emit_byte(0xC0 | dstenc << 3 | srcenc);
1521 }
1523 void Assembler::movaps(XMMRegister dst, XMMRegister src) {
1524 NOT_LP64(assert(VM_Version::supports_sse(), ""));
1525 int dstenc = dst->encoding();
1526 int srcenc = src->encoding();
1527 if (dstenc < 8) {
1528 if (srcenc >= 8) {
1529 prefix(REX_B);
1530 srcenc -= 8;
1531 }
1532 } else {
1533 if (srcenc < 8) {
1534 prefix(REX_R);
1535 } else {
1536 prefix(REX_RB);
1537 srcenc -= 8;
1538 }
1539 dstenc -= 8;
1540 }
1541 emit_byte(0x0F);
1542 emit_byte(0x28);
1543 emit_byte(0xC0 | dstenc << 3 | srcenc);
1544 }
1546 void Assembler::movb(Register dst, Address src) {
1547 NOT_LP64(assert(dst->has_byte_register(), "must have byte register"));
1548 InstructionMark im(this);
1549 prefix(src, dst, true);
1550 emit_byte(0x8A);
1551 emit_operand(dst, src);
1552 }
1555 void Assembler::movb(Address dst, int imm8) {
1556 InstructionMark im(this);
1557 prefix(dst);
1558 emit_byte(0xC6);
1559 emit_operand(rax, dst, 1);
1560 emit_byte(imm8);
1561 }
1564 void Assembler::movb(Address dst, Register src) {
1565 assert(src->has_byte_register(), "must have byte register");
1566 InstructionMark im(this);
1567 prefix(dst, src, true);
1568 emit_byte(0x88);
1569 emit_operand(src, dst);
1570 }
1572 void Assembler::movdl(XMMRegister dst, Register src) {
1573 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1574 emit_byte(0x66);
1575 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1576 emit_byte(0x0F);
1577 emit_byte(0x6E);
1578 emit_byte(0xC0 | encode);
1579 }
1581 void Assembler::movdl(Register dst, XMMRegister src) {
1582 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1583 emit_byte(0x66);
1584 // swap src/dst to get correct prefix
1585 int encode = prefix_and_encode(src->encoding(), dst->encoding());
1586 emit_byte(0x0F);
1587 emit_byte(0x7E);
1588 emit_byte(0xC0 | encode);
1589 }
1591 void Assembler::movdqa(XMMRegister dst, Address src) {
1592 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1593 InstructionMark im(this);
1594 emit_byte(0x66);
1595 prefix(src, dst);
1596 emit_byte(0x0F);
1597 emit_byte(0x6F);
1598 emit_operand(dst, src);
1599 }
1601 void Assembler::movdqa(XMMRegister dst, XMMRegister src) {
1602 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1603 emit_byte(0x66);
1604 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
1605 emit_byte(0x0F);
1606 emit_byte(0x6F);
1607 emit_byte(0xC0 | encode);
1608 }
1610 void Assembler::movdqa(Address dst, XMMRegister src) {
1611 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1612 InstructionMark im(this);
1613 emit_byte(0x66);
1614 prefix(dst, src);
1615 emit_byte(0x0F);
1616 emit_byte(0x7F);
1617 emit_operand(src, dst);
1618 }
1620 void Assembler::movdqu(XMMRegister dst, Address src) {
1621 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1622 InstructionMark im(this);
1623 emit_byte(0xF3);
1624 prefix(src, dst);
1625 emit_byte(0x0F);
1626 emit_byte(0x6F);
1627 emit_operand(dst, src);
1628 }
1630 void Assembler::movdqu(XMMRegister dst, XMMRegister src) {
1631 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1632 emit_byte(0xF3);
1633 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
1634 emit_byte(0x0F);
1635 emit_byte(0x6F);
1636 emit_byte(0xC0 | encode);
1637 }
1639 void Assembler::movdqu(Address dst, XMMRegister src) {
1640 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1641 InstructionMark im(this);
1642 emit_byte(0xF3);
1643 prefix(dst, src);
1644 emit_byte(0x0F);
1645 emit_byte(0x7F);
1646 emit_operand(src, dst);
1647 }
1649 // Uses zero extension on 64bit
1651 void Assembler::movl(Register dst, int32_t imm32) {
1652 int encode = prefix_and_encode(dst->encoding());
1653 emit_byte(0xB8 | encode);
1654 emit_long(imm32);
1655 }
1657 void Assembler::movl(Register dst, Register src) {
1658 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1659 emit_byte(0x8B);
1660 emit_byte(0xC0 | encode);
1661 }
1663 void Assembler::movl(Register dst, Address src) {
1664 InstructionMark im(this);
1665 prefix(src, dst);
1666 emit_byte(0x8B);
1667 emit_operand(dst, src);
1668 }
1670 void Assembler::movl(Address dst, int32_t imm32) {
1671 InstructionMark im(this);
1672 prefix(dst);
1673 emit_byte(0xC7);
1674 emit_operand(rax, dst, 4);
1675 emit_long(imm32);
1676 }
1678 void Assembler::movl(Address dst, Register src) {
1679 InstructionMark im(this);
1680 prefix(dst, src);
1681 emit_byte(0x89);
1682 emit_operand(src, dst);
1683 }
1685 // New cpus require to use movsd and movss to avoid partial register stall
1686 // when loading from memory. But for old Opteron use movlpd instead of movsd.
1687 // The selection is done in MacroAssembler::movdbl() and movflt().
1688 void Assembler::movlpd(XMMRegister dst, Address src) {
1689 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1690 InstructionMark im(this);
1691 emit_byte(0x66);
1692 prefix(src, dst);
1693 emit_byte(0x0F);
1694 emit_byte(0x12);
1695 emit_operand(dst, src);
1696 }
1698 void Assembler::movq( MMXRegister dst, Address src ) {
1699 assert( VM_Version::supports_mmx(), "" );
1700 emit_byte(0x0F);
1701 emit_byte(0x6F);
1702 emit_operand(dst, src);
1703 }
1705 void Assembler::movq( Address dst, MMXRegister src ) {
1706 assert( VM_Version::supports_mmx(), "" );
1707 emit_byte(0x0F);
1708 emit_byte(0x7F);
1709 // workaround gcc (3.2.1-7a) bug
1710 // In that version of gcc with only an emit_operand(MMX, Address)
1711 // gcc will tail jump and try and reverse the parameters completely
1712 // obliterating dst in the process. By having a version available
1713 // that doesn't need to swap the args at the tail jump the bug is
1714 // avoided.
1715 emit_operand(dst, src);
1716 }
1718 void Assembler::movq(XMMRegister dst, Address src) {
1719 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1720 InstructionMark im(this);
1721 emit_byte(0xF3);
1722 prefix(src, dst);
1723 emit_byte(0x0F);
1724 emit_byte(0x7E);
1725 emit_operand(dst, src);
1726 }
1728 void Assembler::movq(Address dst, XMMRegister src) {
1729 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1730 InstructionMark im(this);
1731 emit_byte(0x66);
1732 prefix(dst, src);
1733 emit_byte(0x0F);
1734 emit_byte(0xD6);
1735 emit_operand(src, dst);
1736 }
1738 void Assembler::movsbl(Register dst, Address src) { // movsxb
1739 InstructionMark im(this);
1740 prefix(src, dst);
1741 emit_byte(0x0F);
1742 emit_byte(0xBE);
1743 emit_operand(dst, src);
1744 }
1746 void Assembler::movsbl(Register dst, Register src) { // movsxb
1747 NOT_LP64(assert(src->has_byte_register(), "must have byte register"));
1748 int encode = prefix_and_encode(dst->encoding(), src->encoding(), true);
1749 emit_byte(0x0F);
1750 emit_byte(0xBE);
1751 emit_byte(0xC0 | encode);
1752 }
1754 void Assembler::movsd(XMMRegister dst, XMMRegister src) {
1755 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1756 emit_byte(0xF2);
1757 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1758 emit_byte(0x0F);
1759 emit_byte(0x10);
1760 emit_byte(0xC0 | encode);
1761 }
1763 void Assembler::movsd(XMMRegister dst, Address src) {
1764 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1765 InstructionMark im(this);
1766 emit_byte(0xF2);
1767 prefix(src, dst);
1768 emit_byte(0x0F);
1769 emit_byte(0x10);
1770 emit_operand(dst, src);
1771 }
1773 void Assembler::movsd(Address dst, XMMRegister src) {
1774 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1775 InstructionMark im(this);
1776 emit_byte(0xF2);
1777 prefix(dst, src);
1778 emit_byte(0x0F);
1779 emit_byte(0x11);
1780 emit_operand(src, dst);
1781 }
1783 void Assembler::movss(XMMRegister dst, XMMRegister src) {
1784 NOT_LP64(assert(VM_Version::supports_sse(), ""));
1785 emit_byte(0xF3);
1786 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1787 emit_byte(0x0F);
1788 emit_byte(0x10);
1789 emit_byte(0xC0 | encode);
1790 }
1792 void Assembler::movss(XMMRegister dst, Address src) {
1793 NOT_LP64(assert(VM_Version::supports_sse(), ""));
1794 InstructionMark im(this);
1795 emit_byte(0xF3);
1796 prefix(src, dst);
1797 emit_byte(0x0F);
1798 emit_byte(0x10);
1799 emit_operand(dst, src);
1800 }
1802 void Assembler::movss(Address dst, XMMRegister src) {
1803 NOT_LP64(assert(VM_Version::supports_sse(), ""));
1804 InstructionMark im(this);
1805 emit_byte(0xF3);
1806 prefix(dst, src);
1807 emit_byte(0x0F);
1808 emit_byte(0x11);
1809 emit_operand(src, dst);
1810 }
1812 void Assembler::movswl(Register dst, Address src) { // movsxw
1813 InstructionMark im(this);
1814 prefix(src, dst);
1815 emit_byte(0x0F);
1816 emit_byte(0xBF);
1817 emit_operand(dst, src);
1818 }
1820 void Assembler::movswl(Register dst, Register src) { // movsxw
1821 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1822 emit_byte(0x0F);
1823 emit_byte(0xBF);
1824 emit_byte(0xC0 | encode);
1825 }
1827 void Assembler::movw(Address dst, int imm16) {
1828 InstructionMark im(this);
1830 emit_byte(0x66); // switch to 16-bit mode
1831 prefix(dst);
1832 emit_byte(0xC7);
1833 emit_operand(rax, dst, 2);
1834 emit_word(imm16);
1835 }
1837 void Assembler::movw(Register dst, Address src) {
1838 InstructionMark im(this);
1839 emit_byte(0x66);
1840 prefix(src, dst);
1841 emit_byte(0x8B);
1842 emit_operand(dst, src);
1843 }
1845 void Assembler::movw(Address dst, Register src) {
1846 InstructionMark im(this);
1847 emit_byte(0x66);
1848 prefix(dst, src);
1849 emit_byte(0x89);
1850 emit_operand(src, dst);
1851 }
1853 void Assembler::movzbl(Register dst, Address src) { // movzxb
1854 InstructionMark im(this);
1855 prefix(src, dst);
1856 emit_byte(0x0F);
1857 emit_byte(0xB6);
1858 emit_operand(dst, src);
1859 }
1861 void Assembler::movzbl(Register dst, Register src) { // movzxb
1862 NOT_LP64(assert(src->has_byte_register(), "must have byte register"));
1863 int encode = prefix_and_encode(dst->encoding(), src->encoding(), true);
1864 emit_byte(0x0F);
1865 emit_byte(0xB6);
1866 emit_byte(0xC0 | encode);
1867 }
1869 void Assembler::movzwl(Register dst, Address src) { // movzxw
1870 InstructionMark im(this);
1871 prefix(src, dst);
1872 emit_byte(0x0F);
1873 emit_byte(0xB7);
1874 emit_operand(dst, src);
1875 }
1877 void Assembler::movzwl(Register dst, Register src) { // movzxw
1878 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1879 emit_byte(0x0F);
1880 emit_byte(0xB7);
1881 emit_byte(0xC0 | encode);
1882 }
1884 void Assembler::mull(Address src) {
1885 InstructionMark im(this);
1886 prefix(src);
1887 emit_byte(0xF7);
1888 emit_operand(rsp, src);
1889 }
1891 void Assembler::mull(Register src) {
1892 int encode = prefix_and_encode(src->encoding());
1893 emit_byte(0xF7);
1894 emit_byte(0xE0 | encode);
1895 }
1897 void Assembler::mulsd(XMMRegister dst, Address src) {
1898 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1899 InstructionMark im(this);
1900 emit_byte(0xF2);
1901 prefix(src, dst);
1902 emit_byte(0x0F);
1903 emit_byte(0x59);
1904 emit_operand(dst, src);
1905 }
1907 void Assembler::mulsd(XMMRegister dst, XMMRegister src) {
1908 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
1909 emit_byte(0xF2);
1910 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1911 emit_byte(0x0F);
1912 emit_byte(0x59);
1913 emit_byte(0xC0 | encode);
1914 }
1916 void Assembler::mulss(XMMRegister dst, Address src) {
1917 NOT_LP64(assert(VM_Version::supports_sse(), ""));
1918 InstructionMark im(this);
1919 emit_byte(0xF3);
1920 prefix(src, dst);
1921 emit_byte(0x0F);
1922 emit_byte(0x59);
1923 emit_operand(dst, src);
1924 }
1926 void Assembler::mulss(XMMRegister dst, XMMRegister src) {
1927 NOT_LP64(assert(VM_Version::supports_sse(), ""));
1928 emit_byte(0xF3);
1929 int encode = prefix_and_encode(dst->encoding(), src->encoding());
1930 emit_byte(0x0F);
1931 emit_byte(0x59);
1932 emit_byte(0xC0 | encode);
1933 }
1935 void Assembler::negl(Register dst) {
1936 int encode = prefix_and_encode(dst->encoding());
1937 emit_byte(0xF7);
1938 emit_byte(0xD8 | encode);
1939 }
1941 void Assembler::nop(int i) {
1942 #ifdef ASSERT
1943 assert(i > 0, " ");
1944 // The fancy nops aren't currently recognized by debuggers making it a
1945 // pain to disassemble code while debugging. If asserts are on clearly
1946 // speed is not an issue so simply use the single byte traditional nop
1947 // to do alignment.
1949 for (; i > 0 ; i--) emit_byte(0x90);
1950 return;
1952 #endif // ASSERT
1954 if (UseAddressNop && VM_Version::is_intel()) {
1955 //
1956 // Using multi-bytes nops "0x0F 0x1F [address]" for Intel
1957 // 1: 0x90
1958 // 2: 0x66 0x90
1959 // 3: 0x66 0x66 0x90 (don't use "0x0F 0x1F 0x00" - need patching safe padding)
1960 // 4: 0x0F 0x1F 0x40 0x00
1961 // 5: 0x0F 0x1F 0x44 0x00 0x00
1962 // 6: 0x66 0x0F 0x1F 0x44 0x00 0x00
1963 // 7: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00
1964 // 8: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
1965 // 9: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
1966 // 10: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
1967 // 11: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
1969 // The rest coding is Intel specific - don't use consecutive address nops
1971 // 12: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90
1972 // 13: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90
1973 // 14: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90
1974 // 15: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90
1976 while(i >= 15) {
1977 // For Intel don't generate consecutive addess nops (mix with regular nops)
1978 i -= 15;
1979 emit_byte(0x66); // size prefix
1980 emit_byte(0x66); // size prefix
1981 emit_byte(0x66); // size prefix
1982 addr_nop_8();
1983 emit_byte(0x66); // size prefix
1984 emit_byte(0x66); // size prefix
1985 emit_byte(0x66); // size prefix
1986 emit_byte(0x90); // nop
1987 }
1988 switch (i) {
1989 case 14:
1990 emit_byte(0x66); // size prefix
1991 case 13:
1992 emit_byte(0x66); // size prefix
1993 case 12:
1994 addr_nop_8();
1995 emit_byte(0x66); // size prefix
1996 emit_byte(0x66); // size prefix
1997 emit_byte(0x66); // size prefix
1998 emit_byte(0x90); // nop
1999 break;
2000 case 11:
2001 emit_byte(0x66); // size prefix
2002 case 10:
2003 emit_byte(0x66); // size prefix
2004 case 9:
2005 emit_byte(0x66); // size prefix
2006 case 8:
2007 addr_nop_8();
2008 break;
2009 case 7:
2010 addr_nop_7();
2011 break;
2012 case 6:
2013 emit_byte(0x66); // size prefix
2014 case 5:
2015 addr_nop_5();
2016 break;
2017 case 4:
2018 addr_nop_4();
2019 break;
2020 case 3:
2021 // Don't use "0x0F 0x1F 0x00" - need patching safe padding
2022 emit_byte(0x66); // size prefix
2023 case 2:
2024 emit_byte(0x66); // size prefix
2025 case 1:
2026 emit_byte(0x90); // nop
2027 break;
2028 default:
2029 assert(i == 0, " ");
2030 }
2031 return;
2032 }
2033 if (UseAddressNop && VM_Version::is_amd()) {
2034 //
2035 // Using multi-bytes nops "0x0F 0x1F [address]" for AMD.
2036 // 1: 0x90
2037 // 2: 0x66 0x90
2038 // 3: 0x66 0x66 0x90 (don't use "0x0F 0x1F 0x00" - need patching safe padding)
2039 // 4: 0x0F 0x1F 0x40 0x00
2040 // 5: 0x0F 0x1F 0x44 0x00 0x00
2041 // 6: 0x66 0x0F 0x1F 0x44 0x00 0x00
2042 // 7: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00
2043 // 8: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2044 // 9: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2045 // 10: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2046 // 11: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2048 // The rest coding is AMD specific - use consecutive address nops
2050 // 12: 0x66 0x0F 0x1F 0x44 0x00 0x00 0x66 0x0F 0x1F 0x44 0x00 0x00
2051 // 13: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 0x66 0x0F 0x1F 0x44 0x00 0x00
2052 // 14: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00
2053 // 15: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00
2054 // 16: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00
2055 // Size prefixes (0x66) are added for larger sizes
2057 while(i >= 22) {
2058 i -= 11;
2059 emit_byte(0x66); // size prefix
2060 emit_byte(0x66); // size prefix
2061 emit_byte(0x66); // size prefix
2062 addr_nop_8();
2063 }
2064 // Generate first nop for size between 21-12
2065 switch (i) {
2066 case 21:
2067 i -= 1;
2068 emit_byte(0x66); // size prefix
2069 case 20:
2070 case 19:
2071 i -= 1;
2072 emit_byte(0x66); // size prefix
2073 case 18:
2074 case 17:
2075 i -= 1;
2076 emit_byte(0x66); // size prefix
2077 case 16:
2078 case 15:
2079 i -= 8;
2080 addr_nop_8();
2081 break;
2082 case 14:
2083 case 13:
2084 i -= 7;
2085 addr_nop_7();
2086 break;
2087 case 12:
2088 i -= 6;
2089 emit_byte(0x66); // size prefix
2090 addr_nop_5();
2091 break;
2092 default:
2093 assert(i < 12, " ");
2094 }
2096 // Generate second nop for size between 11-1
2097 switch (i) {
2098 case 11:
2099 emit_byte(0x66); // size prefix
2100 case 10:
2101 emit_byte(0x66); // size prefix
2102 case 9:
2103 emit_byte(0x66); // size prefix
2104 case 8:
2105 addr_nop_8();
2106 break;
2107 case 7:
2108 addr_nop_7();
2109 break;
2110 case 6:
2111 emit_byte(0x66); // size prefix
2112 case 5:
2113 addr_nop_5();
2114 break;
2115 case 4:
2116 addr_nop_4();
2117 break;
2118 case 3:
2119 // Don't use "0x0F 0x1F 0x00" - need patching safe padding
2120 emit_byte(0x66); // size prefix
2121 case 2:
2122 emit_byte(0x66); // size prefix
2123 case 1:
2124 emit_byte(0x90); // nop
2125 break;
2126 default:
2127 assert(i == 0, " ");
2128 }
2129 return;
2130 }
2132 // Using nops with size prefixes "0x66 0x90".
2133 // From AMD Optimization Guide:
2134 // 1: 0x90
2135 // 2: 0x66 0x90
2136 // 3: 0x66 0x66 0x90
2137 // 4: 0x66 0x66 0x66 0x90
2138 // 5: 0x66 0x66 0x90 0x66 0x90
2139 // 6: 0x66 0x66 0x90 0x66 0x66 0x90
2140 // 7: 0x66 0x66 0x66 0x90 0x66 0x66 0x90
2141 // 8: 0x66 0x66 0x66 0x90 0x66 0x66 0x66 0x90
2142 // 9: 0x66 0x66 0x90 0x66 0x66 0x90 0x66 0x66 0x90
2143 // 10: 0x66 0x66 0x66 0x90 0x66 0x66 0x90 0x66 0x66 0x90
2144 //
2145 while(i > 12) {
2146 i -= 4;
2147 emit_byte(0x66); // size prefix
2148 emit_byte(0x66);
2149 emit_byte(0x66);
2150 emit_byte(0x90); // nop
2151 }
2152 // 1 - 12 nops
2153 if(i > 8) {
2154 if(i > 9) {
2155 i -= 1;
2156 emit_byte(0x66);
2157 }
2158 i -= 3;
2159 emit_byte(0x66);
2160 emit_byte(0x66);
2161 emit_byte(0x90);
2162 }
2163 // 1 - 8 nops
2164 if(i > 4) {
2165 if(i > 6) {
2166 i -= 1;
2167 emit_byte(0x66);
2168 }
2169 i -= 3;
2170 emit_byte(0x66);
2171 emit_byte(0x66);
2172 emit_byte(0x90);
2173 }
2174 switch (i) {
2175 case 4:
2176 emit_byte(0x66);
2177 case 3:
2178 emit_byte(0x66);
2179 case 2:
2180 emit_byte(0x66);
2181 case 1:
2182 emit_byte(0x90);
2183 break;
2184 default:
2185 assert(i == 0, " ");
2186 }
2187 }
2189 void Assembler::notl(Register dst) {
2190 int encode = prefix_and_encode(dst->encoding());
2191 emit_byte(0xF7);
2192 emit_byte(0xD0 | encode );
2193 }
2195 void Assembler::orl(Address dst, int32_t imm32) {
2196 InstructionMark im(this);
2197 prefix(dst);
2198 emit_byte(0x81);
2199 emit_operand(rcx, dst, 4);
2200 emit_long(imm32);
2201 }
2203 void Assembler::orl(Register dst, int32_t imm32) {
2204 prefix(dst);
2205 emit_arith(0x81, 0xC8, dst, imm32);
2206 }
2209 void Assembler::orl(Register dst, Address src) {
2210 InstructionMark im(this);
2211 prefix(src, dst);
2212 emit_byte(0x0B);
2213 emit_operand(dst, src);
2214 }
2217 void Assembler::orl(Register dst, Register src) {
2218 (void) prefix_and_encode(dst->encoding(), src->encoding());
2219 emit_arith(0x0B, 0xC0, dst, src);
2220 }
2222 void Assembler::pcmpestri(XMMRegister dst, Address src, int imm8) {
2223 assert(VM_Version::supports_sse4_2(), "");
2225 InstructionMark im(this);
2226 emit_byte(0x66);
2227 prefix(src, dst);
2228 emit_byte(0x0F);
2229 emit_byte(0x3A);
2230 emit_byte(0x61);
2231 emit_operand(dst, src);
2232 emit_byte(imm8);
2233 }
2235 void Assembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) {
2236 assert(VM_Version::supports_sse4_2(), "");
2238 emit_byte(0x66);
2239 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
2240 emit_byte(0x0F);
2241 emit_byte(0x3A);
2242 emit_byte(0x61);
2243 emit_byte(0xC0 | encode);
2244 emit_byte(imm8);
2245 }
2247 // generic
2248 void Assembler::pop(Register dst) {
2249 int encode = prefix_and_encode(dst->encoding());
2250 emit_byte(0x58 | encode);
2251 }
2253 void Assembler::popcntl(Register dst, Address src) {
2254 assert(VM_Version::supports_popcnt(), "must support");
2255 InstructionMark im(this);
2256 emit_byte(0xF3);
2257 prefix(src, dst);
2258 emit_byte(0x0F);
2259 emit_byte(0xB8);
2260 emit_operand(dst, src);
2261 }
2263 void Assembler::popcntl(Register dst, Register src) {
2264 assert(VM_Version::supports_popcnt(), "must support");
2265 emit_byte(0xF3);
2266 int encode = prefix_and_encode(dst->encoding(), src->encoding());
2267 emit_byte(0x0F);
2268 emit_byte(0xB8);
2269 emit_byte(0xC0 | encode);
2270 }
2272 void Assembler::popf() {
2273 emit_byte(0x9D);
2274 }
2276 #ifndef _LP64 // no 32bit push/pop on amd64
2277 void Assembler::popl(Address dst) {
2278 // NOTE: this will adjust stack by 8byte on 64bits
2279 InstructionMark im(this);
2280 prefix(dst);
2281 emit_byte(0x8F);
2282 emit_operand(rax, dst);
2283 }
2284 #endif
2286 void Assembler::prefetch_prefix(Address src) {
2287 prefix(src);
2288 emit_byte(0x0F);
2289 }
2291 void Assembler::prefetchnta(Address src) {
2292 NOT_LP64(assert(VM_Version::supports_sse2(), "must support"));
2293 InstructionMark im(this);
2294 prefetch_prefix(src);
2295 emit_byte(0x18);
2296 emit_operand(rax, src); // 0, src
2297 }
2299 void Assembler::prefetchr(Address src) {
2300 NOT_LP64(assert(VM_Version::supports_3dnow(), "must support"));
2301 InstructionMark im(this);
2302 prefetch_prefix(src);
2303 emit_byte(0x0D);
2304 emit_operand(rax, src); // 0, src
2305 }
2307 void Assembler::prefetcht0(Address src) {
2308 NOT_LP64(assert(VM_Version::supports_sse(), "must support"));
2309 InstructionMark im(this);
2310 prefetch_prefix(src);
2311 emit_byte(0x18);
2312 emit_operand(rcx, src); // 1, src
2313 }
2315 void Assembler::prefetcht1(Address src) {
2316 NOT_LP64(assert(VM_Version::supports_sse(), "must support"));
2317 InstructionMark im(this);
2318 prefetch_prefix(src);
2319 emit_byte(0x18);
2320 emit_operand(rdx, src); // 2, src
2321 }
2323 void Assembler::prefetcht2(Address src) {
2324 NOT_LP64(assert(VM_Version::supports_sse(), "must support"));
2325 InstructionMark im(this);
2326 prefetch_prefix(src);
2327 emit_byte(0x18);
2328 emit_operand(rbx, src); // 3, src
2329 }
2331 void Assembler::prefetchw(Address src) {
2332 NOT_LP64(assert(VM_Version::supports_3dnow(), "must support"));
2333 InstructionMark im(this);
2334 prefetch_prefix(src);
2335 emit_byte(0x0D);
2336 emit_operand(rcx, src); // 1, src
2337 }
2339 void Assembler::prefix(Prefix p) {
2340 a_byte(p);
2341 }
2343 void Assembler::pshufd(XMMRegister dst, XMMRegister src, int mode) {
2344 assert(isByte(mode), "invalid value");
2345 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2347 emit_byte(0x66);
2348 int encode = prefix_and_encode(dst->encoding(), src->encoding());
2349 emit_byte(0x0F);
2350 emit_byte(0x70);
2351 emit_byte(0xC0 | encode);
2352 emit_byte(mode & 0xFF);
2354 }
2356 void Assembler::pshufd(XMMRegister dst, Address src, int mode) {
2357 assert(isByte(mode), "invalid value");
2358 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2360 InstructionMark im(this);
2361 emit_byte(0x66);
2362 prefix(src, dst);
2363 emit_byte(0x0F);
2364 emit_byte(0x70);
2365 emit_operand(dst, src);
2366 emit_byte(mode & 0xFF);
2367 }
2369 void Assembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) {
2370 assert(isByte(mode), "invalid value");
2371 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2373 emit_byte(0xF2);
2374 int encode = prefix_and_encode(dst->encoding(), src->encoding());
2375 emit_byte(0x0F);
2376 emit_byte(0x70);
2377 emit_byte(0xC0 | encode);
2378 emit_byte(mode & 0xFF);
2379 }
2381 void Assembler::pshuflw(XMMRegister dst, Address src, int mode) {
2382 assert(isByte(mode), "invalid value");
2383 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2385 InstructionMark im(this);
2386 emit_byte(0xF2);
2387 prefix(src, dst); // QQ new
2388 emit_byte(0x0F);
2389 emit_byte(0x70);
2390 emit_operand(dst, src);
2391 emit_byte(mode & 0xFF);
2392 }
2394 void Assembler::psrlq(XMMRegister dst, int shift) {
2395 // HMM Table D-1 says sse2 or mmx
2396 NOT_LP64(assert(VM_Version::supports_sse(), ""));
2398 int encode = prefixq_and_encode(xmm2->encoding(), dst->encoding());
2399 emit_byte(0x66);
2400 emit_byte(0x0F);
2401 emit_byte(0x73);
2402 emit_byte(0xC0 | encode);
2403 emit_byte(shift);
2404 }
2406 void Assembler::ptest(XMMRegister dst, Address src) {
2407 assert(VM_Version::supports_sse4_1(), "");
2409 InstructionMark im(this);
2410 emit_byte(0x66);
2411 prefix(src, dst);
2412 emit_byte(0x0F);
2413 emit_byte(0x38);
2414 emit_byte(0x17);
2415 emit_operand(dst, src);
2416 }
2418 void Assembler::ptest(XMMRegister dst, XMMRegister src) {
2419 assert(VM_Version::supports_sse4_1(), "");
2421 emit_byte(0x66);
2422 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
2423 emit_byte(0x0F);
2424 emit_byte(0x38);
2425 emit_byte(0x17);
2426 emit_byte(0xC0 | encode);
2427 }
2429 void Assembler::punpcklbw(XMMRegister dst, XMMRegister src) {
2430 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2431 emit_byte(0x66);
2432 int encode = prefix_and_encode(dst->encoding(), src->encoding());
2433 emit_byte(0x0F);
2434 emit_byte(0x60);
2435 emit_byte(0xC0 | encode);
2436 }
2438 void Assembler::push(int32_t imm32) {
2439 // in 64bits we push 64bits onto the stack but only
2440 // take a 32bit immediate
2441 emit_byte(0x68);
2442 emit_long(imm32);
2443 }
2445 void Assembler::push(Register src) {
2446 int encode = prefix_and_encode(src->encoding());
2448 emit_byte(0x50 | encode);
2449 }
2451 void Assembler::pushf() {
2452 emit_byte(0x9C);
2453 }
2455 #ifndef _LP64 // no 32bit push/pop on amd64
2456 void Assembler::pushl(Address src) {
2457 // Note this will push 64bit on 64bit
2458 InstructionMark im(this);
2459 prefix(src);
2460 emit_byte(0xFF);
2461 emit_operand(rsi, src);
2462 }
2463 #endif
2465 void Assembler::pxor(XMMRegister dst, Address src) {
2466 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2467 InstructionMark im(this);
2468 emit_byte(0x66);
2469 prefix(src, dst);
2470 emit_byte(0x0F);
2471 emit_byte(0xEF);
2472 emit_operand(dst, src);
2473 }
2475 void Assembler::pxor(XMMRegister dst, XMMRegister src) {
2476 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2477 InstructionMark im(this);
2478 emit_byte(0x66);
2479 int encode = prefix_and_encode(dst->encoding(), src->encoding());
2480 emit_byte(0x0F);
2481 emit_byte(0xEF);
2482 emit_byte(0xC0 | encode);
2483 }
2485 void Assembler::rcll(Register dst, int imm8) {
2486 assert(isShiftCount(imm8), "illegal shift count");
2487 int encode = prefix_and_encode(dst->encoding());
2488 if (imm8 == 1) {
2489 emit_byte(0xD1);
2490 emit_byte(0xD0 | encode);
2491 } else {
2492 emit_byte(0xC1);
2493 emit_byte(0xD0 | encode);
2494 emit_byte(imm8);
2495 }
2496 }
2498 // copies data from [esi] to [edi] using rcx pointer sized words
2499 // generic
2500 void Assembler::rep_mov() {
2501 emit_byte(0xF3);
2502 // MOVSQ
2503 LP64_ONLY(prefix(REX_W));
2504 emit_byte(0xA5);
2505 }
2507 // sets rcx pointer sized words with rax, value at [edi]
2508 // generic
2509 void Assembler::rep_set() { // rep_set
2510 emit_byte(0xF3);
2511 // STOSQ
2512 LP64_ONLY(prefix(REX_W));
2513 emit_byte(0xAB);
2514 }
2516 // scans rcx pointer sized words at [edi] for occurance of rax,
2517 // generic
2518 void Assembler::repne_scan() { // repne_scan
2519 emit_byte(0xF2);
2520 // SCASQ
2521 LP64_ONLY(prefix(REX_W));
2522 emit_byte(0xAF);
2523 }
2525 #ifdef _LP64
2526 // scans rcx 4 byte words at [edi] for occurance of rax,
2527 // generic
2528 void Assembler::repne_scanl() { // repne_scan
2529 emit_byte(0xF2);
2530 // SCASL
2531 emit_byte(0xAF);
2532 }
2533 #endif
2535 void Assembler::ret(int imm16) {
2536 if (imm16 == 0) {
2537 emit_byte(0xC3);
2538 } else {
2539 emit_byte(0xC2);
2540 emit_word(imm16);
2541 }
2542 }
2544 void Assembler::sahf() {
2545 #ifdef _LP64
2546 // Not supported in 64bit mode
2547 ShouldNotReachHere();
2548 #endif
2549 emit_byte(0x9E);
2550 }
2552 void Assembler::sarl(Register dst, int imm8) {
2553 int encode = prefix_and_encode(dst->encoding());
2554 assert(isShiftCount(imm8), "illegal shift count");
2555 if (imm8 == 1) {
2556 emit_byte(0xD1);
2557 emit_byte(0xF8 | encode);
2558 } else {
2559 emit_byte(0xC1);
2560 emit_byte(0xF8 | encode);
2561 emit_byte(imm8);
2562 }
2563 }
2565 void Assembler::sarl(Register dst) {
2566 int encode = prefix_and_encode(dst->encoding());
2567 emit_byte(0xD3);
2568 emit_byte(0xF8 | encode);
2569 }
2571 void Assembler::sbbl(Address dst, int32_t imm32) {
2572 InstructionMark im(this);
2573 prefix(dst);
2574 emit_arith_operand(0x81, rbx, dst, imm32);
2575 }
2577 void Assembler::sbbl(Register dst, int32_t imm32) {
2578 prefix(dst);
2579 emit_arith(0x81, 0xD8, dst, imm32);
2580 }
2583 void Assembler::sbbl(Register dst, Address src) {
2584 InstructionMark im(this);
2585 prefix(src, dst);
2586 emit_byte(0x1B);
2587 emit_operand(dst, src);
2588 }
2590 void Assembler::sbbl(Register dst, Register src) {
2591 (void) prefix_and_encode(dst->encoding(), src->encoding());
2592 emit_arith(0x1B, 0xC0, dst, src);
2593 }
2595 void Assembler::setb(Condition cc, Register dst) {
2596 assert(0 <= cc && cc < 16, "illegal cc");
2597 int encode = prefix_and_encode(dst->encoding(), true);
2598 emit_byte(0x0F);
2599 emit_byte(0x90 | cc);
2600 emit_byte(0xC0 | encode);
2601 }
2603 void Assembler::shll(Register dst, int imm8) {
2604 assert(isShiftCount(imm8), "illegal shift count");
2605 int encode = prefix_and_encode(dst->encoding());
2606 if (imm8 == 1 ) {
2607 emit_byte(0xD1);
2608 emit_byte(0xE0 | encode);
2609 } else {
2610 emit_byte(0xC1);
2611 emit_byte(0xE0 | encode);
2612 emit_byte(imm8);
2613 }
2614 }
2616 void Assembler::shll(Register dst) {
2617 int encode = prefix_and_encode(dst->encoding());
2618 emit_byte(0xD3);
2619 emit_byte(0xE0 | encode);
2620 }
2622 void Assembler::shrl(Register dst, int imm8) {
2623 assert(isShiftCount(imm8), "illegal shift count");
2624 int encode = prefix_and_encode(dst->encoding());
2625 emit_byte(0xC1);
2626 emit_byte(0xE8 | encode);
2627 emit_byte(imm8);
2628 }
2630 void Assembler::shrl(Register dst) {
2631 int encode = prefix_and_encode(dst->encoding());
2632 emit_byte(0xD3);
2633 emit_byte(0xE8 | encode);
2634 }
2636 // copies a single word from [esi] to [edi]
2637 void Assembler::smovl() {
2638 emit_byte(0xA5);
2639 }
2641 void Assembler::sqrtsd(XMMRegister dst, XMMRegister src) {
2642 // HMM Table D-1 says sse2
2643 // NOT_LP64(assert(VM_Version::supports_sse(), ""));
2644 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2645 emit_byte(0xF2);
2646 int encode = prefix_and_encode(dst->encoding(), src->encoding());
2647 emit_byte(0x0F);
2648 emit_byte(0x51);
2649 emit_byte(0xC0 | encode);
2650 }
2652 void Assembler::sqrtsd(XMMRegister dst, Address src) {
2653 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2654 InstructionMark im(this);
2655 emit_byte(0xF2);
2656 prefix(src, dst);
2657 emit_byte(0x0F);
2658 emit_byte(0x51);
2659 emit_operand(dst, src);
2660 }
2662 void Assembler::sqrtss(XMMRegister dst, XMMRegister src) {
2663 // HMM Table D-1 says sse2
2664 // NOT_LP64(assert(VM_Version::supports_sse(), ""));
2665 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2666 emit_byte(0xF3);
2667 int encode = prefix_and_encode(dst->encoding(), src->encoding());
2668 emit_byte(0x0F);
2669 emit_byte(0x51);
2670 emit_byte(0xC0 | encode);
2671 }
2673 void Assembler::sqrtss(XMMRegister dst, Address src) {
2674 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2675 InstructionMark im(this);
2676 emit_byte(0xF3);
2677 prefix(src, dst);
2678 emit_byte(0x0F);
2679 emit_byte(0x51);
2680 emit_operand(dst, src);
2681 }
2683 void Assembler::stmxcsr( Address dst) {
2684 NOT_LP64(assert(VM_Version::supports_sse(), ""));
2685 InstructionMark im(this);
2686 prefix(dst);
2687 emit_byte(0x0F);
2688 emit_byte(0xAE);
2689 emit_operand(as_Register(3), dst);
2690 }
2692 void Assembler::subl(Address dst, int32_t imm32) {
2693 InstructionMark im(this);
2694 prefix(dst);
2695 if (is8bit(imm32)) {
2696 emit_byte(0x83);
2697 emit_operand(rbp, dst, 1);
2698 emit_byte(imm32 & 0xFF);
2699 } else {
2700 emit_byte(0x81);
2701 emit_operand(rbp, dst, 4);
2702 emit_long(imm32);
2703 }
2704 }
2706 void Assembler::subl(Register dst, int32_t imm32) {
2707 prefix(dst);
2708 emit_arith(0x81, 0xE8, dst, imm32);
2709 }
2711 void Assembler::subl(Address dst, Register src) {
2712 InstructionMark im(this);
2713 prefix(dst, src);
2714 emit_byte(0x29);
2715 emit_operand(src, dst);
2716 }
2718 void Assembler::subl(Register dst, Address src) {
2719 InstructionMark im(this);
2720 prefix(src, dst);
2721 emit_byte(0x2B);
2722 emit_operand(dst, src);
2723 }
2725 void Assembler::subl(Register dst, Register src) {
2726 (void) prefix_and_encode(dst->encoding(), src->encoding());
2727 emit_arith(0x2B, 0xC0, dst, src);
2728 }
2730 void Assembler::subsd(XMMRegister dst, XMMRegister src) {
2731 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2732 emit_byte(0xF2);
2733 int encode = prefix_and_encode(dst->encoding(), src->encoding());
2734 emit_byte(0x0F);
2735 emit_byte(0x5C);
2736 emit_byte(0xC0 | encode);
2737 }
2739 void Assembler::subsd(XMMRegister dst, Address src) {
2740 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2741 InstructionMark im(this);
2742 emit_byte(0xF2);
2743 prefix(src, dst);
2744 emit_byte(0x0F);
2745 emit_byte(0x5C);
2746 emit_operand(dst, src);
2747 }
2749 void Assembler::subss(XMMRegister dst, XMMRegister src) {
2750 NOT_LP64(assert(VM_Version::supports_sse(), ""));
2751 emit_byte(0xF3);
2752 int encode = prefix_and_encode(dst->encoding(), src->encoding());
2753 emit_byte(0x0F);
2754 emit_byte(0x5C);
2755 emit_byte(0xC0 | encode);
2756 }
2758 void Assembler::subss(XMMRegister dst, Address src) {
2759 NOT_LP64(assert(VM_Version::supports_sse(), ""));
2760 InstructionMark im(this);
2761 emit_byte(0xF3);
2762 prefix(src, dst);
2763 emit_byte(0x0F);
2764 emit_byte(0x5C);
2765 emit_operand(dst, src);
2766 }
2768 void Assembler::testb(Register dst, int imm8) {
2769 NOT_LP64(assert(dst->has_byte_register(), "must have byte register"));
2770 (void) prefix_and_encode(dst->encoding(), true);
2771 emit_arith_b(0xF6, 0xC0, dst, imm8);
2772 }
2774 void Assembler::testl(Register dst, int32_t imm32) {
2775 // not using emit_arith because test
2776 // doesn't support sign-extension of
2777 // 8bit operands
2778 int encode = dst->encoding();
2779 if (encode == 0) {
2780 emit_byte(0xA9);
2781 } else {
2782 encode = prefix_and_encode(encode);
2783 emit_byte(0xF7);
2784 emit_byte(0xC0 | encode);
2785 }
2786 emit_long(imm32);
2787 }
2789 void Assembler::testl(Register dst, Register src) {
2790 (void) prefix_and_encode(dst->encoding(), src->encoding());
2791 emit_arith(0x85, 0xC0, dst, src);
2792 }
2794 void Assembler::testl(Register dst, Address src) {
2795 InstructionMark im(this);
2796 prefix(src, dst);
2797 emit_byte(0x85);
2798 emit_operand(dst, src);
2799 }
2801 void Assembler::ucomisd(XMMRegister dst, Address src) {
2802 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2803 emit_byte(0x66);
2804 ucomiss(dst, src);
2805 }
2807 void Assembler::ucomisd(XMMRegister dst, XMMRegister src) {
2808 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2809 emit_byte(0x66);
2810 ucomiss(dst, src);
2811 }
2813 void Assembler::ucomiss(XMMRegister dst, Address src) {
2814 NOT_LP64(assert(VM_Version::supports_sse(), ""));
2816 InstructionMark im(this);
2817 prefix(src, dst);
2818 emit_byte(0x0F);
2819 emit_byte(0x2E);
2820 emit_operand(dst, src);
2821 }
2823 void Assembler::ucomiss(XMMRegister dst, XMMRegister src) {
2824 NOT_LP64(assert(VM_Version::supports_sse(), ""));
2825 int encode = prefix_and_encode(dst->encoding(), src->encoding());
2826 emit_byte(0x0F);
2827 emit_byte(0x2E);
2828 emit_byte(0xC0 | encode);
2829 }
2832 void Assembler::xaddl(Address dst, Register src) {
2833 InstructionMark im(this);
2834 prefix(dst, src);
2835 emit_byte(0x0F);
2836 emit_byte(0xC1);
2837 emit_operand(src, dst);
2838 }
2840 void Assembler::xchgl(Register dst, Address src) { // xchg
2841 InstructionMark im(this);
2842 prefix(src, dst);
2843 emit_byte(0x87);
2844 emit_operand(dst, src);
2845 }
2847 void Assembler::xchgl(Register dst, Register src) {
2848 int encode = prefix_and_encode(dst->encoding(), src->encoding());
2849 emit_byte(0x87);
2850 emit_byte(0xc0 | encode);
2851 }
2853 void Assembler::xorl(Register dst, int32_t imm32) {
2854 prefix(dst);
2855 emit_arith(0x81, 0xF0, dst, imm32);
2856 }
2858 void Assembler::xorl(Register dst, Address src) {
2859 InstructionMark im(this);
2860 prefix(src, dst);
2861 emit_byte(0x33);
2862 emit_operand(dst, src);
2863 }
2865 void Assembler::xorl(Register dst, Register src) {
2866 (void) prefix_and_encode(dst->encoding(), src->encoding());
2867 emit_arith(0x33, 0xC0, dst, src);
2868 }
2870 void Assembler::xorpd(XMMRegister dst, XMMRegister src) {
2871 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2872 emit_byte(0x66);
2873 xorps(dst, src);
2874 }
2876 void Assembler::xorpd(XMMRegister dst, Address src) {
2877 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
2878 InstructionMark im(this);
2879 emit_byte(0x66);
2880 prefix(src, dst);
2881 emit_byte(0x0F);
2882 emit_byte(0x57);
2883 emit_operand(dst, src);
2884 }
2887 void Assembler::xorps(XMMRegister dst, XMMRegister src) {
2888 NOT_LP64(assert(VM_Version::supports_sse(), ""));
2889 int encode = prefix_and_encode(dst->encoding(), src->encoding());
2890 emit_byte(0x0F);
2891 emit_byte(0x57);
2892 emit_byte(0xC0 | encode);
2893 }
2895 void Assembler::xorps(XMMRegister dst, Address src) {
2896 NOT_LP64(assert(VM_Version::supports_sse(), ""));
2897 InstructionMark im(this);
2898 prefix(src, dst);
2899 emit_byte(0x0F);
2900 emit_byte(0x57);
2901 emit_operand(dst, src);
2902 }
2904 #ifndef _LP64
2905 // 32bit only pieces of the assembler
2907 void Assembler::cmp_literal32(Register src1, int32_t imm32, RelocationHolder const& rspec) {
2908 // NO PREFIX AS NEVER 64BIT
2909 InstructionMark im(this);
2910 emit_byte(0x81);
2911 emit_byte(0xF8 | src1->encoding());
2912 emit_data(imm32, rspec, 0);
2913 }
2915 void Assembler::cmp_literal32(Address src1, int32_t imm32, RelocationHolder const& rspec) {
2916 // NO PREFIX AS NEVER 64BIT (not even 32bit versions of 64bit regs
2917 InstructionMark im(this);
2918 emit_byte(0x81);
2919 emit_operand(rdi, src1);
2920 emit_data(imm32, rspec, 0);
2921 }
2923 // The 64-bit (32bit platform) cmpxchg compares the value at adr with the contents of rdx:rax,
2924 // and stores rcx:rbx into adr if so; otherwise, the value at adr is loaded
2925 // into rdx:rax. The ZF is set if the compared values were equal, and cleared otherwise.
2926 void Assembler::cmpxchg8(Address adr) {
2927 InstructionMark im(this);
2928 emit_byte(0x0F);
2929 emit_byte(0xc7);
2930 emit_operand(rcx, adr);
2931 }
2933 void Assembler::decl(Register dst) {
2934 // Don't use it directly. Use MacroAssembler::decrementl() instead.
2935 emit_byte(0x48 | dst->encoding());
2936 }
2938 #endif // _LP64
2940 // 64bit typically doesn't use the x87 but needs to for the trig funcs
2942 void Assembler::fabs() {
2943 emit_byte(0xD9);
2944 emit_byte(0xE1);
2945 }
2947 void Assembler::fadd(int i) {
2948 emit_farith(0xD8, 0xC0, i);
2949 }
2951 void Assembler::fadd_d(Address src) {
2952 InstructionMark im(this);
2953 emit_byte(0xDC);
2954 emit_operand32(rax, src);
2955 }
2957 void Assembler::fadd_s(Address src) {
2958 InstructionMark im(this);
2959 emit_byte(0xD8);
2960 emit_operand32(rax, src);
2961 }
2963 void Assembler::fadda(int i) {
2964 emit_farith(0xDC, 0xC0, i);
2965 }
2967 void Assembler::faddp(int i) {
2968 emit_farith(0xDE, 0xC0, i);
2969 }
2971 void Assembler::fchs() {
2972 emit_byte(0xD9);
2973 emit_byte(0xE0);
2974 }
2976 void Assembler::fcom(int i) {
2977 emit_farith(0xD8, 0xD0, i);
2978 }
2980 void Assembler::fcomp(int i) {
2981 emit_farith(0xD8, 0xD8, i);
2982 }
2984 void Assembler::fcomp_d(Address src) {
2985 InstructionMark im(this);
2986 emit_byte(0xDC);
2987 emit_operand32(rbx, src);
2988 }
2990 void Assembler::fcomp_s(Address src) {
2991 InstructionMark im(this);
2992 emit_byte(0xD8);
2993 emit_operand32(rbx, src);
2994 }
2996 void Assembler::fcompp() {
2997 emit_byte(0xDE);
2998 emit_byte(0xD9);
2999 }
3001 void Assembler::fcos() {
3002 emit_byte(0xD9);
3003 emit_byte(0xFF);
3004 }
3006 void Assembler::fdecstp() {
3007 emit_byte(0xD9);
3008 emit_byte(0xF6);
3009 }
3011 void Assembler::fdiv(int i) {
3012 emit_farith(0xD8, 0xF0, i);
3013 }
3015 void Assembler::fdiv_d(Address src) {
3016 InstructionMark im(this);
3017 emit_byte(0xDC);
3018 emit_operand32(rsi, src);
3019 }
3021 void Assembler::fdiv_s(Address src) {
3022 InstructionMark im(this);
3023 emit_byte(0xD8);
3024 emit_operand32(rsi, src);
3025 }
3027 void Assembler::fdiva(int i) {
3028 emit_farith(0xDC, 0xF8, i);
3029 }
3031 // Note: The Intel manual (Pentium Processor User's Manual, Vol.3, 1994)
3032 // is erroneous for some of the floating-point instructions below.
3034 void Assembler::fdivp(int i) {
3035 emit_farith(0xDE, 0xF8, i); // ST(0) <- ST(0) / ST(1) and pop (Intel manual wrong)
3036 }
3038 void Assembler::fdivr(int i) {
3039 emit_farith(0xD8, 0xF8, i);
3040 }
3042 void Assembler::fdivr_d(Address src) {
3043 InstructionMark im(this);
3044 emit_byte(0xDC);
3045 emit_operand32(rdi, src);
3046 }
3048 void Assembler::fdivr_s(Address src) {
3049 InstructionMark im(this);
3050 emit_byte(0xD8);
3051 emit_operand32(rdi, src);
3052 }
3054 void Assembler::fdivra(int i) {
3055 emit_farith(0xDC, 0xF0, i);
3056 }
3058 void Assembler::fdivrp(int i) {
3059 emit_farith(0xDE, 0xF0, i); // ST(0) <- ST(1) / ST(0) and pop (Intel manual wrong)
3060 }
3062 void Assembler::ffree(int i) {
3063 emit_farith(0xDD, 0xC0, i);
3064 }
3066 void Assembler::fild_d(Address adr) {
3067 InstructionMark im(this);
3068 emit_byte(0xDF);
3069 emit_operand32(rbp, adr);
3070 }
3072 void Assembler::fild_s(Address adr) {
3073 InstructionMark im(this);
3074 emit_byte(0xDB);
3075 emit_operand32(rax, adr);
3076 }
3078 void Assembler::fincstp() {
3079 emit_byte(0xD9);
3080 emit_byte(0xF7);
3081 }
3083 void Assembler::finit() {
3084 emit_byte(0x9B);
3085 emit_byte(0xDB);
3086 emit_byte(0xE3);
3087 }
3089 void Assembler::fist_s(Address adr) {
3090 InstructionMark im(this);
3091 emit_byte(0xDB);
3092 emit_operand32(rdx, adr);
3093 }
3095 void Assembler::fistp_d(Address adr) {
3096 InstructionMark im(this);
3097 emit_byte(0xDF);
3098 emit_operand32(rdi, adr);
3099 }
3101 void Assembler::fistp_s(Address adr) {
3102 InstructionMark im(this);
3103 emit_byte(0xDB);
3104 emit_operand32(rbx, adr);
3105 }
3107 void Assembler::fld1() {
3108 emit_byte(0xD9);
3109 emit_byte(0xE8);
3110 }
3112 void Assembler::fld_d(Address adr) {
3113 InstructionMark im(this);
3114 emit_byte(0xDD);
3115 emit_operand32(rax, adr);
3116 }
3118 void Assembler::fld_s(Address adr) {
3119 InstructionMark im(this);
3120 emit_byte(0xD9);
3121 emit_operand32(rax, adr);
3122 }
3125 void Assembler::fld_s(int index) {
3126 emit_farith(0xD9, 0xC0, index);
3127 }
3129 void Assembler::fld_x(Address adr) {
3130 InstructionMark im(this);
3131 emit_byte(0xDB);
3132 emit_operand32(rbp, adr);
3133 }
3135 void Assembler::fldcw(Address src) {
3136 InstructionMark im(this);
3137 emit_byte(0xd9);
3138 emit_operand32(rbp, src);
3139 }
3141 void Assembler::fldenv(Address src) {
3142 InstructionMark im(this);
3143 emit_byte(0xD9);
3144 emit_operand32(rsp, src);
3145 }
3147 void Assembler::fldlg2() {
3148 emit_byte(0xD9);
3149 emit_byte(0xEC);
3150 }
3152 void Assembler::fldln2() {
3153 emit_byte(0xD9);
3154 emit_byte(0xED);
3155 }
3157 void Assembler::fldz() {
3158 emit_byte(0xD9);
3159 emit_byte(0xEE);
3160 }
3162 void Assembler::flog() {
3163 fldln2();
3164 fxch();
3165 fyl2x();
3166 }
3168 void Assembler::flog10() {
3169 fldlg2();
3170 fxch();
3171 fyl2x();
3172 }
3174 void Assembler::fmul(int i) {
3175 emit_farith(0xD8, 0xC8, i);
3176 }
3178 void Assembler::fmul_d(Address src) {
3179 InstructionMark im(this);
3180 emit_byte(0xDC);
3181 emit_operand32(rcx, src);
3182 }
3184 void Assembler::fmul_s(Address src) {
3185 InstructionMark im(this);
3186 emit_byte(0xD8);
3187 emit_operand32(rcx, src);
3188 }
3190 void Assembler::fmula(int i) {
3191 emit_farith(0xDC, 0xC8, i);
3192 }
3194 void Assembler::fmulp(int i) {
3195 emit_farith(0xDE, 0xC8, i);
3196 }
3198 void Assembler::fnsave(Address dst) {
3199 InstructionMark im(this);
3200 emit_byte(0xDD);
3201 emit_operand32(rsi, dst);
3202 }
3204 void Assembler::fnstcw(Address src) {
3205 InstructionMark im(this);
3206 emit_byte(0x9B);
3207 emit_byte(0xD9);
3208 emit_operand32(rdi, src);
3209 }
3211 void Assembler::fnstsw_ax() {
3212 emit_byte(0xdF);
3213 emit_byte(0xE0);
3214 }
3216 void Assembler::fprem() {
3217 emit_byte(0xD9);
3218 emit_byte(0xF8);
3219 }
3221 void Assembler::fprem1() {
3222 emit_byte(0xD9);
3223 emit_byte(0xF5);
3224 }
3226 void Assembler::frstor(Address src) {
3227 InstructionMark im(this);
3228 emit_byte(0xDD);
3229 emit_operand32(rsp, src);
3230 }
3232 void Assembler::fsin() {
3233 emit_byte(0xD9);
3234 emit_byte(0xFE);
3235 }
3237 void Assembler::fsqrt() {
3238 emit_byte(0xD9);
3239 emit_byte(0xFA);
3240 }
3242 void Assembler::fst_d(Address adr) {
3243 InstructionMark im(this);
3244 emit_byte(0xDD);
3245 emit_operand32(rdx, adr);
3246 }
3248 void Assembler::fst_s(Address adr) {
3249 InstructionMark im(this);
3250 emit_byte(0xD9);
3251 emit_operand32(rdx, adr);
3252 }
3254 void Assembler::fstp_d(Address adr) {
3255 InstructionMark im(this);
3256 emit_byte(0xDD);
3257 emit_operand32(rbx, adr);
3258 }
3260 void Assembler::fstp_d(int index) {
3261 emit_farith(0xDD, 0xD8, index);
3262 }
3264 void Assembler::fstp_s(Address adr) {
3265 InstructionMark im(this);
3266 emit_byte(0xD9);
3267 emit_operand32(rbx, adr);
3268 }
3270 void Assembler::fstp_x(Address adr) {
3271 InstructionMark im(this);
3272 emit_byte(0xDB);
3273 emit_operand32(rdi, adr);
3274 }
3276 void Assembler::fsub(int i) {
3277 emit_farith(0xD8, 0xE0, i);
3278 }
3280 void Assembler::fsub_d(Address src) {
3281 InstructionMark im(this);
3282 emit_byte(0xDC);
3283 emit_operand32(rsp, src);
3284 }
3286 void Assembler::fsub_s(Address src) {
3287 InstructionMark im(this);
3288 emit_byte(0xD8);
3289 emit_operand32(rsp, src);
3290 }
3292 void Assembler::fsuba(int i) {
3293 emit_farith(0xDC, 0xE8, i);
3294 }
3296 void Assembler::fsubp(int i) {
3297 emit_farith(0xDE, 0xE8, i); // ST(0) <- ST(0) - ST(1) and pop (Intel manual wrong)
3298 }
3300 void Assembler::fsubr(int i) {
3301 emit_farith(0xD8, 0xE8, i);
3302 }
3304 void Assembler::fsubr_d(Address src) {
3305 InstructionMark im(this);
3306 emit_byte(0xDC);
3307 emit_operand32(rbp, src);
3308 }
3310 void Assembler::fsubr_s(Address src) {
3311 InstructionMark im(this);
3312 emit_byte(0xD8);
3313 emit_operand32(rbp, src);
3314 }
3316 void Assembler::fsubra(int i) {
3317 emit_farith(0xDC, 0xE0, i);
3318 }
3320 void Assembler::fsubrp(int i) {
3321 emit_farith(0xDE, 0xE0, i); // ST(0) <- ST(1) - ST(0) and pop (Intel manual wrong)
3322 }
3324 void Assembler::ftan() {
3325 emit_byte(0xD9);
3326 emit_byte(0xF2);
3327 emit_byte(0xDD);
3328 emit_byte(0xD8);
3329 }
3331 void Assembler::ftst() {
3332 emit_byte(0xD9);
3333 emit_byte(0xE4);
3334 }
3336 void Assembler::fucomi(int i) {
3337 // make sure the instruction is supported (introduced for P6, together with cmov)
3338 guarantee(VM_Version::supports_cmov(), "illegal instruction");
3339 emit_farith(0xDB, 0xE8, i);
3340 }
3342 void Assembler::fucomip(int i) {
3343 // make sure the instruction is supported (introduced for P6, together with cmov)
3344 guarantee(VM_Version::supports_cmov(), "illegal instruction");
3345 emit_farith(0xDF, 0xE8, i);
3346 }
3348 void Assembler::fwait() {
3349 emit_byte(0x9B);
3350 }
3352 void Assembler::fxch(int i) {
3353 emit_farith(0xD9, 0xC8, i);
3354 }
3356 void Assembler::fyl2x() {
3357 emit_byte(0xD9);
3358 emit_byte(0xF1);
3359 }
3362 #ifndef _LP64
3364 void Assembler::incl(Register dst) {
3365 // Don't use it directly. Use MacroAssembler::incrementl() instead.
3366 emit_byte(0x40 | dst->encoding());
3367 }
3369 void Assembler::lea(Register dst, Address src) {
3370 leal(dst, src);
3371 }
3373 void Assembler::mov_literal32(Address dst, int32_t imm32, RelocationHolder const& rspec) {
3374 InstructionMark im(this);
3375 emit_byte(0xC7);
3376 emit_operand(rax, dst);
3377 emit_data((int)imm32, rspec, 0);
3378 }
3380 void Assembler::mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec) {
3381 InstructionMark im(this);
3382 int encode = prefix_and_encode(dst->encoding());
3383 emit_byte(0xB8 | encode);
3384 emit_data((int)imm32, rspec, 0);
3385 }
3387 void Assembler::popa() { // 32bit
3388 emit_byte(0x61);
3389 }
3391 void Assembler::push_literal32(int32_t imm32, RelocationHolder const& rspec) {
3392 InstructionMark im(this);
3393 emit_byte(0x68);
3394 emit_data(imm32, rspec, 0);
3395 }
3397 void Assembler::pusha() { // 32bit
3398 emit_byte(0x60);
3399 }
3401 void Assembler::set_byte_if_not_zero(Register dst) {
3402 emit_byte(0x0F);
3403 emit_byte(0x95);
3404 emit_byte(0xE0 | dst->encoding());
3405 }
3407 void Assembler::shldl(Register dst, Register src) {
3408 emit_byte(0x0F);
3409 emit_byte(0xA5);
3410 emit_byte(0xC0 | src->encoding() << 3 | dst->encoding());
3411 }
3413 void Assembler::shrdl(Register dst, Register src) {
3414 emit_byte(0x0F);
3415 emit_byte(0xAD);
3416 emit_byte(0xC0 | src->encoding() << 3 | dst->encoding());
3417 }
3419 #else // LP64
3421 void Assembler::set_byte_if_not_zero(Register dst) {
3422 int enc = prefix_and_encode(dst->encoding(), true);
3423 emit_byte(0x0F);
3424 emit_byte(0x95);
3425 emit_byte(0xE0 | enc);
3426 }
3428 // 64bit only pieces of the assembler
3429 // This should only be used by 64bit instructions that can use rip-relative
3430 // it cannot be used by instructions that want an immediate value.
3432 bool Assembler::reachable(AddressLiteral adr) {
3433 int64_t disp;
3434 // None will force a 64bit literal to the code stream. Likely a placeholder
3435 // for something that will be patched later and we need to certain it will
3436 // always be reachable.
3437 if (adr.reloc() == relocInfo::none) {
3438 return false;
3439 }
3440 if (adr.reloc() == relocInfo::internal_word_type) {
3441 // This should be rip relative and easily reachable.
3442 return true;
3443 }
3444 if (adr.reloc() == relocInfo::virtual_call_type ||
3445 adr.reloc() == relocInfo::opt_virtual_call_type ||
3446 adr.reloc() == relocInfo::static_call_type ||
3447 adr.reloc() == relocInfo::static_stub_type ) {
3448 // This should be rip relative within the code cache and easily
3449 // reachable until we get huge code caches. (At which point
3450 // ic code is going to have issues).
3451 return true;
3452 }
3453 if (adr.reloc() != relocInfo::external_word_type &&
3454 adr.reloc() != relocInfo::poll_return_type && // these are really external_word but need special
3455 adr.reloc() != relocInfo::poll_type && // relocs to identify them
3456 adr.reloc() != relocInfo::runtime_call_type ) {
3457 return false;
3458 }
3460 // Stress the correction code
3461 if (ForceUnreachable) {
3462 // Must be runtimecall reloc, see if it is in the codecache
3463 // Flipping stuff in the codecache to be unreachable causes issues
3464 // with things like inline caches where the additional instructions
3465 // are not handled.
3466 if (CodeCache::find_blob(adr._target) == NULL) {
3467 return false;
3468 }
3469 }
3470 // For external_word_type/runtime_call_type if it is reachable from where we
3471 // are now (possibly a temp buffer) and where we might end up
3472 // anywhere in the codeCache then we are always reachable.
3473 // This would have to change if we ever save/restore shared code
3474 // to be more pessimistic.
3476 disp = (int64_t)adr._target - ((int64_t)CodeCache::low_bound() + sizeof(int));
3477 if (!is_simm32(disp)) return false;
3478 disp = (int64_t)adr._target - ((int64_t)CodeCache::high_bound() + sizeof(int));
3479 if (!is_simm32(disp)) return false;
3481 disp = (int64_t)adr._target - ((int64_t)_code_pos + sizeof(int));
3483 // Because rip relative is a disp + address_of_next_instruction and we
3484 // don't know the value of address_of_next_instruction we apply a fudge factor
3485 // to make sure we will be ok no matter the size of the instruction we get placed into.
3486 // We don't have to fudge the checks above here because they are already worst case.
3488 // 12 == override/rex byte, opcode byte, rm byte, sib byte, a 4-byte disp , 4-byte literal
3489 // + 4 because better safe than sorry.
3490 const int fudge = 12 + 4;
3491 if (disp < 0) {
3492 disp -= fudge;
3493 } else {
3494 disp += fudge;
3495 }
3496 return is_simm32(disp);
3497 }
3499 void Assembler::emit_data64(jlong data,
3500 relocInfo::relocType rtype,
3501 int format) {
3502 if (rtype == relocInfo::none) {
3503 emit_long64(data);
3504 } else {
3505 emit_data64(data, Relocation::spec_simple(rtype), format);
3506 }
3507 }
3509 void Assembler::emit_data64(jlong data,
3510 RelocationHolder const& rspec,
3511 int format) {
3512 assert(imm_operand == 0, "default format must be immediate in this file");
3513 assert(imm_operand == format, "must be immediate");
3514 assert(inst_mark() != NULL, "must be inside InstructionMark");
3515 // Do not use AbstractAssembler::relocate, which is not intended for
3516 // embedded words. Instead, relocate to the enclosing instruction.
3517 code_section()->relocate(inst_mark(), rspec, format);
3518 #ifdef ASSERT
3519 check_relocation(rspec, format);
3520 #endif
3521 emit_long64(data);
3522 }
3524 int Assembler::prefix_and_encode(int reg_enc, bool byteinst) {
3525 if (reg_enc >= 8) {
3526 prefix(REX_B);
3527 reg_enc -= 8;
3528 } else if (byteinst && reg_enc >= 4) {
3529 prefix(REX);
3530 }
3531 return reg_enc;
3532 }
3534 int Assembler::prefixq_and_encode(int reg_enc) {
3535 if (reg_enc < 8) {
3536 prefix(REX_W);
3537 } else {
3538 prefix(REX_WB);
3539 reg_enc -= 8;
3540 }
3541 return reg_enc;
3542 }
3544 int Assembler::prefix_and_encode(int dst_enc, int src_enc, bool byteinst) {
3545 if (dst_enc < 8) {
3546 if (src_enc >= 8) {
3547 prefix(REX_B);
3548 src_enc -= 8;
3549 } else if (byteinst && src_enc >= 4) {
3550 prefix(REX);
3551 }
3552 } else {
3553 if (src_enc < 8) {
3554 prefix(REX_R);
3555 } else {
3556 prefix(REX_RB);
3557 src_enc -= 8;
3558 }
3559 dst_enc -= 8;
3560 }
3561 return dst_enc << 3 | src_enc;
3562 }
3564 int Assembler::prefixq_and_encode(int dst_enc, int src_enc) {
3565 if (dst_enc < 8) {
3566 if (src_enc < 8) {
3567 prefix(REX_W);
3568 } else {
3569 prefix(REX_WB);
3570 src_enc -= 8;
3571 }
3572 } else {
3573 if (src_enc < 8) {
3574 prefix(REX_WR);
3575 } else {
3576 prefix(REX_WRB);
3577 src_enc -= 8;
3578 }
3579 dst_enc -= 8;
3580 }
3581 return dst_enc << 3 | src_enc;
3582 }
3584 void Assembler::prefix(Register reg) {
3585 if (reg->encoding() >= 8) {
3586 prefix(REX_B);
3587 }
3588 }
3590 void Assembler::prefix(Address adr) {
3591 if (adr.base_needs_rex()) {
3592 if (adr.index_needs_rex()) {
3593 prefix(REX_XB);
3594 } else {
3595 prefix(REX_B);
3596 }
3597 } else {
3598 if (adr.index_needs_rex()) {
3599 prefix(REX_X);
3600 }
3601 }
3602 }
3604 void Assembler::prefixq(Address adr) {
3605 if (adr.base_needs_rex()) {
3606 if (adr.index_needs_rex()) {
3607 prefix(REX_WXB);
3608 } else {
3609 prefix(REX_WB);
3610 }
3611 } else {
3612 if (adr.index_needs_rex()) {
3613 prefix(REX_WX);
3614 } else {
3615 prefix(REX_W);
3616 }
3617 }
3618 }
3621 void Assembler::prefix(Address adr, Register reg, bool byteinst) {
3622 if (reg->encoding() < 8) {
3623 if (adr.base_needs_rex()) {
3624 if (adr.index_needs_rex()) {
3625 prefix(REX_XB);
3626 } else {
3627 prefix(REX_B);
3628 }
3629 } else {
3630 if (adr.index_needs_rex()) {
3631 prefix(REX_X);
3632 } else if (reg->encoding() >= 4 ) {
3633 prefix(REX);
3634 }
3635 }
3636 } else {
3637 if (adr.base_needs_rex()) {
3638 if (adr.index_needs_rex()) {
3639 prefix(REX_RXB);
3640 } else {
3641 prefix(REX_RB);
3642 }
3643 } else {
3644 if (adr.index_needs_rex()) {
3645 prefix(REX_RX);
3646 } else {
3647 prefix(REX_R);
3648 }
3649 }
3650 }
3651 }
3653 void Assembler::prefixq(Address adr, Register src) {
3654 if (src->encoding() < 8) {
3655 if (adr.base_needs_rex()) {
3656 if (adr.index_needs_rex()) {
3657 prefix(REX_WXB);
3658 } else {
3659 prefix(REX_WB);
3660 }
3661 } else {
3662 if (adr.index_needs_rex()) {
3663 prefix(REX_WX);
3664 } else {
3665 prefix(REX_W);
3666 }
3667 }
3668 } else {
3669 if (adr.base_needs_rex()) {
3670 if (adr.index_needs_rex()) {
3671 prefix(REX_WRXB);
3672 } else {
3673 prefix(REX_WRB);
3674 }
3675 } else {
3676 if (adr.index_needs_rex()) {
3677 prefix(REX_WRX);
3678 } else {
3679 prefix(REX_WR);
3680 }
3681 }
3682 }
3683 }
3685 void Assembler::prefix(Address adr, XMMRegister reg) {
3686 if (reg->encoding() < 8) {
3687 if (adr.base_needs_rex()) {
3688 if (adr.index_needs_rex()) {
3689 prefix(REX_XB);
3690 } else {
3691 prefix(REX_B);
3692 }
3693 } else {
3694 if (adr.index_needs_rex()) {
3695 prefix(REX_X);
3696 }
3697 }
3698 } else {
3699 if (adr.base_needs_rex()) {
3700 if (adr.index_needs_rex()) {
3701 prefix(REX_RXB);
3702 } else {
3703 prefix(REX_RB);
3704 }
3705 } else {
3706 if (adr.index_needs_rex()) {
3707 prefix(REX_RX);
3708 } else {
3709 prefix(REX_R);
3710 }
3711 }
3712 }
3713 }
3715 void Assembler::adcq(Register dst, int32_t imm32) {
3716 (void) prefixq_and_encode(dst->encoding());
3717 emit_arith(0x81, 0xD0, dst, imm32);
3718 }
3720 void Assembler::adcq(Register dst, Address src) {
3721 InstructionMark im(this);
3722 prefixq(src, dst);
3723 emit_byte(0x13);
3724 emit_operand(dst, src);
3725 }
3727 void Assembler::adcq(Register dst, Register src) {
3728 (int) prefixq_and_encode(dst->encoding(), src->encoding());
3729 emit_arith(0x13, 0xC0, dst, src);
3730 }
3732 void Assembler::addq(Address dst, int32_t imm32) {
3733 InstructionMark im(this);
3734 prefixq(dst);
3735 emit_arith_operand(0x81, rax, dst,imm32);
3736 }
3738 void Assembler::addq(Address dst, Register src) {
3739 InstructionMark im(this);
3740 prefixq(dst, src);
3741 emit_byte(0x01);
3742 emit_operand(src, dst);
3743 }
3745 void Assembler::addq(Register dst, int32_t imm32) {
3746 (void) prefixq_and_encode(dst->encoding());
3747 emit_arith(0x81, 0xC0, dst, imm32);
3748 }
3750 void Assembler::addq(Register dst, Address src) {
3751 InstructionMark im(this);
3752 prefixq(src, dst);
3753 emit_byte(0x03);
3754 emit_operand(dst, src);
3755 }
3757 void Assembler::addq(Register dst, Register src) {
3758 (void) prefixq_and_encode(dst->encoding(), src->encoding());
3759 emit_arith(0x03, 0xC0, dst, src);
3760 }
3762 void Assembler::andq(Register dst, int32_t imm32) {
3763 (void) prefixq_and_encode(dst->encoding());
3764 emit_arith(0x81, 0xE0, dst, imm32);
3765 }
3767 void Assembler::andq(Register dst, Address src) {
3768 InstructionMark im(this);
3769 prefixq(src, dst);
3770 emit_byte(0x23);
3771 emit_operand(dst, src);
3772 }
3774 void Assembler::andq(Register dst, Register src) {
3775 (int) prefixq_and_encode(dst->encoding(), src->encoding());
3776 emit_arith(0x23, 0xC0, dst, src);
3777 }
3779 void Assembler::bsfq(Register dst, Register src) {
3780 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3781 emit_byte(0x0F);
3782 emit_byte(0xBC);
3783 emit_byte(0xC0 | encode);
3784 }
3786 void Assembler::bsrq(Register dst, Register src) {
3787 assert(!VM_Version::supports_lzcnt(), "encoding is treated as LZCNT");
3788 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3789 emit_byte(0x0F);
3790 emit_byte(0xBD);
3791 emit_byte(0xC0 | encode);
3792 }
3794 void Assembler::bswapq(Register reg) {
3795 int encode = prefixq_and_encode(reg->encoding());
3796 emit_byte(0x0F);
3797 emit_byte(0xC8 | encode);
3798 }
3800 void Assembler::cdqq() {
3801 prefix(REX_W);
3802 emit_byte(0x99);
3803 }
3805 void Assembler::clflush(Address adr) {
3806 prefix(adr);
3807 emit_byte(0x0F);
3808 emit_byte(0xAE);
3809 emit_operand(rdi, adr);
3810 }
3812 void Assembler::cmovq(Condition cc, Register dst, Register src) {
3813 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3814 emit_byte(0x0F);
3815 emit_byte(0x40 | cc);
3816 emit_byte(0xC0 | encode);
3817 }
3819 void Assembler::cmovq(Condition cc, Register dst, Address src) {
3820 InstructionMark im(this);
3821 prefixq(src, dst);
3822 emit_byte(0x0F);
3823 emit_byte(0x40 | cc);
3824 emit_operand(dst, src);
3825 }
3827 void Assembler::cmpq(Address dst, int32_t imm32) {
3828 InstructionMark im(this);
3829 prefixq(dst);
3830 emit_byte(0x81);
3831 emit_operand(rdi, dst, 4);
3832 emit_long(imm32);
3833 }
3835 void Assembler::cmpq(Register dst, int32_t imm32) {
3836 (void) prefixq_and_encode(dst->encoding());
3837 emit_arith(0x81, 0xF8, dst, imm32);
3838 }
3840 void Assembler::cmpq(Address dst, Register src) {
3841 InstructionMark im(this);
3842 prefixq(dst, src);
3843 emit_byte(0x3B);
3844 emit_operand(src, dst);
3845 }
3847 void Assembler::cmpq(Register dst, Register src) {
3848 (void) prefixq_and_encode(dst->encoding(), src->encoding());
3849 emit_arith(0x3B, 0xC0, dst, src);
3850 }
3852 void Assembler::cmpq(Register dst, Address src) {
3853 InstructionMark im(this);
3854 prefixq(src, dst);
3855 emit_byte(0x3B);
3856 emit_operand(dst, src);
3857 }
3859 void Assembler::cmpxchgq(Register reg, Address adr) {
3860 InstructionMark im(this);
3861 prefixq(adr, reg);
3862 emit_byte(0x0F);
3863 emit_byte(0xB1);
3864 emit_operand(reg, adr);
3865 }
3867 void Assembler::cvtsi2sdq(XMMRegister dst, Register src) {
3868 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
3869 emit_byte(0xF2);
3870 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3871 emit_byte(0x0F);
3872 emit_byte(0x2A);
3873 emit_byte(0xC0 | encode);
3874 }
3876 void Assembler::cvtsi2ssq(XMMRegister dst, Register src) {
3877 NOT_LP64(assert(VM_Version::supports_sse(), ""));
3878 emit_byte(0xF3);
3879 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3880 emit_byte(0x0F);
3881 emit_byte(0x2A);
3882 emit_byte(0xC0 | encode);
3883 }
3885 void Assembler::cvttsd2siq(Register dst, XMMRegister src) {
3886 NOT_LP64(assert(VM_Version::supports_sse2(), ""));
3887 emit_byte(0xF2);
3888 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3889 emit_byte(0x0F);
3890 emit_byte(0x2C);
3891 emit_byte(0xC0 | encode);
3892 }
3894 void Assembler::cvttss2siq(Register dst, XMMRegister src) {
3895 NOT_LP64(assert(VM_Version::supports_sse(), ""));
3896 emit_byte(0xF3);
3897 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3898 emit_byte(0x0F);
3899 emit_byte(0x2C);
3900 emit_byte(0xC0 | encode);
3901 }
3903 void Assembler::decl(Register dst) {
3904 // Don't use it directly. Use MacroAssembler::decrementl() instead.
3905 // Use two-byte form (one-byte form is a REX prefix in 64-bit mode)
3906 int encode = prefix_and_encode(dst->encoding());
3907 emit_byte(0xFF);
3908 emit_byte(0xC8 | encode);
3909 }
3911 void Assembler::decq(Register dst) {
3912 // Don't use it directly. Use MacroAssembler::decrementq() instead.
3913 // Use two-byte form (one-byte from is a REX prefix in 64-bit mode)
3914 int encode = prefixq_and_encode(dst->encoding());
3915 emit_byte(0xFF);
3916 emit_byte(0xC8 | encode);
3917 }
3919 void Assembler::decq(Address dst) {
3920 // Don't use it directly. Use MacroAssembler::decrementq() instead.
3921 InstructionMark im(this);
3922 prefixq(dst);
3923 emit_byte(0xFF);
3924 emit_operand(rcx, dst);
3925 }
3927 void Assembler::fxrstor(Address src) {
3928 prefixq(src);
3929 emit_byte(0x0F);
3930 emit_byte(0xAE);
3931 emit_operand(as_Register(1), src);
3932 }
3934 void Assembler::fxsave(Address dst) {
3935 prefixq(dst);
3936 emit_byte(0x0F);
3937 emit_byte(0xAE);
3938 emit_operand(as_Register(0), dst);
3939 }
3941 void Assembler::idivq(Register src) {
3942 int encode = prefixq_and_encode(src->encoding());
3943 emit_byte(0xF7);
3944 emit_byte(0xF8 | encode);
3945 }
3947 void Assembler::imulq(Register dst, Register src) {
3948 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3949 emit_byte(0x0F);
3950 emit_byte(0xAF);
3951 emit_byte(0xC0 | encode);
3952 }
3954 void Assembler::imulq(Register dst, Register src, int value) {
3955 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
3956 if (is8bit(value)) {
3957 emit_byte(0x6B);
3958 emit_byte(0xC0 | encode);
3959 emit_byte(value & 0xFF);
3960 } else {
3961 emit_byte(0x69);
3962 emit_byte(0xC0 | encode);
3963 emit_long(value);
3964 }
3965 }
3967 void Assembler::incl(Register dst) {
3968 // Don't use it directly. Use MacroAssembler::incrementl() instead.
3969 // Use two-byte form (one-byte from is a REX prefix in 64-bit mode)
3970 int encode = prefix_and_encode(dst->encoding());
3971 emit_byte(0xFF);
3972 emit_byte(0xC0 | encode);
3973 }
3975 void Assembler::incq(Register dst) {
3976 // Don't use it directly. Use MacroAssembler::incrementq() instead.
3977 // Use two-byte form (one-byte from is a REX prefix in 64-bit mode)
3978 int encode = prefixq_and_encode(dst->encoding());
3979 emit_byte(0xFF);
3980 emit_byte(0xC0 | encode);
3981 }
3983 void Assembler::incq(Address dst) {
3984 // Don't use it directly. Use MacroAssembler::incrementq() instead.
3985 InstructionMark im(this);
3986 prefixq(dst);
3987 emit_byte(0xFF);
3988 emit_operand(rax, dst);
3989 }
3991 void Assembler::lea(Register dst, Address src) {
3992 leaq(dst, src);
3993 }
3995 void Assembler::leaq(Register dst, Address src) {
3996 InstructionMark im(this);
3997 prefixq(src, dst);
3998 emit_byte(0x8D);
3999 emit_operand(dst, src);
4000 }
4002 void Assembler::mov64(Register dst, int64_t imm64) {
4003 InstructionMark im(this);
4004 int encode = prefixq_and_encode(dst->encoding());
4005 emit_byte(0xB8 | encode);
4006 emit_long64(imm64);
4007 }
4009 void Assembler::mov_literal64(Register dst, intptr_t imm64, RelocationHolder const& rspec) {
4010 InstructionMark im(this);
4011 int encode = prefixq_and_encode(dst->encoding());
4012 emit_byte(0xB8 | encode);
4013 emit_data64(imm64, rspec);
4014 }
4016 void Assembler::mov_narrow_oop(Register dst, int32_t imm32, RelocationHolder const& rspec) {
4017 InstructionMark im(this);
4018 int encode = prefix_and_encode(dst->encoding());
4019 emit_byte(0xB8 | encode);
4020 emit_data((int)imm32, rspec, narrow_oop_operand);
4021 }
4023 void Assembler::mov_narrow_oop(Address dst, int32_t imm32, RelocationHolder const& rspec) {
4024 InstructionMark im(this);
4025 prefix(dst);
4026 emit_byte(0xC7);
4027 emit_operand(rax, dst, 4);
4028 emit_data((int)imm32, rspec, narrow_oop_operand);
4029 }
4031 void Assembler::cmp_narrow_oop(Register src1, int32_t imm32, RelocationHolder const& rspec) {
4032 InstructionMark im(this);
4033 int encode = prefix_and_encode(src1->encoding());
4034 emit_byte(0x81);
4035 emit_byte(0xF8 | encode);
4036 emit_data((int)imm32, rspec, narrow_oop_operand);
4037 }
4039 void Assembler::cmp_narrow_oop(Address src1, int32_t imm32, RelocationHolder const& rspec) {
4040 InstructionMark im(this);
4041 prefix(src1);
4042 emit_byte(0x81);
4043 emit_operand(rax, src1, 4);
4044 emit_data((int)imm32, rspec, narrow_oop_operand);
4045 }
4047 void Assembler::lzcntq(Register dst, Register src) {
4048 assert(VM_Version::supports_lzcnt(), "encoding is treated as BSR");
4049 emit_byte(0xF3);
4050 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4051 emit_byte(0x0F);
4052 emit_byte(0xBD);
4053 emit_byte(0xC0 | encode);
4054 }
4056 void Assembler::movdq(XMMRegister dst, Register src) {
4057 // table D-1 says MMX/SSE2
4058 NOT_LP64(assert(VM_Version::supports_sse2() || VM_Version::supports_mmx(), ""));
4059 emit_byte(0x66);
4060 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4061 emit_byte(0x0F);
4062 emit_byte(0x6E);
4063 emit_byte(0xC0 | encode);
4064 }
4066 void Assembler::movdq(Register dst, XMMRegister src) {
4067 // table D-1 says MMX/SSE2
4068 NOT_LP64(assert(VM_Version::supports_sse2() || VM_Version::supports_mmx(), ""));
4069 emit_byte(0x66);
4070 // swap src/dst to get correct prefix
4071 int encode = prefixq_and_encode(src->encoding(), dst->encoding());
4072 emit_byte(0x0F);
4073 emit_byte(0x7E);
4074 emit_byte(0xC0 | encode);
4075 }
4077 void Assembler::movq(Register dst, Register src) {
4078 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4079 emit_byte(0x8B);
4080 emit_byte(0xC0 | encode);
4081 }
4083 void Assembler::movq(Register dst, Address src) {
4084 InstructionMark im(this);
4085 prefixq(src, dst);
4086 emit_byte(0x8B);
4087 emit_operand(dst, src);
4088 }
4090 void Assembler::movq(Address dst, Register src) {
4091 InstructionMark im(this);
4092 prefixq(dst, src);
4093 emit_byte(0x89);
4094 emit_operand(src, dst);
4095 }
4097 void Assembler::movsbq(Register dst, Address src) {
4098 InstructionMark im(this);
4099 prefixq(src, dst);
4100 emit_byte(0x0F);
4101 emit_byte(0xBE);
4102 emit_operand(dst, src);
4103 }
4105 void Assembler::movsbq(Register dst, Register src) {
4106 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4107 emit_byte(0x0F);
4108 emit_byte(0xBE);
4109 emit_byte(0xC0 | encode);
4110 }
4112 void Assembler::movslq(Register dst, int32_t imm32) {
4113 // dbx shows movslq(rcx, 3) as movq $0x0000000049000000,(%rbx)
4114 // and movslq(r8, 3); as movl $0x0000000048000000,(%rbx)
4115 // as a result we shouldn't use until tested at runtime...
4116 ShouldNotReachHere();
4117 InstructionMark im(this);
4118 int encode = prefixq_and_encode(dst->encoding());
4119 emit_byte(0xC7 | encode);
4120 emit_long(imm32);
4121 }
4123 void Assembler::movslq(Address dst, int32_t imm32) {
4124 assert(is_simm32(imm32), "lost bits");
4125 InstructionMark im(this);
4126 prefixq(dst);
4127 emit_byte(0xC7);
4128 emit_operand(rax, dst, 4);
4129 emit_long(imm32);
4130 }
4132 void Assembler::movslq(Register dst, Address src) {
4133 InstructionMark im(this);
4134 prefixq(src, dst);
4135 emit_byte(0x63);
4136 emit_operand(dst, src);
4137 }
4139 void Assembler::movslq(Register dst, Register src) {
4140 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4141 emit_byte(0x63);
4142 emit_byte(0xC0 | encode);
4143 }
4145 void Assembler::movswq(Register dst, Address src) {
4146 InstructionMark im(this);
4147 prefixq(src, dst);
4148 emit_byte(0x0F);
4149 emit_byte(0xBF);
4150 emit_operand(dst, src);
4151 }
4153 void Assembler::movswq(Register dst, Register src) {
4154 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4155 emit_byte(0x0F);
4156 emit_byte(0xBF);
4157 emit_byte(0xC0 | encode);
4158 }
4160 void Assembler::movzbq(Register dst, Address src) {
4161 InstructionMark im(this);
4162 prefixq(src, dst);
4163 emit_byte(0x0F);
4164 emit_byte(0xB6);
4165 emit_operand(dst, src);
4166 }
4168 void Assembler::movzbq(Register dst, Register src) {
4169 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4170 emit_byte(0x0F);
4171 emit_byte(0xB6);
4172 emit_byte(0xC0 | encode);
4173 }
4175 void Assembler::movzwq(Register dst, Address src) {
4176 InstructionMark im(this);
4177 prefixq(src, dst);
4178 emit_byte(0x0F);
4179 emit_byte(0xB7);
4180 emit_operand(dst, src);
4181 }
4183 void Assembler::movzwq(Register dst, Register src) {
4184 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4185 emit_byte(0x0F);
4186 emit_byte(0xB7);
4187 emit_byte(0xC0 | encode);
4188 }
4190 void Assembler::negq(Register dst) {
4191 int encode = prefixq_and_encode(dst->encoding());
4192 emit_byte(0xF7);
4193 emit_byte(0xD8 | encode);
4194 }
4196 void Assembler::notq(Register dst) {
4197 int encode = prefixq_and_encode(dst->encoding());
4198 emit_byte(0xF7);
4199 emit_byte(0xD0 | encode);
4200 }
4202 void Assembler::orq(Address dst, int32_t imm32) {
4203 InstructionMark im(this);
4204 prefixq(dst);
4205 emit_byte(0x81);
4206 emit_operand(rcx, dst, 4);
4207 emit_long(imm32);
4208 }
4210 void Assembler::orq(Register dst, int32_t imm32) {
4211 (void) prefixq_and_encode(dst->encoding());
4212 emit_arith(0x81, 0xC8, dst, imm32);
4213 }
4215 void Assembler::orq(Register dst, Address src) {
4216 InstructionMark im(this);
4217 prefixq(src, dst);
4218 emit_byte(0x0B);
4219 emit_operand(dst, src);
4220 }
4222 void Assembler::orq(Register dst, Register src) {
4223 (void) prefixq_and_encode(dst->encoding(), src->encoding());
4224 emit_arith(0x0B, 0xC0, dst, src);
4225 }
4227 void Assembler::popa() { // 64bit
4228 movq(r15, Address(rsp, 0));
4229 movq(r14, Address(rsp, wordSize));
4230 movq(r13, Address(rsp, 2 * wordSize));
4231 movq(r12, Address(rsp, 3 * wordSize));
4232 movq(r11, Address(rsp, 4 * wordSize));
4233 movq(r10, Address(rsp, 5 * wordSize));
4234 movq(r9, Address(rsp, 6 * wordSize));
4235 movq(r8, Address(rsp, 7 * wordSize));
4236 movq(rdi, Address(rsp, 8 * wordSize));
4237 movq(rsi, Address(rsp, 9 * wordSize));
4238 movq(rbp, Address(rsp, 10 * wordSize));
4239 // skip rsp
4240 movq(rbx, Address(rsp, 12 * wordSize));
4241 movq(rdx, Address(rsp, 13 * wordSize));
4242 movq(rcx, Address(rsp, 14 * wordSize));
4243 movq(rax, Address(rsp, 15 * wordSize));
4245 addq(rsp, 16 * wordSize);
4246 }
4248 void Assembler::popcntq(Register dst, Address src) {
4249 assert(VM_Version::supports_popcnt(), "must support");
4250 InstructionMark im(this);
4251 emit_byte(0xF3);
4252 prefixq(src, dst);
4253 emit_byte(0x0F);
4254 emit_byte(0xB8);
4255 emit_operand(dst, src);
4256 }
4258 void Assembler::popcntq(Register dst, Register src) {
4259 assert(VM_Version::supports_popcnt(), "must support");
4260 emit_byte(0xF3);
4261 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4262 emit_byte(0x0F);
4263 emit_byte(0xB8);
4264 emit_byte(0xC0 | encode);
4265 }
4267 void Assembler::popq(Address dst) {
4268 InstructionMark im(this);
4269 prefixq(dst);
4270 emit_byte(0x8F);
4271 emit_operand(rax, dst);
4272 }
4274 void Assembler::pusha() { // 64bit
4275 // we have to store original rsp. ABI says that 128 bytes
4276 // below rsp are local scratch.
4277 movq(Address(rsp, -5 * wordSize), rsp);
4279 subq(rsp, 16 * wordSize);
4281 movq(Address(rsp, 15 * wordSize), rax);
4282 movq(Address(rsp, 14 * wordSize), rcx);
4283 movq(Address(rsp, 13 * wordSize), rdx);
4284 movq(Address(rsp, 12 * wordSize), rbx);
4285 // skip rsp
4286 movq(Address(rsp, 10 * wordSize), rbp);
4287 movq(Address(rsp, 9 * wordSize), rsi);
4288 movq(Address(rsp, 8 * wordSize), rdi);
4289 movq(Address(rsp, 7 * wordSize), r8);
4290 movq(Address(rsp, 6 * wordSize), r9);
4291 movq(Address(rsp, 5 * wordSize), r10);
4292 movq(Address(rsp, 4 * wordSize), r11);
4293 movq(Address(rsp, 3 * wordSize), r12);
4294 movq(Address(rsp, 2 * wordSize), r13);
4295 movq(Address(rsp, wordSize), r14);
4296 movq(Address(rsp, 0), r15);
4297 }
4299 void Assembler::pushq(Address src) {
4300 InstructionMark im(this);
4301 prefixq(src);
4302 emit_byte(0xFF);
4303 emit_operand(rsi, src);
4304 }
4306 void Assembler::rclq(Register dst, int imm8) {
4307 assert(isShiftCount(imm8 >> 1), "illegal shift count");
4308 int encode = prefixq_and_encode(dst->encoding());
4309 if (imm8 == 1) {
4310 emit_byte(0xD1);
4311 emit_byte(0xD0 | encode);
4312 } else {
4313 emit_byte(0xC1);
4314 emit_byte(0xD0 | encode);
4315 emit_byte(imm8);
4316 }
4317 }
4318 void Assembler::sarq(Register dst, int imm8) {
4319 assert(isShiftCount(imm8 >> 1), "illegal shift count");
4320 int encode = prefixq_and_encode(dst->encoding());
4321 if (imm8 == 1) {
4322 emit_byte(0xD1);
4323 emit_byte(0xF8 | encode);
4324 } else {
4325 emit_byte(0xC1);
4326 emit_byte(0xF8 | encode);
4327 emit_byte(imm8);
4328 }
4329 }
4331 void Assembler::sarq(Register dst) {
4332 int encode = prefixq_and_encode(dst->encoding());
4333 emit_byte(0xD3);
4334 emit_byte(0xF8 | encode);
4335 }
4336 void Assembler::sbbq(Address dst, int32_t imm32) {
4337 InstructionMark im(this);
4338 prefixq(dst);
4339 emit_arith_operand(0x81, rbx, dst, imm32);
4340 }
4342 void Assembler::sbbq(Register dst, int32_t imm32) {
4343 (void) prefixq_and_encode(dst->encoding());
4344 emit_arith(0x81, 0xD8, dst, imm32);
4345 }
4347 void Assembler::sbbq(Register dst, Address src) {
4348 InstructionMark im(this);
4349 prefixq(src, dst);
4350 emit_byte(0x1B);
4351 emit_operand(dst, src);
4352 }
4354 void Assembler::sbbq(Register dst, Register src) {
4355 (void) prefixq_and_encode(dst->encoding(), src->encoding());
4356 emit_arith(0x1B, 0xC0, dst, src);
4357 }
4359 void Assembler::shlq(Register dst, int imm8) {
4360 assert(isShiftCount(imm8 >> 1), "illegal shift count");
4361 int encode = prefixq_and_encode(dst->encoding());
4362 if (imm8 == 1) {
4363 emit_byte(0xD1);
4364 emit_byte(0xE0 | encode);
4365 } else {
4366 emit_byte(0xC1);
4367 emit_byte(0xE0 | encode);
4368 emit_byte(imm8);
4369 }
4370 }
4372 void Assembler::shlq(Register dst) {
4373 int encode = prefixq_and_encode(dst->encoding());
4374 emit_byte(0xD3);
4375 emit_byte(0xE0 | encode);
4376 }
4378 void Assembler::shrq(Register dst, int imm8) {
4379 assert(isShiftCount(imm8 >> 1), "illegal shift count");
4380 int encode = prefixq_and_encode(dst->encoding());
4381 emit_byte(0xC1);
4382 emit_byte(0xE8 | encode);
4383 emit_byte(imm8);
4384 }
4386 void Assembler::shrq(Register dst) {
4387 int encode = prefixq_and_encode(dst->encoding());
4388 emit_byte(0xD3);
4389 emit_byte(0xE8 | encode);
4390 }
4392 void Assembler::subq(Address dst, int32_t imm32) {
4393 InstructionMark im(this);
4394 prefixq(dst);
4395 if (is8bit(imm32)) {
4396 emit_byte(0x83);
4397 emit_operand(rbp, dst, 1);
4398 emit_byte(imm32 & 0xFF);
4399 } else {
4400 emit_byte(0x81);
4401 emit_operand(rbp, dst, 4);
4402 emit_long(imm32);
4403 }
4404 }
4406 void Assembler::subq(Register dst, int32_t imm32) {
4407 (void) prefixq_and_encode(dst->encoding());
4408 emit_arith(0x81, 0xE8, dst, imm32);
4409 }
4411 void Assembler::subq(Address dst, Register src) {
4412 InstructionMark im(this);
4413 prefixq(dst, src);
4414 emit_byte(0x29);
4415 emit_operand(src, dst);
4416 }
4418 void Assembler::subq(Register dst, Address src) {
4419 InstructionMark im(this);
4420 prefixq(src, dst);
4421 emit_byte(0x2B);
4422 emit_operand(dst, src);
4423 }
4425 void Assembler::subq(Register dst, Register src) {
4426 (void) prefixq_and_encode(dst->encoding(), src->encoding());
4427 emit_arith(0x2B, 0xC0, dst, src);
4428 }
4430 void Assembler::testq(Register dst, int32_t imm32) {
4431 // not using emit_arith because test
4432 // doesn't support sign-extension of
4433 // 8bit operands
4434 int encode = dst->encoding();
4435 if (encode == 0) {
4436 prefix(REX_W);
4437 emit_byte(0xA9);
4438 } else {
4439 encode = prefixq_and_encode(encode);
4440 emit_byte(0xF7);
4441 emit_byte(0xC0 | encode);
4442 }
4443 emit_long(imm32);
4444 }
4446 void Assembler::testq(Register dst, Register src) {
4447 (void) prefixq_and_encode(dst->encoding(), src->encoding());
4448 emit_arith(0x85, 0xC0, dst, src);
4449 }
4451 void Assembler::xaddq(Address dst, Register src) {
4452 InstructionMark im(this);
4453 prefixq(dst, src);
4454 emit_byte(0x0F);
4455 emit_byte(0xC1);
4456 emit_operand(src, dst);
4457 }
4459 void Assembler::xchgq(Register dst, Address src) {
4460 InstructionMark im(this);
4461 prefixq(src, dst);
4462 emit_byte(0x87);
4463 emit_operand(dst, src);
4464 }
4466 void Assembler::xchgq(Register dst, Register src) {
4467 int encode = prefixq_and_encode(dst->encoding(), src->encoding());
4468 emit_byte(0x87);
4469 emit_byte(0xc0 | encode);
4470 }
4472 void Assembler::xorq(Register dst, Register src) {
4473 (void) prefixq_and_encode(dst->encoding(), src->encoding());
4474 emit_arith(0x33, 0xC0, dst, src);
4475 }
4477 void Assembler::xorq(Register dst, Address src) {
4478 InstructionMark im(this);
4479 prefixq(src, dst);
4480 emit_byte(0x33);
4481 emit_operand(dst, src);
4482 }
4484 #endif // !LP64
4486 static Assembler::Condition reverse[] = {
4487 Assembler::noOverflow /* overflow = 0x0 */ ,
4488 Assembler::overflow /* noOverflow = 0x1 */ ,
4489 Assembler::aboveEqual /* carrySet = 0x2, below = 0x2 */ ,
4490 Assembler::below /* aboveEqual = 0x3, carryClear = 0x3 */ ,
4491 Assembler::notZero /* zero = 0x4, equal = 0x4 */ ,
4492 Assembler::zero /* notZero = 0x5, notEqual = 0x5 */ ,
4493 Assembler::above /* belowEqual = 0x6 */ ,
4494 Assembler::belowEqual /* above = 0x7 */ ,
4495 Assembler::positive /* negative = 0x8 */ ,
4496 Assembler::negative /* positive = 0x9 */ ,
4497 Assembler::noParity /* parity = 0xa */ ,
4498 Assembler::parity /* noParity = 0xb */ ,
4499 Assembler::greaterEqual /* less = 0xc */ ,
4500 Assembler::less /* greaterEqual = 0xd */ ,
4501 Assembler::greater /* lessEqual = 0xe */ ,
4502 Assembler::lessEqual /* greater = 0xf, */
4504 };
4507 // Implementation of MacroAssembler
4509 // First all the versions that have distinct versions depending on 32/64 bit
4510 // Unless the difference is trivial (1 line or so).
4512 #ifndef _LP64
4514 // 32bit versions
4516 Address MacroAssembler::as_Address(AddressLiteral adr) {
4517 return Address(adr.target(), adr.rspec());
4518 }
4520 Address MacroAssembler::as_Address(ArrayAddress adr) {
4521 return Address::make_array(adr);
4522 }
4524 int MacroAssembler::biased_locking_enter(Register lock_reg,
4525 Register obj_reg,
4526 Register swap_reg,
4527 Register tmp_reg,
4528 bool swap_reg_contains_mark,
4529 Label& done,
4530 Label* slow_case,
4531 BiasedLockingCounters* counters) {
4532 assert(UseBiasedLocking, "why call this otherwise?");
4533 assert(swap_reg == rax, "swap_reg must be rax, for cmpxchg");
4534 assert_different_registers(lock_reg, obj_reg, swap_reg);
4536 if (PrintBiasedLockingStatistics && counters == NULL)
4537 counters = BiasedLocking::counters();
4539 bool need_tmp_reg = false;
4540 if (tmp_reg == noreg) {
4541 need_tmp_reg = true;
4542 tmp_reg = lock_reg;
4543 } else {
4544 assert_different_registers(lock_reg, obj_reg, swap_reg, tmp_reg);
4545 }
4546 assert(markOopDesc::age_shift == markOopDesc::lock_bits + markOopDesc::biased_lock_bits, "biased locking makes assumptions about bit layout");
4547 Address mark_addr (obj_reg, oopDesc::mark_offset_in_bytes());
4548 Address klass_addr (obj_reg, oopDesc::klass_offset_in_bytes());
4549 Address saved_mark_addr(lock_reg, 0);
4551 // Biased locking
4552 // See whether the lock is currently biased toward our thread and
4553 // whether the epoch is still valid
4554 // Note that the runtime guarantees sufficient alignment of JavaThread
4555 // pointers to allow age to be placed into low bits
4556 // First check to see whether biasing is even enabled for this object
4557 Label cas_label;
4558 int null_check_offset = -1;
4559 if (!swap_reg_contains_mark) {
4560 null_check_offset = offset();
4561 movl(swap_reg, mark_addr);
4562 }
4563 if (need_tmp_reg) {
4564 push(tmp_reg);
4565 }
4566 movl(tmp_reg, swap_reg);
4567 andl(tmp_reg, markOopDesc::biased_lock_mask_in_place);
4568 cmpl(tmp_reg, markOopDesc::biased_lock_pattern);
4569 if (need_tmp_reg) {
4570 pop(tmp_reg);
4571 }
4572 jcc(Assembler::notEqual, cas_label);
4573 // The bias pattern is present in the object's header. Need to check
4574 // whether the bias owner and the epoch are both still current.
4575 // Note that because there is no current thread register on x86 we
4576 // need to store off the mark word we read out of the object to
4577 // avoid reloading it and needing to recheck invariants below. This
4578 // store is unfortunate but it makes the overall code shorter and
4579 // simpler.
4580 movl(saved_mark_addr, swap_reg);
4581 if (need_tmp_reg) {
4582 push(tmp_reg);
4583 }
4584 get_thread(tmp_reg);
4585 xorl(swap_reg, tmp_reg);
4586 if (swap_reg_contains_mark) {
4587 null_check_offset = offset();
4588 }
4589 movl(tmp_reg, klass_addr);
4590 xorl(swap_reg, Address(tmp_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
4591 andl(swap_reg, ~((int) markOopDesc::age_mask_in_place));
4592 if (need_tmp_reg) {
4593 pop(tmp_reg);
4594 }
4595 if (counters != NULL) {
4596 cond_inc32(Assembler::zero,
4597 ExternalAddress((address)counters->biased_lock_entry_count_addr()));
4598 }
4599 jcc(Assembler::equal, done);
4601 Label try_revoke_bias;
4602 Label try_rebias;
4604 // At this point we know that the header has the bias pattern and
4605 // that we are not the bias owner in the current epoch. We need to
4606 // figure out more details about the state of the header in order to
4607 // know what operations can be legally performed on the object's
4608 // header.
4610 // If the low three bits in the xor result aren't clear, that means
4611 // the prototype header is no longer biased and we have to revoke
4612 // the bias on this object.
4613 testl(swap_reg, markOopDesc::biased_lock_mask_in_place);
4614 jcc(Assembler::notZero, try_revoke_bias);
4616 // Biasing is still enabled for this data type. See whether the
4617 // epoch of the current bias is still valid, meaning that the epoch
4618 // bits of the mark word are equal to the epoch bits of the
4619 // prototype header. (Note that the prototype header's epoch bits
4620 // only change at a safepoint.) If not, attempt to rebias the object
4621 // toward the current thread. Note that we must be absolutely sure
4622 // that the current epoch is invalid in order to do this because
4623 // otherwise the manipulations it performs on the mark word are
4624 // illegal.
4625 testl(swap_reg, markOopDesc::epoch_mask_in_place);
4626 jcc(Assembler::notZero, try_rebias);
4628 // The epoch of the current bias is still valid but we know nothing
4629 // about the owner; it might be set or it might be clear. Try to
4630 // acquire the bias of the object using an atomic operation. If this
4631 // fails we will go in to the runtime to revoke the object's bias.
4632 // Note that we first construct the presumed unbiased header so we
4633 // don't accidentally blow away another thread's valid bias.
4634 movl(swap_reg, saved_mark_addr);
4635 andl(swap_reg,
4636 markOopDesc::biased_lock_mask_in_place | markOopDesc::age_mask_in_place | markOopDesc::epoch_mask_in_place);
4637 if (need_tmp_reg) {
4638 push(tmp_reg);
4639 }
4640 get_thread(tmp_reg);
4641 orl(tmp_reg, swap_reg);
4642 if (os::is_MP()) {
4643 lock();
4644 }
4645 cmpxchgptr(tmp_reg, Address(obj_reg, 0));
4646 if (need_tmp_reg) {
4647 pop(tmp_reg);
4648 }
4649 // If the biasing toward our thread failed, this means that
4650 // another thread succeeded in biasing it toward itself and we
4651 // need to revoke that bias. The revocation will occur in the
4652 // interpreter runtime in the slow case.
4653 if (counters != NULL) {
4654 cond_inc32(Assembler::zero,
4655 ExternalAddress((address)counters->anonymously_biased_lock_entry_count_addr()));
4656 }
4657 if (slow_case != NULL) {
4658 jcc(Assembler::notZero, *slow_case);
4659 }
4660 jmp(done);
4662 bind(try_rebias);
4663 // At this point we know the epoch has expired, meaning that the
4664 // current "bias owner", if any, is actually invalid. Under these
4665 // circumstances _only_, we are allowed to use the current header's
4666 // value as the comparison value when doing the cas to acquire the
4667 // bias in the current epoch. In other words, we allow transfer of
4668 // the bias from one thread to another directly in this situation.
4669 //
4670 // FIXME: due to a lack of registers we currently blow away the age
4671 // bits in this situation. Should attempt to preserve them.
4672 if (need_tmp_reg) {
4673 push(tmp_reg);
4674 }
4675 get_thread(tmp_reg);
4676 movl(swap_reg, klass_addr);
4677 orl(tmp_reg, Address(swap_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
4678 movl(swap_reg, saved_mark_addr);
4679 if (os::is_MP()) {
4680 lock();
4681 }
4682 cmpxchgptr(tmp_reg, Address(obj_reg, 0));
4683 if (need_tmp_reg) {
4684 pop(tmp_reg);
4685 }
4686 // If the biasing toward our thread failed, then another thread
4687 // succeeded in biasing it toward itself and we need to revoke that
4688 // bias. The revocation will occur in the runtime in the slow case.
4689 if (counters != NULL) {
4690 cond_inc32(Assembler::zero,
4691 ExternalAddress((address)counters->rebiased_lock_entry_count_addr()));
4692 }
4693 if (slow_case != NULL) {
4694 jcc(Assembler::notZero, *slow_case);
4695 }
4696 jmp(done);
4698 bind(try_revoke_bias);
4699 // The prototype mark in the klass doesn't have the bias bit set any
4700 // more, indicating that objects of this data type are not supposed
4701 // to be biased any more. We are going to try to reset the mark of
4702 // this object to the prototype value and fall through to the
4703 // CAS-based locking scheme. Note that if our CAS fails, it means
4704 // that another thread raced us for the privilege of revoking the
4705 // bias of this particular object, so it's okay to continue in the
4706 // normal locking code.
4707 //
4708 // FIXME: due to a lack of registers we currently blow away the age
4709 // bits in this situation. Should attempt to preserve them.
4710 movl(swap_reg, saved_mark_addr);
4711 if (need_tmp_reg) {
4712 push(tmp_reg);
4713 }
4714 movl(tmp_reg, klass_addr);
4715 movl(tmp_reg, Address(tmp_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
4716 if (os::is_MP()) {
4717 lock();
4718 }
4719 cmpxchgptr(tmp_reg, Address(obj_reg, 0));
4720 if (need_tmp_reg) {
4721 pop(tmp_reg);
4722 }
4723 // Fall through to the normal CAS-based lock, because no matter what
4724 // the result of the above CAS, some thread must have succeeded in
4725 // removing the bias bit from the object's header.
4726 if (counters != NULL) {
4727 cond_inc32(Assembler::zero,
4728 ExternalAddress((address)counters->revoked_lock_entry_count_addr()));
4729 }
4731 bind(cas_label);
4733 return null_check_offset;
4734 }
4735 void MacroAssembler::call_VM_leaf_base(address entry_point,
4736 int number_of_arguments) {
4737 call(RuntimeAddress(entry_point));
4738 increment(rsp, number_of_arguments * wordSize);
4739 }
4741 void MacroAssembler::cmpoop(Address src1, jobject obj) {
4742 cmp_literal32(src1, (int32_t)obj, oop_Relocation::spec_for_immediate());
4743 }
4745 void MacroAssembler::cmpoop(Register src1, jobject obj) {
4746 cmp_literal32(src1, (int32_t)obj, oop_Relocation::spec_for_immediate());
4747 }
4749 void MacroAssembler::extend_sign(Register hi, Register lo) {
4750 // According to Intel Doc. AP-526, "Integer Divide", p.18.
4751 if (VM_Version::is_P6() && hi == rdx && lo == rax) {
4752 cdql();
4753 } else {
4754 movl(hi, lo);
4755 sarl(hi, 31);
4756 }
4757 }
4759 void MacroAssembler::fat_nop() {
4760 // A 5 byte nop that is safe for patching (see patch_verified_entry)
4761 emit_byte(0x26); // es:
4762 emit_byte(0x2e); // cs:
4763 emit_byte(0x64); // fs:
4764 emit_byte(0x65); // gs:
4765 emit_byte(0x90);
4766 }
4768 void MacroAssembler::jC2(Register tmp, Label& L) {
4769 // set parity bit if FPU flag C2 is set (via rax)
4770 save_rax(tmp);
4771 fwait(); fnstsw_ax();
4772 sahf();
4773 restore_rax(tmp);
4774 // branch
4775 jcc(Assembler::parity, L);
4776 }
4778 void MacroAssembler::jnC2(Register tmp, Label& L) {
4779 // set parity bit if FPU flag C2 is set (via rax)
4780 save_rax(tmp);
4781 fwait(); fnstsw_ax();
4782 sahf();
4783 restore_rax(tmp);
4784 // branch
4785 jcc(Assembler::noParity, L);
4786 }
4788 // 32bit can do a case table jump in one instruction but we no longer allow the base
4789 // to be installed in the Address class
4790 void MacroAssembler::jump(ArrayAddress entry) {
4791 jmp(as_Address(entry));
4792 }
4794 // Note: y_lo will be destroyed
4795 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
4796 // Long compare for Java (semantics as described in JVM spec.)
4797 Label high, low, done;
4799 cmpl(x_hi, y_hi);
4800 jcc(Assembler::less, low);
4801 jcc(Assembler::greater, high);
4802 // x_hi is the return register
4803 xorl(x_hi, x_hi);
4804 cmpl(x_lo, y_lo);
4805 jcc(Assembler::below, low);
4806 jcc(Assembler::equal, done);
4808 bind(high);
4809 xorl(x_hi, x_hi);
4810 increment(x_hi);
4811 jmp(done);
4813 bind(low);
4814 xorl(x_hi, x_hi);
4815 decrementl(x_hi);
4817 bind(done);
4818 }
4820 void MacroAssembler::lea(Register dst, AddressLiteral src) {
4821 mov_literal32(dst, (int32_t)src.target(), src.rspec());
4822 }
4824 void MacroAssembler::lea(Address dst, AddressLiteral adr) {
4825 // leal(dst, as_Address(adr));
4826 // see note in movl as to why we must use a move
4827 mov_literal32(dst, (int32_t) adr.target(), adr.rspec());
4828 }
4830 void MacroAssembler::leave() {
4831 mov(rsp, rbp);
4832 pop(rbp);
4833 }
4835 void MacroAssembler::lmul(int x_rsp_offset, int y_rsp_offset) {
4836 // Multiplication of two Java long values stored on the stack
4837 // as illustrated below. Result is in rdx:rax.
4838 //
4839 // rsp ---> [ ?? ] \ \
4840 // .... | y_rsp_offset |
4841 // [ y_lo ] / (in bytes) | x_rsp_offset
4842 // [ y_hi ] | (in bytes)
4843 // .... |
4844 // [ x_lo ] /
4845 // [ x_hi ]
4846 // ....
4847 //
4848 // Basic idea: lo(result) = lo(x_lo * y_lo)
4849 // hi(result) = hi(x_lo * y_lo) + lo(x_hi * y_lo) + lo(x_lo * y_hi)
4850 Address x_hi(rsp, x_rsp_offset + wordSize); Address x_lo(rsp, x_rsp_offset);
4851 Address y_hi(rsp, y_rsp_offset + wordSize); Address y_lo(rsp, y_rsp_offset);
4852 Label quick;
4853 // load x_hi, y_hi and check if quick
4854 // multiplication is possible
4855 movl(rbx, x_hi);
4856 movl(rcx, y_hi);
4857 movl(rax, rbx);
4858 orl(rbx, rcx); // rbx, = 0 <=> x_hi = 0 and y_hi = 0
4859 jcc(Assembler::zero, quick); // if rbx, = 0 do quick multiply
4860 // do full multiplication
4861 // 1st step
4862 mull(y_lo); // x_hi * y_lo
4863 movl(rbx, rax); // save lo(x_hi * y_lo) in rbx,
4864 // 2nd step
4865 movl(rax, x_lo);
4866 mull(rcx); // x_lo * y_hi
4867 addl(rbx, rax); // add lo(x_lo * y_hi) to rbx,
4868 // 3rd step
4869 bind(quick); // note: rbx, = 0 if quick multiply!
4870 movl(rax, x_lo);
4871 mull(y_lo); // x_lo * y_lo
4872 addl(rdx, rbx); // correct hi(x_lo * y_lo)
4873 }
4875 void MacroAssembler::lneg(Register hi, Register lo) {
4876 negl(lo);
4877 adcl(hi, 0);
4878 negl(hi);
4879 }
4881 void MacroAssembler::lshl(Register hi, Register lo) {
4882 // Java shift left long support (semantics as described in JVM spec., p.305)
4883 // (basic idea for shift counts s >= n: x << s == (x << n) << (s - n))
4884 // shift value is in rcx !
4885 assert(hi != rcx, "must not use rcx");
4886 assert(lo != rcx, "must not use rcx");
4887 const Register s = rcx; // shift count
4888 const int n = BitsPerWord;
4889 Label L;
4890 andl(s, 0x3f); // s := s & 0x3f (s < 0x40)
4891 cmpl(s, n); // if (s < n)
4892 jcc(Assembler::less, L); // else (s >= n)
4893 movl(hi, lo); // x := x << n
4894 xorl(lo, lo);
4895 // Note: subl(s, n) is not needed since the Intel shift instructions work rcx mod n!
4896 bind(L); // s (mod n) < n
4897 shldl(hi, lo); // x := x << s
4898 shll(lo);
4899 }
4902 void MacroAssembler::lshr(Register hi, Register lo, bool sign_extension) {
4903 // Java shift right long support (semantics as described in JVM spec., p.306 & p.310)
4904 // (basic idea for shift counts s >= n: x >> s == (x >> n) >> (s - n))
4905 assert(hi != rcx, "must not use rcx");
4906 assert(lo != rcx, "must not use rcx");
4907 const Register s = rcx; // shift count
4908 const int n = BitsPerWord;
4909 Label L;
4910 andl(s, 0x3f); // s := s & 0x3f (s < 0x40)
4911 cmpl(s, n); // if (s < n)
4912 jcc(Assembler::less, L); // else (s >= n)
4913 movl(lo, hi); // x := x >> n
4914 if (sign_extension) sarl(hi, 31);
4915 else xorl(hi, hi);
4916 // Note: subl(s, n) is not needed since the Intel shift instructions work rcx mod n!
4917 bind(L); // s (mod n) < n
4918 shrdl(lo, hi); // x := x >> s
4919 if (sign_extension) sarl(hi);
4920 else shrl(hi);
4921 }
4923 void MacroAssembler::movoop(Register dst, jobject obj) {
4924 mov_literal32(dst, (int32_t)obj, oop_Relocation::spec_for_immediate());
4925 }
4927 void MacroAssembler::movoop(Address dst, jobject obj) {
4928 mov_literal32(dst, (int32_t)obj, oop_Relocation::spec_for_immediate());
4929 }
4931 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
4932 if (src.is_lval()) {
4933 mov_literal32(dst, (intptr_t)src.target(), src.rspec());
4934 } else {
4935 movl(dst, as_Address(src));
4936 }
4937 }
4939 void MacroAssembler::movptr(ArrayAddress dst, Register src) {
4940 movl(as_Address(dst), src);
4941 }
4943 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
4944 movl(dst, as_Address(src));
4945 }
4947 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
4948 void MacroAssembler::movptr(Address dst, intptr_t src) {
4949 movl(dst, src);
4950 }
4953 void MacroAssembler::pop_callee_saved_registers() {
4954 pop(rcx);
4955 pop(rdx);
4956 pop(rdi);
4957 pop(rsi);
4958 }
4960 void MacroAssembler::pop_fTOS() {
4961 fld_d(Address(rsp, 0));
4962 addl(rsp, 2 * wordSize);
4963 }
4965 void MacroAssembler::push_callee_saved_registers() {
4966 push(rsi);
4967 push(rdi);
4968 push(rdx);
4969 push(rcx);
4970 }
4972 void MacroAssembler::push_fTOS() {
4973 subl(rsp, 2 * wordSize);
4974 fstp_d(Address(rsp, 0));
4975 }
4978 void MacroAssembler::pushoop(jobject obj) {
4979 push_literal32((int32_t)obj, oop_Relocation::spec_for_immediate());
4980 }
4983 void MacroAssembler::pushptr(AddressLiteral src) {
4984 if (src.is_lval()) {
4985 push_literal32((int32_t)src.target(), src.rspec());
4986 } else {
4987 pushl(as_Address(src));
4988 }
4989 }
4991 void MacroAssembler::set_word_if_not_zero(Register dst) {
4992 xorl(dst, dst);
4993 set_byte_if_not_zero(dst);
4994 }
4996 static void pass_arg0(MacroAssembler* masm, Register arg) {
4997 masm->push(arg);
4998 }
5000 static void pass_arg1(MacroAssembler* masm, Register arg) {
5001 masm->push(arg);
5002 }
5004 static void pass_arg2(MacroAssembler* masm, Register arg) {
5005 masm->push(arg);
5006 }
5008 static void pass_arg3(MacroAssembler* masm, Register arg) {
5009 masm->push(arg);
5010 }
5012 #ifndef PRODUCT
5013 extern "C" void findpc(intptr_t x);
5014 #endif
5016 void MacroAssembler::debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg) {
5017 // In order to get locks to work, we need to fake a in_VM state
5018 JavaThread* thread = JavaThread::current();
5019 JavaThreadState saved_state = thread->thread_state();
5020 thread->set_thread_state(_thread_in_vm);
5021 if (ShowMessageBoxOnError) {
5022 JavaThread* thread = JavaThread::current();
5023 JavaThreadState saved_state = thread->thread_state();
5024 thread->set_thread_state(_thread_in_vm);
5025 if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
5026 ttyLocker ttyl;
5027 BytecodeCounter::print();
5028 }
5029 // To see where a verify_oop failed, get $ebx+40/X for this frame.
5030 // This is the value of eip which points to where verify_oop will return.
5031 if (os::message_box(msg, "Execution stopped, print registers?")) {
5032 ttyLocker ttyl;
5033 tty->print_cr("eip = 0x%08x", eip);
5034 #ifndef PRODUCT
5035 if ((WizardMode || Verbose) && PrintMiscellaneous) {
5036 tty->cr();
5037 findpc(eip);
5038 tty->cr();
5039 }
5040 #endif
5041 tty->print_cr("rax = 0x%08x", rax);
5042 tty->print_cr("rbx = 0x%08x", rbx);
5043 tty->print_cr("rcx = 0x%08x", rcx);
5044 tty->print_cr("rdx = 0x%08x", rdx);
5045 tty->print_cr("rdi = 0x%08x", rdi);
5046 tty->print_cr("rsi = 0x%08x", rsi);
5047 tty->print_cr("rbp = 0x%08x", rbp);
5048 tty->print_cr("rsp = 0x%08x", rsp);
5049 BREAKPOINT;
5050 assert(false, "start up GDB");
5051 }
5052 } else {
5053 ttyLocker ttyl;
5054 ::tty->print_cr("=============== DEBUG MESSAGE: %s ================\n", msg);
5055 assert(false, "DEBUG MESSAGE");
5056 }
5057 ThreadStateTransition::transition(thread, _thread_in_vm, saved_state);
5058 }
5060 void MacroAssembler::stop(const char* msg) {
5061 ExternalAddress message((address)msg);
5062 // push address of message
5063 pushptr(message.addr());
5064 { Label L; call(L, relocInfo::none); bind(L); } // push eip
5065 pusha(); // push registers
5066 call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
5067 hlt();
5068 }
5070 void MacroAssembler::warn(const char* msg) {
5071 push_CPU_state();
5073 ExternalAddress message((address) msg);
5074 // push address of message
5075 pushptr(message.addr());
5077 call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning)));
5078 addl(rsp, wordSize); // discard argument
5079 pop_CPU_state();
5080 }
5082 #else // _LP64
5084 // 64 bit versions
5086 Address MacroAssembler::as_Address(AddressLiteral adr) {
5087 // amd64 always does this as a pc-rel
5088 // we can be absolute or disp based on the instruction type
5089 // jmp/call are displacements others are absolute
5090 assert(!adr.is_lval(), "must be rval");
5091 assert(reachable(adr), "must be");
5092 return Address((int32_t)(intptr_t)(adr.target() - pc()), adr.target(), adr.reloc());
5094 }
5096 Address MacroAssembler::as_Address(ArrayAddress adr) {
5097 AddressLiteral base = adr.base();
5098 lea(rscratch1, base);
5099 Address index = adr.index();
5100 assert(index._disp == 0, "must not have disp"); // maybe it can?
5101 Address array(rscratch1, index._index, index._scale, index._disp);
5102 return array;
5103 }
5105 int MacroAssembler::biased_locking_enter(Register lock_reg,
5106 Register obj_reg,
5107 Register swap_reg,
5108 Register tmp_reg,
5109 bool swap_reg_contains_mark,
5110 Label& done,
5111 Label* slow_case,
5112 BiasedLockingCounters* counters) {
5113 assert(UseBiasedLocking, "why call this otherwise?");
5114 assert(swap_reg == rax, "swap_reg must be rax for cmpxchgq");
5115 assert(tmp_reg != noreg, "tmp_reg must be supplied");
5116 assert_different_registers(lock_reg, obj_reg, swap_reg, tmp_reg);
5117 assert(markOopDesc::age_shift == markOopDesc::lock_bits + markOopDesc::biased_lock_bits, "biased locking makes assumptions about bit layout");
5118 Address mark_addr (obj_reg, oopDesc::mark_offset_in_bytes());
5119 Address saved_mark_addr(lock_reg, 0);
5121 if (PrintBiasedLockingStatistics && counters == NULL)
5122 counters = BiasedLocking::counters();
5124 // Biased locking
5125 // See whether the lock is currently biased toward our thread and
5126 // whether the epoch is still valid
5127 // Note that the runtime guarantees sufficient alignment of JavaThread
5128 // pointers to allow age to be placed into low bits
5129 // First check to see whether biasing is even enabled for this object
5130 Label cas_label;
5131 int null_check_offset = -1;
5132 if (!swap_reg_contains_mark) {
5133 null_check_offset = offset();
5134 movq(swap_reg, mark_addr);
5135 }
5136 movq(tmp_reg, swap_reg);
5137 andq(tmp_reg, markOopDesc::biased_lock_mask_in_place);
5138 cmpq(tmp_reg, markOopDesc::biased_lock_pattern);
5139 jcc(Assembler::notEqual, cas_label);
5140 // The bias pattern is present in the object's header. Need to check
5141 // whether the bias owner and the epoch are both still current.
5142 load_prototype_header(tmp_reg, obj_reg);
5143 orq(tmp_reg, r15_thread);
5144 xorq(tmp_reg, swap_reg);
5145 andq(tmp_reg, ~((int) markOopDesc::age_mask_in_place));
5146 if (counters != NULL) {
5147 cond_inc32(Assembler::zero,
5148 ExternalAddress((address) counters->anonymously_biased_lock_entry_count_addr()));
5149 }
5150 jcc(Assembler::equal, done);
5152 Label try_revoke_bias;
5153 Label try_rebias;
5155 // At this point we know that the header has the bias pattern and
5156 // that we are not the bias owner in the current epoch. We need to
5157 // figure out more details about the state of the header in order to
5158 // know what operations can be legally performed on the object's
5159 // header.
5161 // If the low three bits in the xor result aren't clear, that means
5162 // the prototype header is no longer biased and we have to revoke
5163 // the bias on this object.
5164 testq(tmp_reg, markOopDesc::biased_lock_mask_in_place);
5165 jcc(Assembler::notZero, try_revoke_bias);
5167 // Biasing is still enabled for this data type. See whether the
5168 // epoch of the current bias is still valid, meaning that the epoch
5169 // bits of the mark word are equal to the epoch bits of the
5170 // prototype header. (Note that the prototype header's epoch bits
5171 // only change at a safepoint.) If not, attempt to rebias the object
5172 // toward the current thread. Note that we must be absolutely sure
5173 // that the current epoch is invalid in order to do this because
5174 // otherwise the manipulations it performs on the mark word are
5175 // illegal.
5176 testq(tmp_reg, markOopDesc::epoch_mask_in_place);
5177 jcc(Assembler::notZero, try_rebias);
5179 // The epoch of the current bias is still valid but we know nothing
5180 // about the owner; it might be set or it might be clear. Try to
5181 // acquire the bias of the object using an atomic operation. If this
5182 // fails we will go in to the runtime to revoke the object's bias.
5183 // Note that we first construct the presumed unbiased header so we
5184 // don't accidentally blow away another thread's valid bias.
5185 andq(swap_reg,
5186 markOopDesc::biased_lock_mask_in_place | markOopDesc::age_mask_in_place | markOopDesc::epoch_mask_in_place);
5187 movq(tmp_reg, swap_reg);
5188 orq(tmp_reg, r15_thread);
5189 if (os::is_MP()) {
5190 lock();
5191 }
5192 cmpxchgq(tmp_reg, Address(obj_reg, 0));
5193 // If the biasing toward our thread failed, this means that
5194 // another thread succeeded in biasing it toward itself and we
5195 // need to revoke that bias. The revocation will occur in the
5196 // interpreter runtime in the slow case.
5197 if (counters != NULL) {
5198 cond_inc32(Assembler::zero,
5199 ExternalAddress((address) counters->anonymously_biased_lock_entry_count_addr()));
5200 }
5201 if (slow_case != NULL) {
5202 jcc(Assembler::notZero, *slow_case);
5203 }
5204 jmp(done);
5206 bind(try_rebias);
5207 // At this point we know the epoch has expired, meaning that the
5208 // current "bias owner", if any, is actually invalid. Under these
5209 // circumstances _only_, we are allowed to use the current header's
5210 // value as the comparison value when doing the cas to acquire the
5211 // bias in the current epoch. In other words, we allow transfer of
5212 // the bias from one thread to another directly in this situation.
5213 //
5214 // FIXME: due to a lack of registers we currently blow away the age
5215 // bits in this situation. Should attempt to preserve them.
5216 load_prototype_header(tmp_reg, obj_reg);
5217 orq(tmp_reg, r15_thread);
5218 if (os::is_MP()) {
5219 lock();
5220 }
5221 cmpxchgq(tmp_reg, Address(obj_reg, 0));
5222 // If the biasing toward our thread failed, then another thread
5223 // succeeded in biasing it toward itself and we need to revoke that
5224 // bias. The revocation will occur in the runtime in the slow case.
5225 if (counters != NULL) {
5226 cond_inc32(Assembler::zero,
5227 ExternalAddress((address) counters->rebiased_lock_entry_count_addr()));
5228 }
5229 if (slow_case != NULL) {
5230 jcc(Assembler::notZero, *slow_case);
5231 }
5232 jmp(done);
5234 bind(try_revoke_bias);
5235 // The prototype mark in the klass doesn't have the bias bit set any
5236 // more, indicating that objects of this data type are not supposed
5237 // to be biased any more. We are going to try to reset the mark of
5238 // this object to the prototype value and fall through to the
5239 // CAS-based locking scheme. Note that if our CAS fails, it means
5240 // that another thread raced us for the privilege of revoking the
5241 // bias of this particular object, so it's okay to continue in the
5242 // normal locking code.
5243 //
5244 // FIXME: due to a lack of registers we currently blow away the age
5245 // bits in this situation. Should attempt to preserve them.
5246 load_prototype_header(tmp_reg, obj_reg);
5247 if (os::is_MP()) {
5248 lock();
5249 }
5250 cmpxchgq(tmp_reg, Address(obj_reg, 0));
5251 // Fall through to the normal CAS-based lock, because no matter what
5252 // the result of the above CAS, some thread must have succeeded in
5253 // removing the bias bit from the object's header.
5254 if (counters != NULL) {
5255 cond_inc32(Assembler::zero,
5256 ExternalAddress((address) counters->revoked_lock_entry_count_addr()));
5257 }
5259 bind(cas_label);
5261 return null_check_offset;
5262 }
5264 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) {
5265 Label L, E;
5267 #ifdef _WIN64
5268 // Windows always allocates space for it's register args
5269 assert(num_args <= 4, "only register arguments supported");
5270 subq(rsp, frame::arg_reg_save_area_bytes);
5271 #endif
5273 // Align stack if necessary
5274 testl(rsp, 15);
5275 jcc(Assembler::zero, L);
5277 subq(rsp, 8);
5278 {
5279 call(RuntimeAddress(entry_point));
5280 }
5281 addq(rsp, 8);
5282 jmp(E);
5284 bind(L);
5285 {
5286 call(RuntimeAddress(entry_point));
5287 }
5289 bind(E);
5291 #ifdef _WIN64
5292 // restore stack pointer
5293 addq(rsp, frame::arg_reg_save_area_bytes);
5294 #endif
5296 }
5298 void MacroAssembler::cmp64(Register src1, AddressLiteral src2) {
5299 assert(!src2.is_lval(), "should use cmpptr");
5301 if (reachable(src2)) {
5302 cmpq(src1, as_Address(src2));
5303 } else {
5304 lea(rscratch1, src2);
5305 Assembler::cmpq(src1, Address(rscratch1, 0));
5306 }
5307 }
5309 int MacroAssembler::corrected_idivq(Register reg) {
5310 // Full implementation of Java ldiv and lrem; checks for special
5311 // case as described in JVM spec., p.243 & p.271. The function
5312 // returns the (pc) offset of the idivl instruction - may be needed
5313 // for implicit exceptions.
5314 //
5315 // normal case special case
5316 //
5317 // input : rax: dividend min_long
5318 // reg: divisor (may not be eax/edx) -1
5319 //
5320 // output: rax: quotient (= rax idiv reg) min_long
5321 // rdx: remainder (= rax irem reg) 0
5322 assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register");
5323 static const int64_t min_long = 0x8000000000000000;
5324 Label normal_case, special_case;
5326 // check for special case
5327 cmp64(rax, ExternalAddress((address) &min_long));
5328 jcc(Assembler::notEqual, normal_case);
5329 xorl(rdx, rdx); // prepare rdx for possible special case (where
5330 // remainder = 0)
5331 cmpq(reg, -1);
5332 jcc(Assembler::equal, special_case);
5334 // handle normal case
5335 bind(normal_case);
5336 cdqq();
5337 int idivq_offset = offset();
5338 idivq(reg);
5340 // normal and special case exit
5341 bind(special_case);
5343 return idivq_offset;
5344 }
5346 void MacroAssembler::decrementq(Register reg, int value) {
5347 if (value == min_jint) { subq(reg, value); return; }
5348 if (value < 0) { incrementq(reg, -value); return; }
5349 if (value == 0) { ; return; }
5350 if (value == 1 && UseIncDec) { decq(reg) ; return; }
5351 /* else */ { subq(reg, value) ; return; }
5352 }
5354 void MacroAssembler::decrementq(Address dst, int value) {
5355 if (value == min_jint) { subq(dst, value); return; }
5356 if (value < 0) { incrementq(dst, -value); return; }
5357 if (value == 0) { ; return; }
5358 if (value == 1 && UseIncDec) { decq(dst) ; return; }
5359 /* else */ { subq(dst, value) ; return; }
5360 }
5362 void MacroAssembler::fat_nop() {
5363 // A 5 byte nop that is safe for patching (see patch_verified_entry)
5364 // Recommened sequence from 'Software Optimization Guide for the AMD
5365 // Hammer Processor'
5366 emit_byte(0x66);
5367 emit_byte(0x66);
5368 emit_byte(0x90);
5369 emit_byte(0x66);
5370 emit_byte(0x90);
5371 }
5373 void MacroAssembler::incrementq(Register reg, int value) {
5374 if (value == min_jint) { addq(reg, value); return; }
5375 if (value < 0) { decrementq(reg, -value); return; }
5376 if (value == 0) { ; return; }
5377 if (value == 1 && UseIncDec) { incq(reg) ; return; }
5378 /* else */ { addq(reg, value) ; return; }
5379 }
5381 void MacroAssembler::incrementq(Address dst, int value) {
5382 if (value == min_jint) { addq(dst, value); return; }
5383 if (value < 0) { decrementq(dst, -value); return; }
5384 if (value == 0) { ; return; }
5385 if (value == 1 && UseIncDec) { incq(dst) ; return; }
5386 /* else */ { addq(dst, value) ; return; }
5387 }
5389 // 32bit can do a case table jump in one instruction but we no longer allow the base
5390 // to be installed in the Address class
5391 void MacroAssembler::jump(ArrayAddress entry) {
5392 lea(rscratch1, entry.base());
5393 Address dispatch = entry.index();
5394 assert(dispatch._base == noreg, "must be");
5395 dispatch._base = rscratch1;
5396 jmp(dispatch);
5397 }
5399 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
5400 ShouldNotReachHere(); // 64bit doesn't use two regs
5401 cmpq(x_lo, y_lo);
5402 }
5404 void MacroAssembler::lea(Register dst, AddressLiteral src) {
5405 mov_literal64(dst, (intptr_t)src.target(), src.rspec());
5406 }
5408 void MacroAssembler::lea(Address dst, AddressLiteral adr) {
5409 mov_literal64(rscratch1, (intptr_t)adr.target(), adr.rspec());
5410 movptr(dst, rscratch1);
5411 }
5413 void MacroAssembler::leave() {
5414 // %%% is this really better? Why not on 32bit too?
5415 emit_byte(0xC9); // LEAVE
5416 }
5418 void MacroAssembler::lneg(Register hi, Register lo) {
5419 ShouldNotReachHere(); // 64bit doesn't use two regs
5420 negq(lo);
5421 }
5423 void MacroAssembler::movoop(Register dst, jobject obj) {
5424 mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate());
5425 }
5427 void MacroAssembler::movoop(Address dst, jobject obj) {
5428 mov_literal64(rscratch1, (intptr_t)obj, oop_Relocation::spec_for_immediate());
5429 movq(dst, rscratch1);
5430 }
5432 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
5433 if (src.is_lval()) {
5434 mov_literal64(dst, (intptr_t)src.target(), src.rspec());
5435 } else {
5436 if (reachable(src)) {
5437 movq(dst, as_Address(src));
5438 } else {
5439 lea(rscratch1, src);
5440 movq(dst, Address(rscratch1,0));
5441 }
5442 }
5443 }
5445 void MacroAssembler::movptr(ArrayAddress dst, Register src) {
5446 movq(as_Address(dst), src);
5447 }
5449 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
5450 movq(dst, as_Address(src));
5451 }
5453 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
5454 void MacroAssembler::movptr(Address dst, intptr_t src) {
5455 mov64(rscratch1, src);
5456 movq(dst, rscratch1);
5457 }
5459 // These are mostly for initializing NULL
5460 void MacroAssembler::movptr(Address dst, int32_t src) {
5461 movslq(dst, src);
5462 }
5464 void MacroAssembler::movptr(Register dst, int32_t src) {
5465 mov64(dst, (intptr_t)src);
5466 }
5468 void MacroAssembler::pushoop(jobject obj) {
5469 movoop(rscratch1, obj);
5470 push(rscratch1);
5471 }
5473 void MacroAssembler::pushptr(AddressLiteral src) {
5474 lea(rscratch1, src);
5475 if (src.is_lval()) {
5476 push(rscratch1);
5477 } else {
5478 pushq(Address(rscratch1, 0));
5479 }
5480 }
5482 void MacroAssembler::reset_last_Java_frame(bool clear_fp,
5483 bool clear_pc) {
5484 // we must set sp to zero to clear frame
5485 movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
5486 // must clear fp, so that compiled frames are not confused; it is
5487 // possible that we need it only for debugging
5488 if (clear_fp) {
5489 movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
5490 }
5492 if (clear_pc) {
5493 movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
5494 }
5495 }
5497 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
5498 Register last_java_fp,
5499 address last_java_pc) {
5500 // determine last_java_sp register
5501 if (!last_java_sp->is_valid()) {
5502 last_java_sp = rsp;
5503 }
5505 // last_java_fp is optional
5506 if (last_java_fp->is_valid()) {
5507 movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()),
5508 last_java_fp);
5509 }
5511 // last_java_pc is optional
5512 if (last_java_pc != NULL) {
5513 Address java_pc(r15_thread,
5514 JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset());
5515 lea(rscratch1, InternalAddress(last_java_pc));
5516 movptr(java_pc, rscratch1);
5517 }
5519 movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
5520 }
5522 static void pass_arg0(MacroAssembler* masm, Register arg) {
5523 if (c_rarg0 != arg ) {
5524 masm->mov(c_rarg0, arg);
5525 }
5526 }
5528 static void pass_arg1(MacroAssembler* masm, Register arg) {
5529 if (c_rarg1 != arg ) {
5530 masm->mov(c_rarg1, arg);
5531 }
5532 }
5534 static void pass_arg2(MacroAssembler* masm, Register arg) {
5535 if (c_rarg2 != arg ) {
5536 masm->mov(c_rarg2, arg);
5537 }
5538 }
5540 static void pass_arg3(MacroAssembler* masm, Register arg) {
5541 if (c_rarg3 != arg ) {
5542 masm->mov(c_rarg3, arg);
5543 }
5544 }
5546 void MacroAssembler::stop(const char* msg) {
5547 address rip = pc();
5548 pusha(); // get regs on stack
5549 lea(c_rarg0, ExternalAddress((address) msg));
5550 lea(c_rarg1, InternalAddress(rip));
5551 movq(c_rarg2, rsp); // pass pointer to regs array
5552 andq(rsp, -16); // align stack as required by ABI
5553 call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
5554 hlt();
5555 }
5557 void MacroAssembler::warn(const char* msg) {
5558 push(rsp);
5559 andq(rsp, -16); // align stack as required by push_CPU_state and call
5561 push_CPU_state(); // keeps alignment at 16 bytes
5562 lea(c_rarg0, ExternalAddress((address) msg));
5563 call_VM_leaf(CAST_FROM_FN_PTR(address, warning), c_rarg0);
5564 pop_CPU_state();
5565 pop(rsp);
5566 }
5568 #ifndef PRODUCT
5569 extern "C" void findpc(intptr_t x);
5570 #endif
5572 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[]) {
5573 // In order to get locks to work, we need to fake a in_VM state
5574 if (ShowMessageBoxOnError ) {
5575 JavaThread* thread = JavaThread::current();
5576 JavaThreadState saved_state = thread->thread_state();
5577 thread->set_thread_state(_thread_in_vm);
5578 #ifndef PRODUCT
5579 if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
5580 ttyLocker ttyl;
5581 BytecodeCounter::print();
5582 }
5583 #endif
5584 // To see where a verify_oop failed, get $ebx+40/X for this frame.
5585 // XXX correct this offset for amd64
5586 // This is the value of eip which points to where verify_oop will return.
5587 if (os::message_box(msg, "Execution stopped, print registers?")) {
5588 ttyLocker ttyl;
5589 tty->print_cr("rip = 0x%016lx", pc);
5590 #ifndef PRODUCT
5591 tty->cr();
5592 findpc(pc);
5593 tty->cr();
5594 #endif
5595 tty->print_cr("rax = 0x%016lx", regs[15]);
5596 tty->print_cr("rbx = 0x%016lx", regs[12]);
5597 tty->print_cr("rcx = 0x%016lx", regs[14]);
5598 tty->print_cr("rdx = 0x%016lx", regs[13]);
5599 tty->print_cr("rdi = 0x%016lx", regs[8]);
5600 tty->print_cr("rsi = 0x%016lx", regs[9]);
5601 tty->print_cr("rbp = 0x%016lx", regs[10]);
5602 tty->print_cr("rsp = 0x%016lx", regs[11]);
5603 tty->print_cr("r8 = 0x%016lx", regs[7]);
5604 tty->print_cr("r9 = 0x%016lx", regs[6]);
5605 tty->print_cr("r10 = 0x%016lx", regs[5]);
5606 tty->print_cr("r11 = 0x%016lx", regs[4]);
5607 tty->print_cr("r12 = 0x%016lx", regs[3]);
5608 tty->print_cr("r13 = 0x%016lx", regs[2]);
5609 tty->print_cr("r14 = 0x%016lx", regs[1]);
5610 tty->print_cr("r15 = 0x%016lx", regs[0]);
5611 BREAKPOINT;
5612 }
5613 ThreadStateTransition::transition(thread, _thread_in_vm, saved_state);
5614 } else {
5615 ttyLocker ttyl;
5616 ::tty->print_cr("=============== DEBUG MESSAGE: %s ================\n",
5617 msg);
5618 }
5619 }
5621 #endif // _LP64
5623 // Now versions that are common to 32/64 bit
5625 void MacroAssembler::addptr(Register dst, int32_t imm32) {
5626 LP64_ONLY(addq(dst, imm32)) NOT_LP64(addl(dst, imm32));
5627 }
5629 void MacroAssembler::addptr(Register dst, Register src) {
5630 LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src));
5631 }
5633 void MacroAssembler::addptr(Address dst, Register src) {
5634 LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src));
5635 }
5637 void MacroAssembler::align(int modulus) {
5638 if (offset() % modulus != 0) {
5639 nop(modulus - (offset() % modulus));
5640 }
5641 }
5643 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src) {
5644 if (reachable(src)) {
5645 andpd(dst, as_Address(src));
5646 } else {
5647 lea(rscratch1, src);
5648 andpd(dst, Address(rscratch1, 0));
5649 }
5650 }
5652 void MacroAssembler::andptr(Register dst, int32_t imm32) {
5653 LP64_ONLY(andq(dst, imm32)) NOT_LP64(andl(dst, imm32));
5654 }
5656 void MacroAssembler::atomic_incl(AddressLiteral counter_addr) {
5657 pushf();
5658 if (os::is_MP())
5659 lock();
5660 incrementl(counter_addr);
5661 popf();
5662 }
5664 // Writes to stack successive pages until offset reached to check for
5665 // stack overflow + shadow pages. This clobbers tmp.
5666 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
5667 movptr(tmp, rsp);
5668 // Bang stack for total size given plus shadow page size.
5669 // Bang one page at a time because large size can bang beyond yellow and
5670 // red zones.
5671 Label loop;
5672 bind(loop);
5673 movl(Address(tmp, (-os::vm_page_size())), size );
5674 subptr(tmp, os::vm_page_size());
5675 subl(size, os::vm_page_size());
5676 jcc(Assembler::greater, loop);
5678 // Bang down shadow pages too.
5679 // The -1 because we already subtracted 1 page.
5680 for (int i = 0; i< StackShadowPages-1; i++) {
5681 // this could be any sized move but this is can be a debugging crumb
5682 // so the bigger the better.
5683 movptr(Address(tmp, (-i*os::vm_page_size())), size );
5684 }
5685 }
5687 void MacroAssembler::biased_locking_exit(Register obj_reg, Register temp_reg, Label& done) {
5688 assert(UseBiasedLocking, "why call this otherwise?");
5690 // Check for biased locking unlock case, which is a no-op
5691 // Note: we do not have to check the thread ID for two reasons.
5692 // First, the interpreter checks for IllegalMonitorStateException at
5693 // a higher level. Second, if the bias was revoked while we held the
5694 // lock, the object could not be rebiased toward another thread, so
5695 // the bias bit would be clear.
5696 movptr(temp_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
5697 andptr(temp_reg, markOopDesc::biased_lock_mask_in_place);
5698 cmpptr(temp_reg, markOopDesc::biased_lock_pattern);
5699 jcc(Assembler::equal, done);
5700 }
5702 void MacroAssembler::c2bool(Register x) {
5703 // implements x == 0 ? 0 : 1
5704 // note: must only look at least-significant byte of x
5705 // since C-style booleans are stored in one byte
5706 // only! (was bug)
5707 andl(x, 0xFF);
5708 setb(Assembler::notZero, x);
5709 }
5711 // Wouldn't need if AddressLiteral version had new name
5712 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) {
5713 Assembler::call(L, rtype);
5714 }
5716 void MacroAssembler::call(Register entry) {
5717 Assembler::call(entry);
5718 }
5720 void MacroAssembler::call(AddressLiteral entry) {
5721 if (reachable(entry)) {
5722 Assembler::call_literal(entry.target(), entry.rspec());
5723 } else {
5724 lea(rscratch1, entry);
5725 Assembler::call(rscratch1);
5726 }
5727 }
5729 // Implementation of call_VM versions
5731 void MacroAssembler::call_VM(Register oop_result,
5732 address entry_point,
5733 bool check_exceptions) {
5734 Label C, E;
5735 call(C, relocInfo::none);
5736 jmp(E);
5738 bind(C);
5739 call_VM_helper(oop_result, entry_point, 0, check_exceptions);
5740 ret(0);
5742 bind(E);
5743 }
5745 void MacroAssembler::call_VM(Register oop_result,
5746 address entry_point,
5747 Register arg_1,
5748 bool check_exceptions) {
5749 Label C, E;
5750 call(C, relocInfo::none);
5751 jmp(E);
5753 bind(C);
5754 pass_arg1(this, arg_1);
5755 call_VM_helper(oop_result, entry_point, 1, check_exceptions);
5756 ret(0);
5758 bind(E);
5759 }
5761 void MacroAssembler::call_VM(Register oop_result,
5762 address entry_point,
5763 Register arg_1,
5764 Register arg_2,
5765 bool check_exceptions) {
5766 Label C, E;
5767 call(C, relocInfo::none);
5768 jmp(E);
5770 bind(C);
5772 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
5774 pass_arg2(this, arg_2);
5775 pass_arg1(this, arg_1);
5776 call_VM_helper(oop_result, entry_point, 2, check_exceptions);
5777 ret(0);
5779 bind(E);
5780 }
5782 void MacroAssembler::call_VM(Register oop_result,
5783 address entry_point,
5784 Register arg_1,
5785 Register arg_2,
5786 Register arg_3,
5787 bool check_exceptions) {
5788 Label C, E;
5789 call(C, relocInfo::none);
5790 jmp(E);
5792 bind(C);
5794 LP64_ONLY(assert(arg_1 != c_rarg3, "smashed arg"));
5795 LP64_ONLY(assert(arg_2 != c_rarg3, "smashed arg"));
5796 pass_arg3(this, arg_3);
5798 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
5799 pass_arg2(this, arg_2);
5801 pass_arg1(this, arg_1);
5802 call_VM_helper(oop_result, entry_point, 3, check_exceptions);
5803 ret(0);
5805 bind(E);
5806 }
5808 void MacroAssembler::call_VM(Register oop_result,
5809 Register last_java_sp,
5810 address entry_point,
5811 int number_of_arguments,
5812 bool check_exceptions) {
5813 Register thread = LP64_ONLY(r15_thread) NOT_LP64(noreg);
5814 call_VM_base(oop_result, thread, last_java_sp, entry_point, number_of_arguments, check_exceptions);
5815 }
5817 void MacroAssembler::call_VM(Register oop_result,
5818 Register last_java_sp,
5819 address entry_point,
5820 Register arg_1,
5821 bool check_exceptions) {
5822 pass_arg1(this, arg_1);
5823 call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
5824 }
5826 void MacroAssembler::call_VM(Register oop_result,
5827 Register last_java_sp,
5828 address entry_point,
5829 Register arg_1,
5830 Register arg_2,
5831 bool check_exceptions) {
5833 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
5834 pass_arg2(this, arg_2);
5835 pass_arg1(this, arg_1);
5836 call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
5837 }
5839 void MacroAssembler::call_VM(Register oop_result,
5840 Register last_java_sp,
5841 address entry_point,
5842 Register arg_1,
5843 Register arg_2,
5844 Register arg_3,
5845 bool check_exceptions) {
5846 LP64_ONLY(assert(arg_1 != c_rarg3, "smashed arg"));
5847 LP64_ONLY(assert(arg_2 != c_rarg3, "smashed arg"));
5848 pass_arg3(this, arg_3);
5849 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
5850 pass_arg2(this, arg_2);
5851 pass_arg1(this, arg_1);
5852 call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
5853 }
5855 void MacroAssembler::call_VM_base(Register oop_result,
5856 Register java_thread,
5857 Register last_java_sp,
5858 address entry_point,
5859 int number_of_arguments,
5860 bool check_exceptions) {
5861 // determine java_thread register
5862 if (!java_thread->is_valid()) {
5863 #ifdef _LP64
5864 java_thread = r15_thread;
5865 #else
5866 java_thread = rdi;
5867 get_thread(java_thread);
5868 #endif // LP64
5869 }
5870 // determine last_java_sp register
5871 if (!last_java_sp->is_valid()) {
5872 last_java_sp = rsp;
5873 }
5874 // debugging support
5875 assert(number_of_arguments >= 0 , "cannot have negative number of arguments");
5876 LP64_ONLY(assert(java_thread == r15_thread, "unexpected register"));
5877 #ifdef ASSERT
5878 LP64_ONLY(if (UseCompressedOops) verify_heapbase("call_VM_base");)
5879 #endif // ASSERT
5881 assert(java_thread != oop_result , "cannot use the same register for java_thread & oop_result");
5882 assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
5884 // push java thread (becomes first argument of C function)
5886 NOT_LP64(push(java_thread); number_of_arguments++);
5887 LP64_ONLY(mov(c_rarg0, r15_thread));
5889 // set last Java frame before call
5890 assert(last_java_sp != rbp, "can't use ebp/rbp");
5892 // Only interpreter should have to set fp
5893 set_last_Java_frame(java_thread, last_java_sp, rbp, NULL);
5895 // do the call, remove parameters
5896 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
5898 // restore the thread (cannot use the pushed argument since arguments
5899 // may be overwritten by C code generated by an optimizing compiler);
5900 // however can use the register value directly if it is callee saved.
5901 if (LP64_ONLY(true ||) java_thread == rdi || java_thread == rsi) {
5902 // rdi & rsi (also r15) are callee saved -> nothing to do
5903 #ifdef ASSERT
5904 guarantee(java_thread != rax, "change this code");
5905 push(rax);
5906 { Label L;
5907 get_thread(rax);
5908 cmpptr(java_thread, rax);
5909 jcc(Assembler::equal, L);
5910 stop("MacroAssembler::call_VM_base: rdi not callee saved?");
5911 bind(L);
5912 }
5913 pop(rax);
5914 #endif
5915 } else {
5916 get_thread(java_thread);
5917 }
5918 // reset last Java frame
5919 // Only interpreter should have to clear fp
5920 reset_last_Java_frame(java_thread, true, false);
5922 #ifndef CC_INTERP
5923 // C++ interp handles this in the interpreter
5924 check_and_handle_popframe(java_thread);
5925 check_and_handle_earlyret(java_thread);
5926 #endif /* CC_INTERP */
5928 if (check_exceptions) {
5929 // check for pending exceptions (java_thread is set upon return)
5930 cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
5931 #ifndef _LP64
5932 jump_cc(Assembler::notEqual,
5933 RuntimeAddress(StubRoutines::forward_exception_entry()));
5934 #else
5935 // This used to conditionally jump to forward_exception however it is
5936 // possible if we relocate that the branch will not reach. So we must jump
5937 // around so we can always reach
5939 Label ok;
5940 jcc(Assembler::equal, ok);
5941 jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
5942 bind(ok);
5943 #endif // LP64
5944 }
5946 // get oop result if there is one and reset the value in the thread
5947 if (oop_result->is_valid()) {
5948 movptr(oop_result, Address(java_thread, JavaThread::vm_result_offset()));
5949 movptr(Address(java_thread, JavaThread::vm_result_offset()), NULL_WORD);
5950 verify_oop(oop_result, "broken oop in call_VM_base");
5951 }
5952 }
5954 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
5956 // Calculate the value for last_Java_sp
5957 // somewhat subtle. call_VM does an intermediate call
5958 // which places a return address on the stack just under the
5959 // stack pointer as the user finsihed with it. This allows
5960 // use to retrieve last_Java_pc from last_Java_sp[-1].
5961 // On 32bit we then have to push additional args on the stack to accomplish
5962 // the actual requested call. On 64bit call_VM only can use register args
5963 // so the only extra space is the return address that call_VM created.
5964 // This hopefully explains the calculations here.
5966 #ifdef _LP64
5967 // We've pushed one address, correct last_Java_sp
5968 lea(rax, Address(rsp, wordSize));
5969 #else
5970 lea(rax, Address(rsp, (1 + number_of_arguments) * wordSize));
5971 #endif // LP64
5973 call_VM_base(oop_result, noreg, rax, entry_point, number_of_arguments, check_exceptions);
5975 }
5977 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
5978 call_VM_leaf_base(entry_point, number_of_arguments);
5979 }
5981 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
5982 pass_arg0(this, arg_0);
5983 call_VM_leaf(entry_point, 1);
5984 }
5986 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
5988 LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg"));
5989 pass_arg1(this, arg_1);
5990 pass_arg0(this, arg_0);
5991 call_VM_leaf(entry_point, 2);
5992 }
5994 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
5995 LP64_ONLY(assert(arg_0 != c_rarg2, "smashed arg"));
5996 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
5997 pass_arg2(this, arg_2);
5998 LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg"));
5999 pass_arg1(this, arg_1);
6000 pass_arg0(this, arg_0);
6001 call_VM_leaf(entry_point, 3);
6002 }
6004 void MacroAssembler::check_and_handle_earlyret(Register java_thread) {
6005 }
6007 void MacroAssembler::check_and_handle_popframe(Register java_thread) {
6008 }
6010 void MacroAssembler::cmp32(AddressLiteral src1, int32_t imm) {
6011 if (reachable(src1)) {
6012 cmpl(as_Address(src1), imm);
6013 } else {
6014 lea(rscratch1, src1);
6015 cmpl(Address(rscratch1, 0), imm);
6016 }
6017 }
6019 void MacroAssembler::cmp32(Register src1, AddressLiteral src2) {
6020 assert(!src2.is_lval(), "use cmpptr");
6021 if (reachable(src2)) {
6022 cmpl(src1, as_Address(src2));
6023 } else {
6024 lea(rscratch1, src2);
6025 cmpl(src1, Address(rscratch1, 0));
6026 }
6027 }
6029 void MacroAssembler::cmp32(Register src1, int32_t imm) {
6030 Assembler::cmpl(src1, imm);
6031 }
6033 void MacroAssembler::cmp32(Register src1, Address src2) {
6034 Assembler::cmpl(src1, src2);
6035 }
6037 void MacroAssembler::cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
6038 ucomisd(opr1, opr2);
6040 Label L;
6041 if (unordered_is_less) {
6042 movl(dst, -1);
6043 jcc(Assembler::parity, L);
6044 jcc(Assembler::below , L);
6045 movl(dst, 0);
6046 jcc(Assembler::equal , L);
6047 increment(dst);
6048 } else { // unordered is greater
6049 movl(dst, 1);
6050 jcc(Assembler::parity, L);
6051 jcc(Assembler::above , L);
6052 movl(dst, 0);
6053 jcc(Assembler::equal , L);
6054 decrementl(dst);
6055 }
6056 bind(L);
6057 }
6059 void MacroAssembler::cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
6060 ucomiss(opr1, opr2);
6062 Label L;
6063 if (unordered_is_less) {
6064 movl(dst, -1);
6065 jcc(Assembler::parity, L);
6066 jcc(Assembler::below , L);
6067 movl(dst, 0);
6068 jcc(Assembler::equal , L);
6069 increment(dst);
6070 } else { // unordered is greater
6071 movl(dst, 1);
6072 jcc(Assembler::parity, L);
6073 jcc(Assembler::above , L);
6074 movl(dst, 0);
6075 jcc(Assembler::equal , L);
6076 decrementl(dst);
6077 }
6078 bind(L);
6079 }
6082 void MacroAssembler::cmp8(AddressLiteral src1, int imm) {
6083 if (reachable(src1)) {
6084 cmpb(as_Address(src1), imm);
6085 } else {
6086 lea(rscratch1, src1);
6087 cmpb(Address(rscratch1, 0), imm);
6088 }
6089 }
6091 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2) {
6092 #ifdef _LP64
6093 if (src2.is_lval()) {
6094 movptr(rscratch1, src2);
6095 Assembler::cmpq(src1, rscratch1);
6096 } else if (reachable(src2)) {
6097 cmpq(src1, as_Address(src2));
6098 } else {
6099 lea(rscratch1, src2);
6100 Assembler::cmpq(src1, Address(rscratch1, 0));
6101 }
6102 #else
6103 if (src2.is_lval()) {
6104 cmp_literal32(src1, (int32_t) src2.target(), src2.rspec());
6105 } else {
6106 cmpl(src1, as_Address(src2));
6107 }
6108 #endif // _LP64
6109 }
6111 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2) {
6112 assert(src2.is_lval(), "not a mem-mem compare");
6113 #ifdef _LP64
6114 // moves src2's literal address
6115 movptr(rscratch1, src2);
6116 Assembler::cmpq(src1, rscratch1);
6117 #else
6118 cmp_literal32(src1, (int32_t) src2.target(), src2.rspec());
6119 #endif // _LP64
6120 }
6122 void MacroAssembler::locked_cmpxchgptr(Register reg, AddressLiteral adr) {
6123 if (reachable(adr)) {
6124 if (os::is_MP())
6125 lock();
6126 cmpxchgptr(reg, as_Address(adr));
6127 } else {
6128 lea(rscratch1, adr);
6129 if (os::is_MP())
6130 lock();
6131 cmpxchgptr(reg, Address(rscratch1, 0));
6132 }
6133 }
6135 void MacroAssembler::cmpxchgptr(Register reg, Address adr) {
6136 LP64_ONLY(cmpxchgq(reg, adr)) NOT_LP64(cmpxchgl(reg, adr));
6137 }
6139 void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src) {
6140 if (reachable(src)) {
6141 comisd(dst, as_Address(src));
6142 } else {
6143 lea(rscratch1, src);
6144 comisd(dst, Address(rscratch1, 0));
6145 }
6146 }
6148 void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src) {
6149 if (reachable(src)) {
6150 comiss(dst, as_Address(src));
6151 } else {
6152 lea(rscratch1, src);
6153 comiss(dst, Address(rscratch1, 0));
6154 }
6155 }
6158 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr) {
6159 Condition negated_cond = negate_condition(cond);
6160 Label L;
6161 jcc(negated_cond, L);
6162 atomic_incl(counter_addr);
6163 bind(L);
6164 }
6166 int MacroAssembler::corrected_idivl(Register reg) {
6167 // Full implementation of Java idiv and irem; checks for
6168 // special case as described in JVM spec., p.243 & p.271.
6169 // The function returns the (pc) offset of the idivl
6170 // instruction - may be needed for implicit exceptions.
6171 //
6172 // normal case special case
6173 //
6174 // input : rax,: dividend min_int
6175 // reg: divisor (may not be rax,/rdx) -1
6176 //
6177 // output: rax,: quotient (= rax, idiv reg) min_int
6178 // rdx: remainder (= rax, irem reg) 0
6179 assert(reg != rax && reg != rdx, "reg cannot be rax, or rdx register");
6180 const int min_int = 0x80000000;
6181 Label normal_case, special_case;
6183 // check for special case
6184 cmpl(rax, min_int);
6185 jcc(Assembler::notEqual, normal_case);
6186 xorl(rdx, rdx); // prepare rdx for possible special case (where remainder = 0)
6187 cmpl(reg, -1);
6188 jcc(Assembler::equal, special_case);
6190 // handle normal case
6191 bind(normal_case);
6192 cdql();
6193 int idivl_offset = offset();
6194 idivl(reg);
6196 // normal and special case exit
6197 bind(special_case);
6199 return idivl_offset;
6200 }
6204 void MacroAssembler::decrementl(Register reg, int value) {
6205 if (value == min_jint) {subl(reg, value) ; return; }
6206 if (value < 0) { incrementl(reg, -value); return; }
6207 if (value == 0) { ; return; }
6208 if (value == 1 && UseIncDec) { decl(reg) ; return; }
6209 /* else */ { subl(reg, value) ; return; }
6210 }
6212 void MacroAssembler::decrementl(Address dst, int value) {
6213 if (value == min_jint) {subl(dst, value) ; return; }
6214 if (value < 0) { incrementl(dst, -value); return; }
6215 if (value == 0) { ; return; }
6216 if (value == 1 && UseIncDec) { decl(dst) ; return; }
6217 /* else */ { subl(dst, value) ; return; }
6218 }
6220 void MacroAssembler::division_with_shift (Register reg, int shift_value) {
6221 assert (shift_value > 0, "illegal shift value");
6222 Label _is_positive;
6223 testl (reg, reg);
6224 jcc (Assembler::positive, _is_positive);
6225 int offset = (1 << shift_value) - 1 ;
6227 if (offset == 1) {
6228 incrementl(reg);
6229 } else {
6230 addl(reg, offset);
6231 }
6233 bind (_is_positive);
6234 sarl(reg, shift_value);
6235 }
6237 // !defined(COMPILER2) is because of stupid core builds
6238 #if !defined(_LP64) || defined(COMPILER1) || !defined(COMPILER2)
6239 void MacroAssembler::empty_FPU_stack() {
6240 if (VM_Version::supports_mmx()) {
6241 emms();
6242 } else {
6243 for (int i = 8; i-- > 0; ) ffree(i);
6244 }
6245 }
6246 #endif // !LP64 || C1 || !C2
6249 // Defines obj, preserves var_size_in_bytes
6250 void MacroAssembler::eden_allocate(Register obj,
6251 Register var_size_in_bytes,
6252 int con_size_in_bytes,
6253 Register t1,
6254 Label& slow_case) {
6255 assert(obj == rax, "obj must be in rax, for cmpxchg");
6256 assert_different_registers(obj, var_size_in_bytes, t1);
6257 if (CMSIncrementalMode || !Universe::heap()->supports_inline_contig_alloc()) {
6258 jmp(slow_case);
6259 } else {
6260 Register end = t1;
6261 Label retry;
6262 bind(retry);
6263 ExternalAddress heap_top((address) Universe::heap()->top_addr());
6264 movptr(obj, heap_top);
6265 if (var_size_in_bytes == noreg) {
6266 lea(end, Address(obj, con_size_in_bytes));
6267 } else {
6268 lea(end, Address(obj, var_size_in_bytes, Address::times_1));
6269 }
6270 // if end < obj then we wrapped around => object too long => slow case
6271 cmpptr(end, obj);
6272 jcc(Assembler::below, slow_case);
6273 cmpptr(end, ExternalAddress((address) Universe::heap()->end_addr()));
6274 jcc(Assembler::above, slow_case);
6275 // Compare obj with the top addr, and if still equal, store the new top addr in
6276 // end at the address of the top addr pointer. Sets ZF if was equal, and clears
6277 // it otherwise. Use lock prefix for atomicity on MPs.
6278 locked_cmpxchgptr(end, heap_top);
6279 jcc(Assembler::notEqual, retry);
6280 }
6281 }
6283 void MacroAssembler::enter() {
6284 push(rbp);
6285 mov(rbp, rsp);
6286 }
6288 void MacroAssembler::fcmp(Register tmp) {
6289 fcmp(tmp, 1, true, true);
6290 }
6292 void MacroAssembler::fcmp(Register tmp, int index, bool pop_left, bool pop_right) {
6293 assert(!pop_right || pop_left, "usage error");
6294 if (VM_Version::supports_cmov()) {
6295 assert(tmp == noreg, "unneeded temp");
6296 if (pop_left) {
6297 fucomip(index);
6298 } else {
6299 fucomi(index);
6300 }
6301 if (pop_right) {
6302 fpop();
6303 }
6304 } else {
6305 assert(tmp != noreg, "need temp");
6306 if (pop_left) {
6307 if (pop_right) {
6308 fcompp();
6309 } else {
6310 fcomp(index);
6311 }
6312 } else {
6313 fcom(index);
6314 }
6315 // convert FPU condition into eflags condition via rax,
6316 save_rax(tmp);
6317 fwait(); fnstsw_ax();
6318 sahf();
6319 restore_rax(tmp);
6320 }
6321 // condition codes set as follows:
6322 //
6323 // CF (corresponds to C0) if x < y
6324 // PF (corresponds to C2) if unordered
6325 // ZF (corresponds to C3) if x = y
6326 }
6328 void MacroAssembler::fcmp2int(Register dst, bool unordered_is_less) {
6329 fcmp2int(dst, unordered_is_less, 1, true, true);
6330 }
6332 void MacroAssembler::fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right) {
6333 fcmp(VM_Version::supports_cmov() ? noreg : dst, index, pop_left, pop_right);
6334 Label L;
6335 if (unordered_is_less) {
6336 movl(dst, -1);
6337 jcc(Assembler::parity, L);
6338 jcc(Assembler::below , L);
6339 movl(dst, 0);
6340 jcc(Assembler::equal , L);
6341 increment(dst);
6342 } else { // unordered is greater
6343 movl(dst, 1);
6344 jcc(Assembler::parity, L);
6345 jcc(Assembler::above , L);
6346 movl(dst, 0);
6347 jcc(Assembler::equal , L);
6348 decrementl(dst);
6349 }
6350 bind(L);
6351 }
6353 void MacroAssembler::fld_d(AddressLiteral src) {
6354 fld_d(as_Address(src));
6355 }
6357 void MacroAssembler::fld_s(AddressLiteral src) {
6358 fld_s(as_Address(src));
6359 }
6361 void MacroAssembler::fld_x(AddressLiteral src) {
6362 Assembler::fld_x(as_Address(src));
6363 }
6365 void MacroAssembler::fldcw(AddressLiteral src) {
6366 Assembler::fldcw(as_Address(src));
6367 }
6369 void MacroAssembler::fpop() {
6370 ffree();
6371 fincstp();
6372 }
6374 void MacroAssembler::fremr(Register tmp) {
6375 save_rax(tmp);
6376 { Label L;
6377 bind(L);
6378 fprem();
6379 fwait(); fnstsw_ax();
6380 #ifdef _LP64
6381 testl(rax, 0x400);
6382 jcc(Assembler::notEqual, L);
6383 #else
6384 sahf();
6385 jcc(Assembler::parity, L);
6386 #endif // _LP64
6387 }
6388 restore_rax(tmp);
6389 // Result is in ST0.
6390 // Note: fxch & fpop to get rid of ST1
6391 // (otherwise FPU stack could overflow eventually)
6392 fxch(1);
6393 fpop();
6394 }
6397 void MacroAssembler::incrementl(AddressLiteral dst) {
6398 if (reachable(dst)) {
6399 incrementl(as_Address(dst));
6400 } else {
6401 lea(rscratch1, dst);
6402 incrementl(Address(rscratch1, 0));
6403 }
6404 }
6406 void MacroAssembler::incrementl(ArrayAddress dst) {
6407 incrementl(as_Address(dst));
6408 }
6410 void MacroAssembler::incrementl(Register reg, int value) {
6411 if (value == min_jint) {addl(reg, value) ; return; }
6412 if (value < 0) { decrementl(reg, -value); return; }
6413 if (value == 0) { ; return; }
6414 if (value == 1 && UseIncDec) { incl(reg) ; return; }
6415 /* else */ { addl(reg, value) ; return; }
6416 }
6418 void MacroAssembler::incrementl(Address dst, int value) {
6419 if (value == min_jint) {addl(dst, value) ; return; }
6420 if (value < 0) { decrementl(dst, -value); return; }
6421 if (value == 0) { ; return; }
6422 if (value == 1 && UseIncDec) { incl(dst) ; return; }
6423 /* else */ { addl(dst, value) ; return; }
6424 }
6426 void MacroAssembler::jump(AddressLiteral dst) {
6427 if (reachable(dst)) {
6428 jmp_literal(dst.target(), dst.rspec());
6429 } else {
6430 lea(rscratch1, dst);
6431 jmp(rscratch1);
6432 }
6433 }
6435 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst) {
6436 if (reachable(dst)) {
6437 InstructionMark im(this);
6438 relocate(dst.reloc());
6439 const int short_size = 2;
6440 const int long_size = 6;
6441 int offs = (intptr_t)dst.target() - ((intptr_t)_code_pos);
6442 if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) {
6443 // 0111 tttn #8-bit disp
6444 emit_byte(0x70 | cc);
6445 emit_byte((offs - short_size) & 0xFF);
6446 } else {
6447 // 0000 1111 1000 tttn #32-bit disp
6448 emit_byte(0x0F);
6449 emit_byte(0x80 | cc);
6450 emit_long(offs - long_size);
6451 }
6452 } else {
6453 #ifdef ASSERT
6454 warning("reversing conditional branch");
6455 #endif /* ASSERT */
6456 Label skip;
6457 jccb(reverse[cc], skip);
6458 lea(rscratch1, dst);
6459 Assembler::jmp(rscratch1);
6460 bind(skip);
6461 }
6462 }
6464 void MacroAssembler::ldmxcsr(AddressLiteral src) {
6465 if (reachable(src)) {
6466 Assembler::ldmxcsr(as_Address(src));
6467 } else {
6468 lea(rscratch1, src);
6469 Assembler::ldmxcsr(Address(rscratch1, 0));
6470 }
6471 }
6473 int MacroAssembler::load_signed_byte(Register dst, Address src) {
6474 int off;
6475 if (LP64_ONLY(true ||) VM_Version::is_P6()) {
6476 off = offset();
6477 movsbl(dst, src); // movsxb
6478 } else {
6479 off = load_unsigned_byte(dst, src);
6480 shll(dst, 24);
6481 sarl(dst, 24);
6482 }
6483 return off;
6484 }
6486 // Note: load_signed_short used to be called load_signed_word.
6487 // Although the 'w' in x86 opcodes refers to the term "word" in the assembler
6488 // manual, which means 16 bits, that usage is found nowhere in HotSpot code.
6489 // The term "word" in HotSpot means a 32- or 64-bit machine word.
6490 int MacroAssembler::load_signed_short(Register dst, Address src) {
6491 int off;
6492 if (LP64_ONLY(true ||) VM_Version::is_P6()) {
6493 // This is dubious to me since it seems safe to do a signed 16 => 64 bit
6494 // version but this is what 64bit has always done. This seems to imply
6495 // that users are only using 32bits worth.
6496 off = offset();
6497 movswl(dst, src); // movsxw
6498 } else {
6499 off = load_unsigned_short(dst, src);
6500 shll(dst, 16);
6501 sarl(dst, 16);
6502 }
6503 return off;
6504 }
6506 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
6507 // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
6508 // and "3.9 Partial Register Penalties", p. 22).
6509 int off;
6510 if (LP64_ONLY(true || ) VM_Version::is_P6() || src.uses(dst)) {
6511 off = offset();
6512 movzbl(dst, src); // movzxb
6513 } else {
6514 xorl(dst, dst);
6515 off = offset();
6516 movb(dst, src);
6517 }
6518 return off;
6519 }
6521 // Note: load_unsigned_short used to be called load_unsigned_word.
6522 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
6523 // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
6524 // and "3.9 Partial Register Penalties", p. 22).
6525 int off;
6526 if (LP64_ONLY(true ||) VM_Version::is_P6() || src.uses(dst)) {
6527 off = offset();
6528 movzwl(dst, src); // movzxw
6529 } else {
6530 xorl(dst, dst);
6531 off = offset();
6532 movw(dst, src);
6533 }
6534 return off;
6535 }
6537 void MacroAssembler::load_sized_value(Register dst, Address src,
6538 size_t size_in_bytes, bool is_signed) {
6539 switch (size_in_bytes) {
6540 #ifndef _LP64
6541 // For case 8, caller is responsible for manually loading
6542 // the second word into another register.
6543 case 8: movl(dst, src); break;
6544 #else
6545 case 8: movq(dst, src); break;
6546 #endif
6547 case 4: movl(dst, src); break;
6548 case 2: is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
6549 case 1: is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
6550 default: ShouldNotReachHere();
6551 }
6552 }
6554 void MacroAssembler::mov32(AddressLiteral dst, Register src) {
6555 if (reachable(dst)) {
6556 movl(as_Address(dst), src);
6557 } else {
6558 lea(rscratch1, dst);
6559 movl(Address(rscratch1, 0), src);
6560 }
6561 }
6563 void MacroAssembler::mov32(Register dst, AddressLiteral src) {
6564 if (reachable(src)) {
6565 movl(dst, as_Address(src));
6566 } else {
6567 lea(rscratch1, src);
6568 movl(dst, Address(rscratch1, 0));
6569 }
6570 }
6572 // C++ bool manipulation
6574 void MacroAssembler::movbool(Register dst, Address src) {
6575 if(sizeof(bool) == 1)
6576 movb(dst, src);
6577 else if(sizeof(bool) == 2)
6578 movw(dst, src);
6579 else if(sizeof(bool) == 4)
6580 movl(dst, src);
6581 else
6582 // unsupported
6583 ShouldNotReachHere();
6584 }
6586 void MacroAssembler::movbool(Address dst, bool boolconst) {
6587 if(sizeof(bool) == 1)
6588 movb(dst, (int) boolconst);
6589 else if(sizeof(bool) == 2)
6590 movw(dst, (int) boolconst);
6591 else if(sizeof(bool) == 4)
6592 movl(dst, (int) boolconst);
6593 else
6594 // unsupported
6595 ShouldNotReachHere();
6596 }
6598 void MacroAssembler::movbool(Address dst, Register src) {
6599 if(sizeof(bool) == 1)
6600 movb(dst, src);
6601 else if(sizeof(bool) == 2)
6602 movw(dst, src);
6603 else if(sizeof(bool) == 4)
6604 movl(dst, src);
6605 else
6606 // unsupported
6607 ShouldNotReachHere();
6608 }
6610 void MacroAssembler::movbyte(ArrayAddress dst, int src) {
6611 movb(as_Address(dst), src);
6612 }
6614 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src) {
6615 if (reachable(src)) {
6616 if (UseXmmLoadAndClearUpper) {
6617 movsd (dst, as_Address(src));
6618 } else {
6619 movlpd(dst, as_Address(src));
6620 }
6621 } else {
6622 lea(rscratch1, src);
6623 if (UseXmmLoadAndClearUpper) {
6624 movsd (dst, Address(rscratch1, 0));
6625 } else {
6626 movlpd(dst, Address(rscratch1, 0));
6627 }
6628 }
6629 }
6631 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src) {
6632 if (reachable(src)) {
6633 movss(dst, as_Address(src));
6634 } else {
6635 lea(rscratch1, src);
6636 movss(dst, Address(rscratch1, 0));
6637 }
6638 }
6640 void MacroAssembler::movptr(Register dst, Register src) {
6641 LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src));
6642 }
6644 void MacroAssembler::movptr(Register dst, Address src) {
6645 LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src));
6646 }
6648 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
6649 void MacroAssembler::movptr(Register dst, intptr_t src) {
6650 LP64_ONLY(mov64(dst, src)) NOT_LP64(movl(dst, src));
6651 }
6653 void MacroAssembler::movptr(Address dst, Register src) {
6654 LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src));
6655 }
6657 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src) {
6658 if (reachable(src)) {
6659 movss(dst, as_Address(src));
6660 } else {
6661 lea(rscratch1, src);
6662 movss(dst, Address(rscratch1, 0));
6663 }
6664 }
6666 void MacroAssembler::null_check(Register reg, int offset) {
6667 if (needs_explicit_null_check(offset)) {
6668 // provoke OS NULL exception if reg = NULL by
6669 // accessing M[reg] w/o changing any (non-CC) registers
6670 // NOTE: cmpl is plenty here to provoke a segv
6671 cmpptr(rax, Address(reg, 0));
6672 // Note: should probably use testl(rax, Address(reg, 0));
6673 // may be shorter code (however, this version of
6674 // testl needs to be implemented first)
6675 } else {
6676 // nothing to do, (later) access of M[reg + offset]
6677 // will provoke OS NULL exception if reg = NULL
6678 }
6679 }
6681 void MacroAssembler::os_breakpoint() {
6682 // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
6683 // (e.g., MSVC can't call ps() otherwise)
6684 call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
6685 }
6687 void MacroAssembler::pop_CPU_state() {
6688 pop_FPU_state();
6689 pop_IU_state();
6690 }
6692 void MacroAssembler::pop_FPU_state() {
6693 NOT_LP64(frstor(Address(rsp, 0));)
6694 LP64_ONLY(fxrstor(Address(rsp, 0));)
6695 addptr(rsp, FPUStateSizeInWords * wordSize);
6696 }
6698 void MacroAssembler::pop_IU_state() {
6699 popa();
6700 LP64_ONLY(addq(rsp, 8));
6701 popf();
6702 }
6704 // Save Integer and Float state
6705 // Warning: Stack must be 16 byte aligned (64bit)
6706 void MacroAssembler::push_CPU_state() {
6707 push_IU_state();
6708 push_FPU_state();
6709 }
6711 void MacroAssembler::push_FPU_state() {
6712 subptr(rsp, FPUStateSizeInWords * wordSize);
6713 #ifndef _LP64
6714 fnsave(Address(rsp, 0));
6715 fwait();
6716 #else
6717 fxsave(Address(rsp, 0));
6718 #endif // LP64
6719 }
6721 void MacroAssembler::push_IU_state() {
6722 // Push flags first because pusha kills them
6723 pushf();
6724 // Make sure rsp stays 16-byte aligned
6725 LP64_ONLY(subq(rsp, 8));
6726 pusha();
6727 }
6729 void MacroAssembler::reset_last_Java_frame(Register java_thread, bool clear_fp, bool clear_pc) {
6730 // determine java_thread register
6731 if (!java_thread->is_valid()) {
6732 java_thread = rdi;
6733 get_thread(java_thread);
6734 }
6735 // we must set sp to zero to clear frame
6736 movptr(Address(java_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
6737 if (clear_fp) {
6738 movptr(Address(java_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
6739 }
6741 if (clear_pc)
6742 movptr(Address(java_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
6744 }
6746 void MacroAssembler::restore_rax(Register tmp) {
6747 if (tmp == noreg) pop(rax);
6748 else if (tmp != rax) mov(rax, tmp);
6749 }
6751 void MacroAssembler::round_to(Register reg, int modulus) {
6752 addptr(reg, modulus - 1);
6753 andptr(reg, -modulus);
6754 }
6756 void MacroAssembler::save_rax(Register tmp) {
6757 if (tmp == noreg) push(rax);
6758 else if (tmp != rax) mov(tmp, rax);
6759 }
6761 // Write serialization page so VM thread can do a pseudo remote membar.
6762 // We use the current thread pointer to calculate a thread specific
6763 // offset to write to within the page. This minimizes bus traffic
6764 // due to cache line collision.
6765 void MacroAssembler::serialize_memory(Register thread, Register tmp) {
6766 movl(tmp, thread);
6767 shrl(tmp, os::get_serialize_page_shift_count());
6768 andl(tmp, (os::vm_page_size() - sizeof(int)));
6770 Address index(noreg, tmp, Address::times_1);
6771 ExternalAddress page(os::get_memory_serialize_page());
6773 // Size of store must match masking code above
6774 movl(as_Address(ArrayAddress(page, index)), tmp);
6775 }
6777 // Calls to C land
6778 //
6779 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded
6780 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
6781 // has to be reset to 0. This is required to allow proper stack traversal.
6782 void MacroAssembler::set_last_Java_frame(Register java_thread,
6783 Register last_java_sp,
6784 Register last_java_fp,
6785 address last_java_pc) {
6786 // determine java_thread register
6787 if (!java_thread->is_valid()) {
6788 java_thread = rdi;
6789 get_thread(java_thread);
6790 }
6791 // determine last_java_sp register
6792 if (!last_java_sp->is_valid()) {
6793 last_java_sp = rsp;
6794 }
6796 // last_java_fp is optional
6798 if (last_java_fp->is_valid()) {
6799 movptr(Address(java_thread, JavaThread::last_Java_fp_offset()), last_java_fp);
6800 }
6802 // last_java_pc is optional
6804 if (last_java_pc != NULL) {
6805 lea(Address(java_thread,
6806 JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset()),
6807 InternalAddress(last_java_pc));
6809 }
6810 movptr(Address(java_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
6811 }
6813 void MacroAssembler::shlptr(Register dst, int imm8) {
6814 LP64_ONLY(shlq(dst, imm8)) NOT_LP64(shll(dst, imm8));
6815 }
6817 void MacroAssembler::shrptr(Register dst, int imm8) {
6818 LP64_ONLY(shrq(dst, imm8)) NOT_LP64(shrl(dst, imm8));
6819 }
6821 void MacroAssembler::sign_extend_byte(Register reg) {
6822 if (LP64_ONLY(true ||) (VM_Version::is_P6() && reg->has_byte_register())) {
6823 movsbl(reg, reg); // movsxb
6824 } else {
6825 shll(reg, 24);
6826 sarl(reg, 24);
6827 }
6828 }
6830 void MacroAssembler::sign_extend_short(Register reg) {
6831 if (LP64_ONLY(true ||) VM_Version::is_P6()) {
6832 movswl(reg, reg); // movsxw
6833 } else {
6834 shll(reg, 16);
6835 sarl(reg, 16);
6836 }
6837 }
6839 //////////////////////////////////////////////////////////////////////////////////
6840 #ifndef SERIALGC
6842 void MacroAssembler::g1_write_barrier_pre(Register obj,
6843 #ifndef _LP64
6844 Register thread,
6845 #endif
6846 Register tmp,
6847 Register tmp2,
6848 bool tosca_live) {
6849 LP64_ONLY(Register thread = r15_thread;)
6850 Address in_progress(thread, in_bytes(JavaThread::satb_mark_queue_offset() +
6851 PtrQueue::byte_offset_of_active()));
6853 Address index(thread, in_bytes(JavaThread::satb_mark_queue_offset() +
6854 PtrQueue::byte_offset_of_index()));
6855 Address buffer(thread, in_bytes(JavaThread::satb_mark_queue_offset() +
6856 PtrQueue::byte_offset_of_buf()));
6859 Label done;
6860 Label runtime;
6862 // if (!marking_in_progress) goto done;
6863 if (in_bytes(PtrQueue::byte_width_of_active()) == 4) {
6864 cmpl(in_progress, 0);
6865 } else {
6866 assert(in_bytes(PtrQueue::byte_width_of_active()) == 1, "Assumption");
6867 cmpb(in_progress, 0);
6868 }
6869 jcc(Assembler::equal, done);
6871 // if (x.f == NULL) goto done;
6872 #ifdef _LP64
6873 load_heap_oop(tmp2, Address(obj, 0));
6874 #else
6875 movptr(tmp2, Address(obj, 0));
6876 #endif
6877 cmpptr(tmp2, (int32_t) NULL_WORD);
6878 jcc(Assembler::equal, done);
6880 // Can we store original value in the thread's buffer?
6882 #ifdef _LP64
6883 movslq(tmp, index);
6884 cmpq(tmp, 0);
6885 #else
6886 cmpl(index, 0);
6887 #endif
6888 jcc(Assembler::equal, runtime);
6889 #ifdef _LP64
6890 subq(tmp, wordSize);
6891 movl(index, tmp);
6892 addq(tmp, buffer);
6893 #else
6894 subl(index, wordSize);
6895 movl(tmp, buffer);
6896 addl(tmp, index);
6897 #endif
6898 movptr(Address(tmp, 0), tmp2);
6899 jmp(done);
6900 bind(runtime);
6901 // save the live input values
6902 if(tosca_live) push(rax);
6903 push(obj);
6904 #ifdef _LP64
6905 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), tmp2, r15_thread);
6906 #else
6907 push(thread);
6908 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), tmp2, thread);
6909 pop(thread);
6910 #endif
6911 pop(obj);
6912 if(tosca_live) pop(rax);
6913 bind(done);
6915 }
6917 void MacroAssembler::g1_write_barrier_post(Register store_addr,
6918 Register new_val,
6919 #ifndef _LP64
6920 Register thread,
6921 #endif
6922 Register tmp,
6923 Register tmp2) {
6925 LP64_ONLY(Register thread = r15_thread;)
6926 Address queue_index(thread, in_bytes(JavaThread::dirty_card_queue_offset() +
6927 PtrQueue::byte_offset_of_index()));
6928 Address buffer(thread, in_bytes(JavaThread::dirty_card_queue_offset() +
6929 PtrQueue::byte_offset_of_buf()));
6930 BarrierSet* bs = Universe::heap()->barrier_set();
6931 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
6932 Label done;
6933 Label runtime;
6935 // Does store cross heap regions?
6937 movptr(tmp, store_addr);
6938 xorptr(tmp, new_val);
6939 shrptr(tmp, HeapRegion::LogOfHRGrainBytes);
6940 jcc(Assembler::equal, done);
6942 // crosses regions, storing NULL?
6944 cmpptr(new_val, (int32_t) NULL_WORD);
6945 jcc(Assembler::equal, done);
6947 // storing region crossing non-NULL, is card already dirty?
6949 ExternalAddress cardtable((address) ct->byte_map_base);
6950 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
6951 #ifdef _LP64
6952 const Register card_addr = tmp;
6954 movq(card_addr, store_addr);
6955 shrq(card_addr, CardTableModRefBS::card_shift);
6957 lea(tmp2, cardtable);
6959 // get the address of the card
6960 addq(card_addr, tmp2);
6961 #else
6962 const Register card_index = tmp;
6964 movl(card_index, store_addr);
6965 shrl(card_index, CardTableModRefBS::card_shift);
6967 Address index(noreg, card_index, Address::times_1);
6968 const Register card_addr = tmp;
6969 lea(card_addr, as_Address(ArrayAddress(cardtable, index)));
6970 #endif
6971 cmpb(Address(card_addr, 0), 0);
6972 jcc(Assembler::equal, done);
6974 // storing a region crossing, non-NULL oop, card is clean.
6975 // dirty card and log.
6977 movb(Address(card_addr, 0), 0);
6979 cmpl(queue_index, 0);
6980 jcc(Assembler::equal, runtime);
6981 subl(queue_index, wordSize);
6982 movptr(tmp2, buffer);
6983 #ifdef _LP64
6984 movslq(rscratch1, queue_index);
6985 addq(tmp2, rscratch1);
6986 movq(Address(tmp2, 0), card_addr);
6987 #else
6988 addl(tmp2, queue_index);
6989 movl(Address(tmp2, 0), card_index);
6990 #endif
6991 jmp(done);
6993 bind(runtime);
6994 // save the live input values
6995 push(store_addr);
6996 push(new_val);
6997 #ifdef _LP64
6998 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_post), card_addr, r15_thread);
6999 #else
7000 push(thread);
7001 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_post), card_addr, thread);
7002 pop(thread);
7003 #endif
7004 pop(new_val);
7005 pop(store_addr);
7007 bind(done);
7009 }
7011 #endif // SERIALGC
7012 //////////////////////////////////////////////////////////////////////////////////
7015 void MacroAssembler::store_check(Register obj) {
7016 // Does a store check for the oop in register obj. The content of
7017 // register obj is destroyed afterwards.
7018 store_check_part_1(obj);
7019 store_check_part_2(obj);
7020 }
7022 void MacroAssembler::store_check(Register obj, Address dst) {
7023 store_check(obj);
7024 }
7027 // split the store check operation so that other instructions can be scheduled inbetween
7028 void MacroAssembler::store_check_part_1(Register obj) {
7029 BarrierSet* bs = Universe::heap()->barrier_set();
7030 assert(bs->kind() == BarrierSet::CardTableModRef, "Wrong barrier set kind");
7031 shrptr(obj, CardTableModRefBS::card_shift);
7032 }
7034 void MacroAssembler::store_check_part_2(Register obj) {
7035 BarrierSet* bs = Universe::heap()->barrier_set();
7036 assert(bs->kind() == BarrierSet::CardTableModRef, "Wrong barrier set kind");
7037 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
7038 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
7040 // The calculation for byte_map_base is as follows:
7041 // byte_map_base = _byte_map - (uintptr_t(low_bound) >> card_shift);
7042 // So this essentially converts an address to a displacement and
7043 // it will never need to be relocated. On 64bit however the value may be too
7044 // large for a 32bit displacement
7046 intptr_t disp = (intptr_t) ct->byte_map_base;
7047 if (is_simm32(disp)) {
7048 Address cardtable(noreg, obj, Address::times_1, disp);
7049 movb(cardtable, 0);
7050 } else {
7051 // By doing it as an ExternalAddress disp could be converted to a rip-relative
7052 // displacement and done in a single instruction given favorable mapping and
7053 // a smarter version of as_Address. Worst case it is two instructions which
7054 // is no worse off then loading disp into a register and doing as a simple
7055 // Address() as above.
7056 // We can't do as ExternalAddress as the only style since if disp == 0 we'll
7057 // assert since NULL isn't acceptable in a reloci (see 6644928). In any case
7058 // in some cases we'll get a single instruction version.
7060 ExternalAddress cardtable((address)disp);
7061 Address index(noreg, obj, Address::times_1);
7062 movb(as_Address(ArrayAddress(cardtable, index)), 0);
7063 }
7064 }
7066 void MacroAssembler::subptr(Register dst, int32_t imm32) {
7067 LP64_ONLY(subq(dst, imm32)) NOT_LP64(subl(dst, imm32));
7068 }
7070 void MacroAssembler::subptr(Register dst, Register src) {
7071 LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src));
7072 }
7074 void MacroAssembler::test32(Register src1, AddressLiteral src2) {
7075 // src2 must be rval
7077 if (reachable(src2)) {
7078 testl(src1, as_Address(src2));
7079 } else {
7080 lea(rscratch1, src2);
7081 testl(src1, Address(rscratch1, 0));
7082 }
7083 }
7085 // C++ bool manipulation
7086 void MacroAssembler::testbool(Register dst) {
7087 if(sizeof(bool) == 1)
7088 testb(dst, 0xff);
7089 else if(sizeof(bool) == 2) {
7090 // testw implementation needed for two byte bools
7091 ShouldNotReachHere();
7092 } else if(sizeof(bool) == 4)
7093 testl(dst, dst);
7094 else
7095 // unsupported
7096 ShouldNotReachHere();
7097 }
7099 void MacroAssembler::testptr(Register dst, Register src) {
7100 LP64_ONLY(testq(dst, src)) NOT_LP64(testl(dst, src));
7101 }
7103 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
7104 void MacroAssembler::tlab_allocate(Register obj,
7105 Register var_size_in_bytes,
7106 int con_size_in_bytes,
7107 Register t1,
7108 Register t2,
7109 Label& slow_case) {
7110 assert_different_registers(obj, t1, t2);
7111 assert_different_registers(obj, var_size_in_bytes, t1);
7112 Register end = t2;
7113 Register thread = NOT_LP64(t1) LP64_ONLY(r15_thread);
7115 verify_tlab();
7117 NOT_LP64(get_thread(thread));
7119 movptr(obj, Address(thread, JavaThread::tlab_top_offset()));
7120 if (var_size_in_bytes == noreg) {
7121 lea(end, Address(obj, con_size_in_bytes));
7122 } else {
7123 lea(end, Address(obj, var_size_in_bytes, Address::times_1));
7124 }
7125 cmpptr(end, Address(thread, JavaThread::tlab_end_offset()));
7126 jcc(Assembler::above, slow_case);
7128 // update the tlab top pointer
7129 movptr(Address(thread, JavaThread::tlab_top_offset()), end);
7131 // recover var_size_in_bytes if necessary
7132 if (var_size_in_bytes == end) {
7133 subptr(var_size_in_bytes, obj);
7134 }
7135 verify_tlab();
7136 }
7138 // Preserves rbx, and rdx.
7139 void MacroAssembler::tlab_refill(Label& retry,
7140 Label& try_eden,
7141 Label& slow_case) {
7142 Register top = rax;
7143 Register t1 = rcx;
7144 Register t2 = rsi;
7145 Register thread_reg = NOT_LP64(rdi) LP64_ONLY(r15_thread);
7146 assert_different_registers(top, thread_reg, t1, t2, /* preserve: */ rbx, rdx);
7147 Label do_refill, discard_tlab;
7149 if (CMSIncrementalMode || !Universe::heap()->supports_inline_contig_alloc()) {
7150 // No allocation in the shared eden.
7151 jmp(slow_case);
7152 }
7154 NOT_LP64(get_thread(thread_reg));
7156 movptr(top, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())));
7157 movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_end_offset())));
7159 // calculate amount of free space
7160 subptr(t1, top);
7161 shrptr(t1, LogHeapWordSize);
7163 // Retain tlab and allocate object in shared space if
7164 // the amount free in the tlab is too large to discard.
7165 cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_refill_waste_limit_offset())));
7166 jcc(Assembler::lessEqual, discard_tlab);
7168 // Retain
7169 // %%% yuck as movptr...
7170 movptr(t2, (int32_t) ThreadLocalAllocBuffer::refill_waste_limit_increment());
7171 addptr(Address(thread_reg, in_bytes(JavaThread::tlab_refill_waste_limit_offset())), t2);
7172 if (TLABStats) {
7173 // increment number of slow_allocations
7174 addl(Address(thread_reg, in_bytes(JavaThread::tlab_slow_allocations_offset())), 1);
7175 }
7176 jmp(try_eden);
7178 bind(discard_tlab);
7179 if (TLABStats) {
7180 // increment number of refills
7181 addl(Address(thread_reg, in_bytes(JavaThread::tlab_number_of_refills_offset())), 1);
7182 // accumulate wastage -- t1 is amount free in tlab
7183 addl(Address(thread_reg, in_bytes(JavaThread::tlab_fast_refill_waste_offset())), t1);
7184 }
7186 // if tlab is currently allocated (top or end != null) then
7187 // fill [top, end + alignment_reserve) with array object
7188 testptr (top, top);
7189 jcc(Assembler::zero, do_refill);
7191 // set up the mark word
7192 movptr(Address(top, oopDesc::mark_offset_in_bytes()), (intptr_t)markOopDesc::prototype()->copy_set_hash(0x2));
7193 // set the length to the remaining space
7194 subptr(t1, typeArrayOopDesc::header_size(T_INT));
7195 addptr(t1, (int32_t)ThreadLocalAllocBuffer::alignment_reserve());
7196 shlptr(t1, log2_intptr(HeapWordSize/sizeof(jint)));
7197 movl(Address(top, arrayOopDesc::length_offset_in_bytes()), t1);
7198 // set klass to intArrayKlass
7199 // dubious reloc why not an oop reloc?
7200 movptr(t1, ExternalAddress((address) Universe::intArrayKlassObj_addr()));
7201 // store klass last. concurrent gcs assumes klass length is valid if
7202 // klass field is not null.
7203 store_klass(top, t1);
7205 // refill the tlab with an eden allocation
7206 bind(do_refill);
7207 movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_size_offset())));
7208 shlptr(t1, LogHeapWordSize);
7209 // add object_size ??
7210 eden_allocate(top, t1, 0, t2, slow_case);
7212 // Check that t1 was preserved in eden_allocate.
7213 #ifdef ASSERT
7214 if (UseTLAB) {
7215 Label ok;
7216 Register tsize = rsi;
7217 assert_different_registers(tsize, thread_reg, t1);
7218 push(tsize);
7219 movptr(tsize, Address(thread_reg, in_bytes(JavaThread::tlab_size_offset())));
7220 shlptr(tsize, LogHeapWordSize);
7221 cmpptr(t1, tsize);
7222 jcc(Assembler::equal, ok);
7223 stop("assert(t1 != tlab size)");
7224 should_not_reach_here();
7226 bind(ok);
7227 pop(tsize);
7228 }
7229 #endif
7230 movptr(Address(thread_reg, in_bytes(JavaThread::tlab_start_offset())), top);
7231 movptr(Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())), top);
7232 addptr(top, t1);
7233 subptr(top, (int32_t)ThreadLocalAllocBuffer::alignment_reserve_in_bytes());
7234 movptr(Address(thread_reg, in_bytes(JavaThread::tlab_end_offset())), top);
7235 verify_tlab();
7236 jmp(retry);
7237 }
7239 static const double pi_4 = 0.7853981633974483;
7241 void MacroAssembler::trigfunc(char trig, int num_fpu_regs_in_use) {
7242 // A hand-coded argument reduction for values in fabs(pi/4, pi/2)
7243 // was attempted in this code; unfortunately it appears that the
7244 // switch to 80-bit precision and back causes this to be
7245 // unprofitable compared with simply performing a runtime call if
7246 // the argument is out of the (-pi/4, pi/4) range.
7248 Register tmp = noreg;
7249 if (!VM_Version::supports_cmov()) {
7250 // fcmp needs a temporary so preserve rbx,
7251 tmp = rbx;
7252 push(tmp);
7253 }
7255 Label slow_case, done;
7257 ExternalAddress pi4_adr = (address)&pi_4;
7258 if (reachable(pi4_adr)) {
7259 // x ?<= pi/4
7260 fld_d(pi4_adr);
7261 fld_s(1); // Stack: X PI/4 X
7262 fabs(); // Stack: |X| PI/4 X
7263 fcmp(tmp);
7264 jcc(Assembler::above, slow_case);
7266 // fastest case: -pi/4 <= x <= pi/4
7267 switch(trig) {
7268 case 's':
7269 fsin();
7270 break;
7271 case 'c':
7272 fcos();
7273 break;
7274 case 't':
7275 ftan();
7276 break;
7277 default:
7278 assert(false, "bad intrinsic");
7279 break;
7280 }
7281 jmp(done);
7282 }
7284 // slow case: runtime call
7285 bind(slow_case);
7286 // Preserve registers across runtime call
7287 pusha();
7288 int incoming_argument_and_return_value_offset = -1;
7289 if (num_fpu_regs_in_use > 1) {
7290 // Must preserve all other FPU regs (could alternatively convert
7291 // SharedRuntime::dsin and dcos into assembly routines known not to trash
7292 // FPU state, but can not trust C compiler)
7293 NEEDS_CLEANUP;
7294 // NOTE that in this case we also push the incoming argument to
7295 // the stack and restore it later; we also use this stack slot to
7296 // hold the return value from dsin or dcos.
7297 for (int i = 0; i < num_fpu_regs_in_use; i++) {
7298 subptr(rsp, sizeof(jdouble));
7299 fstp_d(Address(rsp, 0));
7300 }
7301 incoming_argument_and_return_value_offset = sizeof(jdouble)*(num_fpu_regs_in_use-1);
7302 fld_d(Address(rsp, incoming_argument_and_return_value_offset));
7303 }
7304 subptr(rsp, sizeof(jdouble));
7305 fstp_d(Address(rsp, 0));
7306 #ifdef _LP64
7307 movdbl(xmm0, Address(rsp, 0));
7308 #endif // _LP64
7310 // NOTE: we must not use call_VM_leaf here because that requires a
7311 // complete interpreter frame in debug mode -- same bug as 4387334
7312 // MacroAssembler::call_VM_leaf_base is perfectly safe and will
7313 // do proper 64bit abi
7315 NEEDS_CLEANUP;
7316 // Need to add stack banging before this runtime call if it needs to
7317 // be taken; however, there is no generic stack banging routine at
7318 // the MacroAssembler level
7319 switch(trig) {
7320 case 's':
7321 {
7322 MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, SharedRuntime::dsin), 0);
7323 }
7324 break;
7325 case 'c':
7326 {
7327 MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, SharedRuntime::dcos), 0);
7328 }
7329 break;
7330 case 't':
7331 {
7332 MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, SharedRuntime::dtan), 0);
7333 }
7334 break;
7335 default:
7336 assert(false, "bad intrinsic");
7337 break;
7338 }
7339 #ifdef _LP64
7340 movsd(Address(rsp, 0), xmm0);
7341 fld_d(Address(rsp, 0));
7342 #endif // _LP64
7343 addptr(rsp, sizeof(jdouble));
7344 if (num_fpu_regs_in_use > 1) {
7345 // Must save return value to stack and then restore entire FPU stack
7346 fstp_d(Address(rsp, incoming_argument_and_return_value_offset));
7347 for (int i = 0; i < num_fpu_regs_in_use; i++) {
7348 fld_d(Address(rsp, 0));
7349 addptr(rsp, sizeof(jdouble));
7350 }
7351 }
7352 popa();
7354 // Come here with result in F-TOS
7355 bind(done);
7357 if (tmp != noreg) {
7358 pop(tmp);
7359 }
7360 }
7363 // Look up the method for a megamorphic invokeinterface call.
7364 // The target method is determined by <intf_klass, itable_index>.
7365 // The receiver klass is in recv_klass.
7366 // On success, the result will be in method_result, and execution falls through.
7367 // On failure, execution transfers to the given label.
7368 void MacroAssembler::lookup_interface_method(Register recv_klass,
7369 Register intf_klass,
7370 RegisterOrConstant itable_index,
7371 Register method_result,
7372 Register scan_temp,
7373 Label& L_no_such_interface) {
7374 assert_different_registers(recv_klass, intf_klass, method_result, scan_temp);
7375 assert(itable_index.is_constant() || itable_index.as_register() == method_result,
7376 "caller must use same register for non-constant itable index as for method");
7378 // Compute start of first itableOffsetEntry (which is at the end of the vtable)
7379 int vtable_base = instanceKlass::vtable_start_offset() * wordSize;
7380 int itentry_off = itableMethodEntry::method_offset_in_bytes();
7381 int scan_step = itableOffsetEntry::size() * wordSize;
7382 int vte_size = vtableEntry::size() * wordSize;
7383 Address::ScaleFactor times_vte_scale = Address::times_ptr;
7384 assert(vte_size == wordSize, "else adjust times_vte_scale");
7386 movl(scan_temp, Address(recv_klass, instanceKlass::vtable_length_offset() * wordSize));
7388 // %%% Could store the aligned, prescaled offset in the klassoop.
7389 lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
7390 if (HeapWordsPerLong > 1) {
7391 // Round up to align_object_offset boundary
7392 // see code for instanceKlass::start_of_itable!
7393 round_to(scan_temp, BytesPerLong);
7394 }
7396 // Adjust recv_klass by scaled itable_index, so we can free itable_index.
7397 assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
7398 lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
7400 // for (scan = klass->itable(); scan->interface() != NULL; scan += scan_step) {
7401 // if (scan->interface() == intf) {
7402 // result = (klass + scan->offset() + itable_index);
7403 // }
7404 // }
7405 Label search, found_method;
7407 for (int peel = 1; peel >= 0; peel--) {
7408 movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset_in_bytes()));
7409 cmpptr(intf_klass, method_result);
7411 if (peel) {
7412 jccb(Assembler::equal, found_method);
7413 } else {
7414 jccb(Assembler::notEqual, search);
7415 // (invert the test to fall through to found_method...)
7416 }
7418 if (!peel) break;
7420 bind(search);
7422 // Check that the previous entry is non-null. A null entry means that
7423 // the receiver class doesn't implement the interface, and wasn't the
7424 // same as when the caller was compiled.
7425 testptr(method_result, method_result);
7426 jcc(Assembler::zero, L_no_such_interface);
7427 addptr(scan_temp, scan_step);
7428 }
7430 bind(found_method);
7432 // Got a hit.
7433 movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset_in_bytes()));
7434 movptr(method_result, Address(recv_klass, scan_temp, Address::times_1));
7435 }
7438 void MacroAssembler::check_klass_subtype(Register sub_klass,
7439 Register super_klass,
7440 Register temp_reg,
7441 Label& L_success) {
7442 Label L_failure;
7443 check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg, &L_success, &L_failure, NULL);
7444 check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, NULL);
7445 bind(L_failure);
7446 }
7449 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
7450 Register super_klass,
7451 Register temp_reg,
7452 Label* L_success,
7453 Label* L_failure,
7454 Label* L_slow_path,
7455 RegisterOrConstant super_check_offset) {
7456 assert_different_registers(sub_klass, super_klass, temp_reg);
7457 bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
7458 if (super_check_offset.is_register()) {
7459 assert_different_registers(sub_klass, super_klass,
7460 super_check_offset.as_register());
7461 } else if (must_load_sco) {
7462 assert(temp_reg != noreg, "supply either a temp or a register offset");
7463 }
7465 Label L_fallthrough;
7466 int label_nulls = 0;
7467 if (L_success == NULL) { L_success = &L_fallthrough; label_nulls++; }
7468 if (L_failure == NULL) { L_failure = &L_fallthrough; label_nulls++; }
7469 if (L_slow_path == NULL) { L_slow_path = &L_fallthrough; label_nulls++; }
7470 assert(label_nulls <= 1, "at most one NULL in the batch");
7472 int sc_offset = (klassOopDesc::header_size() * HeapWordSize +
7473 Klass::secondary_super_cache_offset_in_bytes());
7474 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
7475 Klass::super_check_offset_offset_in_bytes());
7476 Address super_check_offset_addr(super_klass, sco_offset);
7478 // Hacked jcc, which "knows" that L_fallthrough, at least, is in
7479 // range of a jccb. If this routine grows larger, reconsider at
7480 // least some of these.
7481 #define local_jcc(assembler_cond, label) \
7482 if (&(label) == &L_fallthrough) jccb(assembler_cond, label); \
7483 else jcc( assembler_cond, label) /*omit semi*/
7485 // Hacked jmp, which may only be used just before L_fallthrough.
7486 #define final_jmp(label) \
7487 if (&(label) == &L_fallthrough) { /*do nothing*/ } \
7488 else jmp(label) /*omit semi*/
7490 // If the pointers are equal, we are done (e.g., String[] elements).
7491 // This self-check enables sharing of secondary supertype arrays among
7492 // non-primary types such as array-of-interface. Otherwise, each such
7493 // type would need its own customized SSA.
7494 // We move this check to the front of the fast path because many
7495 // type checks are in fact trivially successful in this manner,
7496 // so we get a nicely predicted branch right at the start of the check.
7497 cmpptr(sub_klass, super_klass);
7498 local_jcc(Assembler::equal, *L_success);
7500 // Check the supertype display:
7501 if (must_load_sco) {
7502 // Positive movl does right thing on LP64.
7503 movl(temp_reg, super_check_offset_addr);
7504 super_check_offset = RegisterOrConstant(temp_reg);
7505 }
7506 Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
7507 cmpptr(super_klass, super_check_addr); // load displayed supertype
7509 // This check has worked decisively for primary supers.
7510 // Secondary supers are sought in the super_cache ('super_cache_addr').
7511 // (Secondary supers are interfaces and very deeply nested subtypes.)
7512 // This works in the same check above because of a tricky aliasing
7513 // between the super_cache and the primary super display elements.
7514 // (The 'super_check_addr' can address either, as the case requires.)
7515 // Note that the cache is updated below if it does not help us find
7516 // what we need immediately.
7517 // So if it was a primary super, we can just fail immediately.
7518 // Otherwise, it's the slow path for us (no success at this point).
7520 if (super_check_offset.is_register()) {
7521 local_jcc(Assembler::equal, *L_success);
7522 cmpl(super_check_offset.as_register(), sc_offset);
7523 if (L_failure == &L_fallthrough) {
7524 local_jcc(Assembler::equal, *L_slow_path);
7525 } else {
7526 local_jcc(Assembler::notEqual, *L_failure);
7527 final_jmp(*L_slow_path);
7528 }
7529 } else if (super_check_offset.as_constant() == sc_offset) {
7530 // Need a slow path; fast failure is impossible.
7531 if (L_slow_path == &L_fallthrough) {
7532 local_jcc(Assembler::equal, *L_success);
7533 } else {
7534 local_jcc(Assembler::notEqual, *L_slow_path);
7535 final_jmp(*L_success);
7536 }
7537 } else {
7538 // No slow path; it's a fast decision.
7539 if (L_failure == &L_fallthrough) {
7540 local_jcc(Assembler::equal, *L_success);
7541 } else {
7542 local_jcc(Assembler::notEqual, *L_failure);
7543 final_jmp(*L_success);
7544 }
7545 }
7547 bind(L_fallthrough);
7549 #undef local_jcc
7550 #undef final_jmp
7551 }
7554 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
7555 Register super_klass,
7556 Register temp_reg,
7557 Register temp2_reg,
7558 Label* L_success,
7559 Label* L_failure,
7560 bool set_cond_codes) {
7561 assert_different_registers(sub_klass, super_klass, temp_reg);
7562 if (temp2_reg != noreg)
7563 assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg);
7564 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
7566 Label L_fallthrough;
7567 int label_nulls = 0;
7568 if (L_success == NULL) { L_success = &L_fallthrough; label_nulls++; }
7569 if (L_failure == NULL) { L_failure = &L_fallthrough; label_nulls++; }
7570 assert(label_nulls <= 1, "at most one NULL in the batch");
7572 // a couple of useful fields in sub_klass:
7573 int ss_offset = (klassOopDesc::header_size() * HeapWordSize +
7574 Klass::secondary_supers_offset_in_bytes());
7575 int sc_offset = (klassOopDesc::header_size() * HeapWordSize +
7576 Klass::secondary_super_cache_offset_in_bytes());
7577 Address secondary_supers_addr(sub_klass, ss_offset);
7578 Address super_cache_addr( sub_klass, sc_offset);
7580 // Do a linear scan of the secondary super-klass chain.
7581 // This code is rarely used, so simplicity is a virtue here.
7582 // The repne_scan instruction uses fixed registers, which we must spill.
7583 // Don't worry too much about pre-existing connections with the input regs.
7585 assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super)
7586 assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter)
7588 // Get super_klass value into rax (even if it was in rdi or rcx).
7589 bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false;
7590 if (super_klass != rax || UseCompressedOops) {
7591 if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; }
7592 mov(rax, super_klass);
7593 }
7594 if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; }
7595 if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; }
7597 #ifndef PRODUCT
7598 int* pst_counter = &SharedRuntime::_partial_subtype_ctr;
7599 ExternalAddress pst_counter_addr((address) pst_counter);
7600 NOT_LP64( incrementl(pst_counter_addr) );
7601 LP64_ONLY( lea(rcx, pst_counter_addr) );
7602 LP64_ONLY( incrementl(Address(rcx, 0)) );
7603 #endif //PRODUCT
7605 // We will consult the secondary-super array.
7606 movptr(rdi, secondary_supers_addr);
7607 // Load the array length. (Positive movl does right thing on LP64.)
7608 movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes()));
7609 // Skip to start of data.
7610 addptr(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
7612 // Scan RCX words at [RDI] for an occurrence of RAX.
7613 // Set NZ/Z based on last compare.
7614 // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does
7615 // not change flags (only scas instruction which is repeated sets flags).
7616 // Set Z = 0 (not equal) before 'repne' to indicate that class was not found.
7617 #ifdef _LP64
7618 // This part is tricky, as values in supers array could be 32 or 64 bit wide
7619 // and we store values in objArrays always encoded, thus we need to encode
7620 // the value of rax before repne. Note that rax is dead after the repne.
7621 if (UseCompressedOops) {
7622 encode_heap_oop_not_null(rax); // Changes flags.
7623 // The superclass is never null; it would be a basic system error if a null
7624 // pointer were to sneak in here. Note that we have already loaded the
7625 // Klass::super_check_offset from the super_klass in the fast path,
7626 // so if there is a null in that register, we are already in the afterlife.
7627 testl(rax,rax); // Set Z = 0
7628 repne_scanl();
7629 } else
7630 #endif // _LP64
7631 {
7632 testptr(rax,rax); // Set Z = 0
7633 repne_scan();
7634 }
7635 // Unspill the temp. registers:
7636 if (pushed_rdi) pop(rdi);
7637 if (pushed_rcx) pop(rcx);
7638 if (pushed_rax) pop(rax);
7640 if (set_cond_codes) {
7641 // Special hack for the AD files: rdi is guaranteed non-zero.
7642 assert(!pushed_rdi, "rdi must be left non-NULL");
7643 // Also, the condition codes are properly set Z/NZ on succeed/failure.
7644 }
7646 if (L_failure == &L_fallthrough)
7647 jccb(Assembler::notEqual, *L_failure);
7648 else jcc(Assembler::notEqual, *L_failure);
7650 // Success. Cache the super we found and proceed in triumph.
7651 movptr(super_cache_addr, super_klass);
7653 if (L_success != &L_fallthrough) {
7654 jmp(*L_success);
7655 }
7657 #undef IS_A_TEMP
7659 bind(L_fallthrough);
7660 }
7663 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src) {
7664 ucomisd(dst, as_Address(src));
7665 }
7667 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src) {
7668 ucomiss(dst, as_Address(src));
7669 }
7671 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src) {
7672 if (reachable(src)) {
7673 xorpd(dst, as_Address(src));
7674 } else {
7675 lea(rscratch1, src);
7676 xorpd(dst, Address(rscratch1, 0));
7677 }
7678 }
7680 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src) {
7681 if (reachable(src)) {
7682 xorps(dst, as_Address(src));
7683 } else {
7684 lea(rscratch1, src);
7685 xorps(dst, Address(rscratch1, 0));
7686 }
7687 }
7689 void MacroAssembler::verify_oop(Register reg, const char* s) {
7690 if (!VerifyOops) return;
7692 // Pass register number to verify_oop_subroutine
7693 char* b = new char[strlen(s) + 50];
7694 sprintf(b, "verify_oop: %s: %s", reg->name(), s);
7695 #ifdef _LP64
7696 push(rscratch1); // save r10, trashed by movptr()
7697 #endif
7698 push(rax); // save rax,
7699 push(reg); // pass register argument
7700 ExternalAddress buffer((address) b);
7701 // avoid using pushptr, as it modifies scratch registers
7702 // and our contract is not to modify anything
7703 movptr(rax, buffer.addr());
7704 push(rax);
7705 // call indirectly to solve generation ordering problem
7706 movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
7707 call(rax);
7708 // Caller pops the arguments (oop, message) and restores rax, r10
7709 }
7712 RegisterOrConstant MacroAssembler::delayed_value_impl(intptr_t* delayed_value_addr,
7713 Register tmp,
7714 int offset) {
7715 intptr_t value = *delayed_value_addr;
7716 if (value != 0)
7717 return RegisterOrConstant(value + offset);
7719 // load indirectly to solve generation ordering problem
7720 movptr(tmp, ExternalAddress((address) delayed_value_addr));
7722 #ifdef ASSERT
7723 { Label L;
7724 testptr(tmp, tmp);
7725 if (WizardMode) {
7726 jcc(Assembler::notZero, L);
7727 char* buf = new char[40];
7728 sprintf(buf, "DelayedValue="INTPTR_FORMAT, delayed_value_addr[1]);
7729 stop(buf);
7730 } else {
7731 jccb(Assembler::notZero, L);
7732 hlt();
7733 }
7734 bind(L);
7735 }
7736 #endif
7738 if (offset != 0)
7739 addptr(tmp, offset);
7741 return RegisterOrConstant(tmp);
7742 }
7745 // registers on entry:
7746 // - rax ('check' register): required MethodType
7747 // - rcx: method handle
7748 // - rdx, rsi, or ?: killable temp
7749 void MacroAssembler::check_method_handle_type(Register mtype_reg, Register mh_reg,
7750 Register temp_reg,
7751 Label& wrong_method_type) {
7752 Address type_addr(mh_reg, delayed_value(java_dyn_MethodHandle::type_offset_in_bytes, temp_reg));
7753 // compare method type against that of the receiver
7754 if (UseCompressedOops) {
7755 load_heap_oop(temp_reg, type_addr);
7756 cmpptr(mtype_reg, temp_reg);
7757 } else {
7758 cmpptr(mtype_reg, type_addr);
7759 }
7760 jcc(Assembler::notEqual, wrong_method_type);
7761 }
7764 // A method handle has a "vmslots" field which gives the size of its
7765 // argument list in JVM stack slots. This field is either located directly
7766 // in every method handle, or else is indirectly accessed through the
7767 // method handle's MethodType. This macro hides the distinction.
7768 void MacroAssembler::load_method_handle_vmslots(Register vmslots_reg, Register mh_reg,
7769 Register temp_reg) {
7770 assert_different_registers(vmslots_reg, mh_reg, temp_reg);
7771 // load mh.type.form.vmslots
7772 if (java_dyn_MethodHandle::vmslots_offset_in_bytes() != 0) {
7773 // hoist vmslots into every mh to avoid dependent load chain
7774 movl(vmslots_reg, Address(mh_reg, delayed_value(java_dyn_MethodHandle::vmslots_offset_in_bytes, temp_reg)));
7775 } else {
7776 Register temp2_reg = vmslots_reg;
7777 load_heap_oop(temp2_reg, Address(mh_reg, delayed_value(java_dyn_MethodHandle::type_offset_in_bytes, temp_reg)));
7778 load_heap_oop(temp2_reg, Address(temp2_reg, delayed_value(java_dyn_MethodType::form_offset_in_bytes, temp_reg)));
7779 movl(vmslots_reg, Address(temp2_reg, delayed_value(java_dyn_MethodTypeForm::vmslots_offset_in_bytes, temp_reg)));
7780 }
7781 }
7784 // registers on entry:
7785 // - rcx: method handle
7786 // - rdx: killable temp (interpreted only)
7787 // - rax: killable temp (compiled only)
7788 void MacroAssembler::jump_to_method_handle_entry(Register mh_reg, Register temp_reg) {
7789 assert(mh_reg == rcx, "caller must put MH object in rcx");
7790 assert_different_registers(mh_reg, temp_reg);
7792 // pick out the interpreted side of the handler
7793 // NOTE: vmentry is not an oop!
7794 movptr(temp_reg, Address(mh_reg, delayed_value(java_dyn_MethodHandle::vmentry_offset_in_bytes, temp_reg)));
7796 // off we go...
7797 jmp(Address(temp_reg, MethodHandleEntry::from_interpreted_entry_offset_in_bytes()));
7799 // for the various stubs which take control at this point,
7800 // see MethodHandles::generate_method_handle_stub
7801 }
7804 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
7805 int extra_slot_offset) {
7806 // cf. TemplateTable::prepare_invoke(), if (load_receiver).
7807 int stackElementSize = Interpreter::stackElementSize;
7808 int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
7809 #ifdef ASSERT
7810 int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
7811 assert(offset1 - offset == stackElementSize, "correct arithmetic");
7812 #endif
7813 Register scale_reg = noreg;
7814 Address::ScaleFactor scale_factor = Address::no_scale;
7815 if (arg_slot.is_constant()) {
7816 offset += arg_slot.as_constant() * stackElementSize;
7817 } else {
7818 scale_reg = arg_slot.as_register();
7819 scale_factor = Address::times(stackElementSize);
7820 }
7821 offset += wordSize; // return PC is on stack
7822 return Address(rsp, scale_reg, scale_factor, offset);
7823 }
7826 void MacroAssembler::verify_oop_addr(Address addr, const char* s) {
7827 if (!VerifyOops) return;
7829 // Address adjust(addr.base(), addr.index(), addr.scale(), addr.disp() + BytesPerWord);
7830 // Pass register number to verify_oop_subroutine
7831 char* b = new char[strlen(s) + 50];
7832 sprintf(b, "verify_oop_addr: %s", s);
7834 #ifdef _LP64
7835 push(rscratch1); // save r10, trashed by movptr()
7836 #endif
7837 push(rax); // save rax,
7838 // addr may contain rsp so we will have to adjust it based on the push
7839 // we just did
7840 // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
7841 // stores rax into addr which is backwards of what was intended.
7842 if (addr.uses(rsp)) {
7843 lea(rax, addr);
7844 pushptr(Address(rax, BytesPerWord));
7845 } else {
7846 pushptr(addr);
7847 }
7849 ExternalAddress buffer((address) b);
7850 // pass msg argument
7851 // avoid using pushptr, as it modifies scratch registers
7852 // and our contract is not to modify anything
7853 movptr(rax, buffer.addr());
7854 push(rax);
7856 // call indirectly to solve generation ordering problem
7857 movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
7858 call(rax);
7859 // Caller pops the arguments (addr, message) and restores rax, r10.
7860 }
7862 void MacroAssembler::verify_tlab() {
7863 #ifdef ASSERT
7864 if (UseTLAB && VerifyOops) {
7865 Label next, ok;
7866 Register t1 = rsi;
7867 Register thread_reg = NOT_LP64(rbx) LP64_ONLY(r15_thread);
7869 push(t1);
7870 NOT_LP64(push(thread_reg));
7871 NOT_LP64(get_thread(thread_reg));
7873 movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())));
7874 cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_start_offset())));
7875 jcc(Assembler::aboveEqual, next);
7876 stop("assert(top >= start)");
7877 should_not_reach_here();
7879 bind(next);
7880 movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_end_offset())));
7881 cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())));
7882 jcc(Assembler::aboveEqual, ok);
7883 stop("assert(top <= end)");
7884 should_not_reach_here();
7886 bind(ok);
7887 NOT_LP64(pop(thread_reg));
7888 pop(t1);
7889 }
7890 #endif
7891 }
7893 class ControlWord {
7894 public:
7895 int32_t _value;
7897 int rounding_control() const { return (_value >> 10) & 3 ; }
7898 int precision_control() const { return (_value >> 8) & 3 ; }
7899 bool precision() const { return ((_value >> 5) & 1) != 0; }
7900 bool underflow() const { return ((_value >> 4) & 1) != 0; }
7901 bool overflow() const { return ((_value >> 3) & 1) != 0; }
7902 bool zero_divide() const { return ((_value >> 2) & 1) != 0; }
7903 bool denormalized() const { return ((_value >> 1) & 1) != 0; }
7904 bool invalid() const { return ((_value >> 0) & 1) != 0; }
7906 void print() const {
7907 // rounding control
7908 const char* rc;
7909 switch (rounding_control()) {
7910 case 0: rc = "round near"; break;
7911 case 1: rc = "round down"; break;
7912 case 2: rc = "round up "; break;
7913 case 3: rc = "chop "; break;
7914 };
7915 // precision control
7916 const char* pc;
7917 switch (precision_control()) {
7918 case 0: pc = "24 bits "; break;
7919 case 1: pc = "reserved"; break;
7920 case 2: pc = "53 bits "; break;
7921 case 3: pc = "64 bits "; break;
7922 };
7923 // flags
7924 char f[9];
7925 f[0] = ' ';
7926 f[1] = ' ';
7927 f[2] = (precision ()) ? 'P' : 'p';
7928 f[3] = (underflow ()) ? 'U' : 'u';
7929 f[4] = (overflow ()) ? 'O' : 'o';
7930 f[5] = (zero_divide ()) ? 'Z' : 'z';
7931 f[6] = (denormalized()) ? 'D' : 'd';
7932 f[7] = (invalid ()) ? 'I' : 'i';
7933 f[8] = '\x0';
7934 // output
7935 printf("%04x masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc);
7936 }
7938 };
7940 class StatusWord {
7941 public:
7942 int32_t _value;
7944 bool busy() const { return ((_value >> 15) & 1) != 0; }
7945 bool C3() const { return ((_value >> 14) & 1) != 0; }
7946 bool C2() const { return ((_value >> 10) & 1) != 0; }
7947 bool C1() const { return ((_value >> 9) & 1) != 0; }
7948 bool C0() const { return ((_value >> 8) & 1) != 0; }
7949 int top() const { return (_value >> 11) & 7 ; }
7950 bool error_status() const { return ((_value >> 7) & 1) != 0; }
7951 bool stack_fault() const { return ((_value >> 6) & 1) != 0; }
7952 bool precision() const { return ((_value >> 5) & 1) != 0; }
7953 bool underflow() const { return ((_value >> 4) & 1) != 0; }
7954 bool overflow() const { return ((_value >> 3) & 1) != 0; }
7955 bool zero_divide() const { return ((_value >> 2) & 1) != 0; }
7956 bool denormalized() const { return ((_value >> 1) & 1) != 0; }
7957 bool invalid() const { return ((_value >> 0) & 1) != 0; }
7959 void print() const {
7960 // condition codes
7961 char c[5];
7962 c[0] = (C3()) ? '3' : '-';
7963 c[1] = (C2()) ? '2' : '-';
7964 c[2] = (C1()) ? '1' : '-';
7965 c[3] = (C0()) ? '0' : '-';
7966 c[4] = '\x0';
7967 // flags
7968 char f[9];
7969 f[0] = (error_status()) ? 'E' : '-';
7970 f[1] = (stack_fault ()) ? 'S' : '-';
7971 f[2] = (precision ()) ? 'P' : '-';
7972 f[3] = (underflow ()) ? 'U' : '-';
7973 f[4] = (overflow ()) ? 'O' : '-';
7974 f[5] = (zero_divide ()) ? 'Z' : '-';
7975 f[6] = (denormalized()) ? 'D' : '-';
7976 f[7] = (invalid ()) ? 'I' : '-';
7977 f[8] = '\x0';
7978 // output
7979 printf("%04x flags = %s, cc = %s, top = %d", _value & 0xFFFF, f, c, top());
7980 }
7982 };
7984 class TagWord {
7985 public:
7986 int32_t _value;
7988 int tag_at(int i) const { return (_value >> (i*2)) & 3; }
7990 void print() const {
7991 printf("%04x", _value & 0xFFFF);
7992 }
7994 };
7996 class FPU_Register {
7997 public:
7998 int32_t _m0;
7999 int32_t _m1;
8000 int16_t _ex;
8002 bool is_indefinite() const {
8003 return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0;
8004 }
8006 void print() const {
8007 char sign = (_ex < 0) ? '-' : '+';
8008 const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : " ";
8009 printf("%c%04hx.%08x%08x %s", sign, _ex, _m1, _m0, kind);
8010 };
8012 };
8014 class FPU_State {
8015 public:
8016 enum {
8017 register_size = 10,
8018 number_of_registers = 8,
8019 register_mask = 7
8020 };
8022 ControlWord _control_word;
8023 StatusWord _status_word;
8024 TagWord _tag_word;
8025 int32_t _error_offset;
8026 int32_t _error_selector;
8027 int32_t _data_offset;
8028 int32_t _data_selector;
8029 int8_t _register[register_size * number_of_registers];
8031 int tag_for_st(int i) const { return _tag_word.tag_at((_status_word.top() + i) & register_mask); }
8032 FPU_Register* st(int i) const { return (FPU_Register*)&_register[register_size * i]; }
8034 const char* tag_as_string(int tag) const {
8035 switch (tag) {
8036 case 0: return "valid";
8037 case 1: return "zero";
8038 case 2: return "special";
8039 case 3: return "empty";
8040 }
8041 ShouldNotReachHere();
8042 return NULL;
8043 }
8045 void print() const {
8046 // print computation registers
8047 { int t = _status_word.top();
8048 for (int i = 0; i < number_of_registers; i++) {
8049 int j = (i - t) & register_mask;
8050 printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j);
8051 st(j)->print();
8052 printf(" %s\n", tag_as_string(_tag_word.tag_at(i)));
8053 }
8054 }
8055 printf("\n");
8056 // print control registers
8057 printf("ctrl = "); _control_word.print(); printf("\n");
8058 printf("stat = "); _status_word .print(); printf("\n");
8059 printf("tags = "); _tag_word .print(); printf("\n");
8060 }
8062 };
8064 class Flag_Register {
8065 public:
8066 int32_t _value;
8068 bool overflow() const { return ((_value >> 11) & 1) != 0; }
8069 bool direction() const { return ((_value >> 10) & 1) != 0; }
8070 bool sign() const { return ((_value >> 7) & 1) != 0; }
8071 bool zero() const { return ((_value >> 6) & 1) != 0; }
8072 bool auxiliary_carry() const { return ((_value >> 4) & 1) != 0; }
8073 bool parity() const { return ((_value >> 2) & 1) != 0; }
8074 bool carry() const { return ((_value >> 0) & 1) != 0; }
8076 void print() const {
8077 // flags
8078 char f[8];
8079 f[0] = (overflow ()) ? 'O' : '-';
8080 f[1] = (direction ()) ? 'D' : '-';
8081 f[2] = (sign ()) ? 'S' : '-';
8082 f[3] = (zero ()) ? 'Z' : '-';
8083 f[4] = (auxiliary_carry()) ? 'A' : '-';
8084 f[5] = (parity ()) ? 'P' : '-';
8085 f[6] = (carry ()) ? 'C' : '-';
8086 f[7] = '\x0';
8087 // output
8088 printf("%08x flags = %s", _value, f);
8089 }
8091 };
8093 class IU_Register {
8094 public:
8095 int32_t _value;
8097 void print() const {
8098 printf("%08x %11d", _value, _value);
8099 }
8101 };
8103 class IU_State {
8104 public:
8105 Flag_Register _eflags;
8106 IU_Register _rdi;
8107 IU_Register _rsi;
8108 IU_Register _rbp;
8109 IU_Register _rsp;
8110 IU_Register _rbx;
8111 IU_Register _rdx;
8112 IU_Register _rcx;
8113 IU_Register _rax;
8115 void print() const {
8116 // computation registers
8117 printf("rax, = "); _rax.print(); printf("\n");
8118 printf("rbx, = "); _rbx.print(); printf("\n");
8119 printf("rcx = "); _rcx.print(); printf("\n");
8120 printf("rdx = "); _rdx.print(); printf("\n");
8121 printf("rdi = "); _rdi.print(); printf("\n");
8122 printf("rsi = "); _rsi.print(); printf("\n");
8123 printf("rbp, = "); _rbp.print(); printf("\n");
8124 printf("rsp = "); _rsp.print(); printf("\n");
8125 printf("\n");
8126 // control registers
8127 printf("flgs = "); _eflags.print(); printf("\n");
8128 }
8129 };
8132 class CPU_State {
8133 public:
8134 FPU_State _fpu_state;
8135 IU_State _iu_state;
8137 void print() const {
8138 printf("--------------------------------------------------\n");
8139 _iu_state .print();
8140 printf("\n");
8141 _fpu_state.print();
8142 printf("--------------------------------------------------\n");
8143 }
8145 };
8148 static void _print_CPU_state(CPU_State* state) {
8149 state->print();
8150 };
8153 void MacroAssembler::print_CPU_state() {
8154 push_CPU_state();
8155 push(rsp); // pass CPU state
8156 call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state)));
8157 addptr(rsp, wordSize); // discard argument
8158 pop_CPU_state();
8159 }
8162 static bool _verify_FPU(int stack_depth, char* s, CPU_State* state) {
8163 static int counter = 0;
8164 FPU_State* fs = &state->_fpu_state;
8165 counter++;
8166 // For leaf calls, only verify that the top few elements remain empty.
8167 // We only need 1 empty at the top for C2 code.
8168 if( stack_depth < 0 ) {
8169 if( fs->tag_for_st(7) != 3 ) {
8170 printf("FPR7 not empty\n");
8171 state->print();
8172 assert(false, "error");
8173 return false;
8174 }
8175 return true; // All other stack states do not matter
8176 }
8178 assert((fs->_control_word._value & 0xffff) == StubRoutines::_fpu_cntrl_wrd_std,
8179 "bad FPU control word");
8181 // compute stack depth
8182 int i = 0;
8183 while (i < FPU_State::number_of_registers && fs->tag_for_st(i) < 3) i++;
8184 int d = i;
8185 while (i < FPU_State::number_of_registers && fs->tag_for_st(i) == 3) i++;
8186 // verify findings
8187 if (i != FPU_State::number_of_registers) {
8188 // stack not contiguous
8189 printf("%s: stack not contiguous at ST%d\n", s, i);
8190 state->print();
8191 assert(false, "error");
8192 return false;
8193 }
8194 // check if computed stack depth corresponds to expected stack depth
8195 if (stack_depth < 0) {
8196 // expected stack depth is -stack_depth or less
8197 if (d > -stack_depth) {
8198 // too many elements on the stack
8199 printf("%s: <= %d stack elements expected but found %d\n", s, -stack_depth, d);
8200 state->print();
8201 assert(false, "error");
8202 return false;
8203 }
8204 } else {
8205 // expected stack depth is stack_depth
8206 if (d != stack_depth) {
8207 // wrong stack depth
8208 printf("%s: %d stack elements expected but found %d\n", s, stack_depth, d);
8209 state->print();
8210 assert(false, "error");
8211 return false;
8212 }
8213 }
8214 // everything is cool
8215 return true;
8216 }
8219 void MacroAssembler::verify_FPU(int stack_depth, const char* s) {
8220 if (!VerifyFPU) return;
8221 push_CPU_state();
8222 push(rsp); // pass CPU state
8223 ExternalAddress msg((address) s);
8224 // pass message string s
8225 pushptr(msg.addr());
8226 push(stack_depth); // pass stack depth
8227 call(RuntimeAddress(CAST_FROM_FN_PTR(address, _verify_FPU)));
8228 addptr(rsp, 3 * wordSize); // discard arguments
8229 // check for error
8230 { Label L;
8231 testl(rax, rax);
8232 jcc(Assembler::notZero, L);
8233 int3(); // break if error condition
8234 bind(L);
8235 }
8236 pop_CPU_state();
8237 }
8239 void MacroAssembler::load_klass(Register dst, Register src) {
8240 #ifdef _LP64
8241 if (UseCompressedOops) {
8242 movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
8243 decode_heap_oop_not_null(dst);
8244 } else
8245 #endif
8246 movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
8247 }
8249 void MacroAssembler::load_prototype_header(Register dst, Register src) {
8250 #ifdef _LP64
8251 if (UseCompressedOops) {
8252 assert (Universe::heap() != NULL, "java heap should be initialized");
8253 movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
8254 if (Universe::narrow_oop_shift() != 0) {
8255 assert(LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
8256 if (LogMinObjAlignmentInBytes == Address::times_8) {
8257 movq(dst, Address(r12_heapbase, dst, Address::times_8, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
8258 } else {
8259 // OK to use shift since we don't need to preserve flags.
8260 shlq(dst, LogMinObjAlignmentInBytes);
8261 movq(dst, Address(r12_heapbase, dst, Address::times_1, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
8262 }
8263 } else {
8264 movq(dst, Address(dst, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
8265 }
8266 } else
8267 #endif
8268 {
8269 movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
8270 movptr(dst, Address(dst, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
8271 }
8272 }
8274 void MacroAssembler::store_klass(Register dst, Register src) {
8275 #ifdef _LP64
8276 if (UseCompressedOops) {
8277 encode_heap_oop_not_null(src);
8278 movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
8279 } else
8280 #endif
8281 movptr(Address(dst, oopDesc::klass_offset_in_bytes()), src);
8282 }
8284 void MacroAssembler::load_heap_oop(Register dst, Address src) {
8285 #ifdef _LP64
8286 if (UseCompressedOops) {
8287 movl(dst, src);
8288 decode_heap_oop(dst);
8289 } else
8290 #endif
8291 movptr(dst, src);
8292 }
8294 void MacroAssembler::store_heap_oop(Address dst, Register src) {
8295 #ifdef _LP64
8296 if (UseCompressedOops) {
8297 assert(!dst.uses(src), "not enough registers");
8298 encode_heap_oop(src);
8299 movl(dst, src);
8300 } else
8301 #endif
8302 movptr(dst, src);
8303 }
8305 // Used for storing NULLs.
8306 void MacroAssembler::store_heap_oop_null(Address dst) {
8307 #ifdef _LP64
8308 if (UseCompressedOops) {
8309 movl(dst, (int32_t)NULL_WORD);
8310 } else {
8311 movslq(dst, (int32_t)NULL_WORD);
8312 }
8313 #else
8314 movl(dst, (int32_t)NULL_WORD);
8315 #endif
8316 }
8318 #ifdef _LP64
8319 void MacroAssembler::store_klass_gap(Register dst, Register src) {
8320 if (UseCompressedOops) {
8321 // Store to klass gap in destination
8322 movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src);
8323 }
8324 }
8326 #ifdef ASSERT
8327 void MacroAssembler::verify_heapbase(const char* msg) {
8328 assert (UseCompressedOops, "should be compressed");
8329 assert (Universe::heap() != NULL, "java heap should be initialized");
8330 if (CheckCompressedOops) {
8331 Label ok;
8332 push(rscratch1); // cmpptr trashes rscratch1
8333 cmpptr(r12_heapbase, ExternalAddress((address)Universe::narrow_oop_base_addr()));
8334 jcc(Assembler::equal, ok);
8335 stop(msg);
8336 bind(ok);
8337 pop(rscratch1);
8338 }
8339 }
8340 #endif
8342 // Algorithm must match oop.inline.hpp encode_heap_oop.
8343 void MacroAssembler::encode_heap_oop(Register r) {
8344 #ifdef ASSERT
8345 verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
8346 #endif
8347 verify_oop(r, "broken oop in encode_heap_oop");
8348 if (Universe::narrow_oop_base() == NULL) {
8349 if (Universe::narrow_oop_shift() != 0) {
8350 assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
8351 shrq(r, LogMinObjAlignmentInBytes);
8352 }
8353 return;
8354 }
8355 testq(r, r);
8356 cmovq(Assembler::equal, r, r12_heapbase);
8357 subq(r, r12_heapbase);
8358 shrq(r, LogMinObjAlignmentInBytes);
8359 }
8361 void MacroAssembler::encode_heap_oop_not_null(Register r) {
8362 #ifdef ASSERT
8363 verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
8364 if (CheckCompressedOops) {
8365 Label ok;
8366 testq(r, r);
8367 jcc(Assembler::notEqual, ok);
8368 stop("null oop passed to encode_heap_oop_not_null");
8369 bind(ok);
8370 }
8371 #endif
8372 verify_oop(r, "broken oop in encode_heap_oop_not_null");
8373 if (Universe::narrow_oop_base() != NULL) {
8374 subq(r, r12_heapbase);
8375 }
8376 if (Universe::narrow_oop_shift() != 0) {
8377 assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
8378 shrq(r, LogMinObjAlignmentInBytes);
8379 }
8380 }
8382 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
8383 #ifdef ASSERT
8384 verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
8385 if (CheckCompressedOops) {
8386 Label ok;
8387 testq(src, src);
8388 jcc(Assembler::notEqual, ok);
8389 stop("null oop passed to encode_heap_oop_not_null2");
8390 bind(ok);
8391 }
8392 #endif
8393 verify_oop(src, "broken oop in encode_heap_oop_not_null2");
8394 if (dst != src) {
8395 movq(dst, src);
8396 }
8397 if (Universe::narrow_oop_base() != NULL) {
8398 subq(dst, r12_heapbase);
8399 }
8400 if (Universe::narrow_oop_shift() != 0) {
8401 assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
8402 shrq(dst, LogMinObjAlignmentInBytes);
8403 }
8404 }
8406 void MacroAssembler::decode_heap_oop(Register r) {
8407 #ifdef ASSERT
8408 verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
8409 #endif
8410 if (Universe::narrow_oop_base() == NULL) {
8411 if (Universe::narrow_oop_shift() != 0) {
8412 assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
8413 shlq(r, LogMinObjAlignmentInBytes);
8414 }
8415 } else {
8416 Label done;
8417 shlq(r, LogMinObjAlignmentInBytes);
8418 jccb(Assembler::equal, done);
8419 addq(r, r12_heapbase);
8420 bind(done);
8421 }
8422 verify_oop(r, "broken oop in decode_heap_oop");
8423 }
8425 void MacroAssembler::decode_heap_oop_not_null(Register r) {
8426 // Note: it will change flags
8427 assert (UseCompressedOops, "should only be used for compressed headers");
8428 assert (Universe::heap() != NULL, "java heap should be initialized");
8429 // Cannot assert, unverified entry point counts instructions (see .ad file)
8430 // vtableStubs also counts instructions in pd_code_size_limit.
8431 // Also do not verify_oop as this is called by verify_oop.
8432 if (Universe::narrow_oop_shift() != 0) {
8433 assert(LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
8434 shlq(r, LogMinObjAlignmentInBytes);
8435 if (Universe::narrow_oop_base() != NULL) {
8436 addq(r, r12_heapbase);
8437 }
8438 } else {
8439 assert (Universe::narrow_oop_base() == NULL, "sanity");
8440 }
8441 }
8443 void MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
8444 // Note: it will change flags
8445 assert (UseCompressedOops, "should only be used for compressed headers");
8446 assert (Universe::heap() != NULL, "java heap should be initialized");
8447 // Cannot assert, unverified entry point counts instructions (see .ad file)
8448 // vtableStubs also counts instructions in pd_code_size_limit.
8449 // Also do not verify_oop as this is called by verify_oop.
8450 if (Universe::narrow_oop_shift() != 0) {
8451 assert(LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong");
8452 if (LogMinObjAlignmentInBytes == Address::times_8) {
8453 leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
8454 } else {
8455 if (dst != src) {
8456 movq(dst, src);
8457 }
8458 shlq(dst, LogMinObjAlignmentInBytes);
8459 if (Universe::narrow_oop_base() != NULL) {
8460 addq(dst, r12_heapbase);
8461 }
8462 }
8463 } else {
8464 assert (Universe::narrow_oop_base() == NULL, "sanity");
8465 if (dst != src) {
8466 movq(dst, src);
8467 }
8468 }
8469 }
8471 void MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
8472 assert (UseCompressedOops, "should only be used for compressed headers");
8473 assert (Universe::heap() != NULL, "java heap should be initialized");
8474 assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
8475 int oop_index = oop_recorder()->find_index(obj);
8476 RelocationHolder rspec = oop_Relocation::spec(oop_index);
8477 mov_narrow_oop(dst, oop_index, rspec);
8478 }
8480 void MacroAssembler::set_narrow_oop(Address dst, jobject obj) {
8481 assert (UseCompressedOops, "should only be used for compressed headers");
8482 assert (Universe::heap() != NULL, "java heap should be initialized");
8483 assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
8484 int oop_index = oop_recorder()->find_index(obj);
8485 RelocationHolder rspec = oop_Relocation::spec(oop_index);
8486 mov_narrow_oop(dst, oop_index, rspec);
8487 }
8489 void MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
8490 assert (UseCompressedOops, "should only be used for compressed headers");
8491 assert (Universe::heap() != NULL, "java heap should be initialized");
8492 assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
8493 int oop_index = oop_recorder()->find_index(obj);
8494 RelocationHolder rspec = oop_Relocation::spec(oop_index);
8495 Assembler::cmp_narrow_oop(dst, oop_index, rspec);
8496 }
8498 void MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) {
8499 assert (UseCompressedOops, "should only be used for compressed headers");
8500 assert (Universe::heap() != NULL, "java heap should be initialized");
8501 assert (oop_recorder() != NULL, "this assembler needs an OopRecorder");
8502 int oop_index = oop_recorder()->find_index(obj);
8503 RelocationHolder rspec = oop_Relocation::spec(oop_index);
8504 Assembler::cmp_narrow_oop(dst, oop_index, rspec);
8505 }
8507 void MacroAssembler::reinit_heapbase() {
8508 if (UseCompressedOops) {
8509 movptr(r12_heapbase, ExternalAddress((address)Universe::narrow_oop_base_addr()));
8510 }
8511 }
8512 #endif // _LP64
8514 // IndexOf substring.
8515 void MacroAssembler::string_indexof(Register str1, Register str2,
8516 Register cnt1, Register cnt2, Register result,
8517 XMMRegister vec, Register tmp) {
8518 assert(UseSSE42Intrinsics, "SSE4.2 is required");
8520 Label RELOAD_SUBSTR, PREP_FOR_SCAN, SCAN_TO_SUBSTR,
8521 SCAN_SUBSTR, RET_NOT_FOUND, CLEANUP;
8523 push(str1); // string addr
8524 push(str2); // substr addr
8525 push(cnt2); // substr count
8526 jmpb(PREP_FOR_SCAN);
8528 // Substr count saved at sp
8529 // Substr saved at sp+1*wordSize
8530 // String saved at sp+2*wordSize
8532 // Reload substr for rescan
8533 bind(RELOAD_SUBSTR);
8534 movl(cnt2, Address(rsp, 0));
8535 movptr(str2, Address(rsp, wordSize));
8536 // We came here after the beginninig of the substring was
8537 // matched but the rest of it was not so we need to search
8538 // again. Start from the next element after the previous match.
8539 subptr(str1, result); // Restore counter
8540 shrl(str1, 1);
8541 addl(cnt1, str1);
8542 decrementl(cnt1);
8543 lea(str1, Address(result, 2)); // Reload string
8545 // Load substr
8546 bind(PREP_FOR_SCAN);
8547 movdqu(vec, Address(str2, 0));
8548 addl(cnt1, 8); // prime the loop
8549 subptr(str1, 16);
8551 // Scan string for substr in 16-byte vectors
8552 bind(SCAN_TO_SUBSTR);
8553 subl(cnt1, 8);
8554 addptr(str1, 16);
8556 // pcmpestri
8557 // inputs:
8558 // xmm - substring
8559 // rax - substring length (elements count)
8560 // mem - scaned string
8561 // rdx - string length (elements count)
8562 // 0xd - mode: 1100 (substring search) + 01 (unsigned shorts)
8563 // outputs:
8564 // rcx - matched index in string
8565 assert(cnt1 == rdx && cnt2 == rax && tmp == rcx, "pcmpestri");
8567 pcmpestri(vec, Address(str1, 0), 0x0d);
8568 jcc(Assembler::above, SCAN_TO_SUBSTR); // CF == 0 && ZF == 0
8569 jccb(Assembler::aboveEqual, RET_NOT_FOUND); // CF == 0
8571 // Fallthrough: found a potential substr
8573 // Make sure string is still long enough
8574 subl(cnt1, tmp);
8575 cmpl(cnt1, cnt2);
8576 jccb(Assembler::negative, RET_NOT_FOUND);
8577 // Compute start addr of substr
8578 lea(str1, Address(str1, tmp, Address::times_2));
8579 movptr(result, str1); // save
8581 // Compare potential substr
8582 addl(cnt1, 8); // prime the loop
8583 addl(cnt2, 8);
8584 subptr(str1, 16);
8585 subptr(str2, 16);
8587 // Scan 16-byte vectors of string and substr
8588 bind(SCAN_SUBSTR);
8589 subl(cnt1, 8);
8590 subl(cnt2, 8);
8591 addptr(str1, 16);
8592 addptr(str2, 16);
8593 movdqu(vec, Address(str2, 0));
8594 pcmpestri(vec, Address(str1, 0), 0x0d);
8595 jcc(Assembler::noOverflow, RELOAD_SUBSTR); // OF == 0
8596 jcc(Assembler::positive, SCAN_SUBSTR); // SF == 0
8598 // Compute substr offset
8599 subptr(result, Address(rsp, 2*wordSize));
8600 shrl(result, 1); // index
8601 jmpb(CLEANUP);
8603 bind(RET_NOT_FOUND);
8604 movl(result, -1);
8606 bind(CLEANUP);
8607 addptr(rsp, 3*wordSize);
8608 }
8610 // Compare strings.
8611 void MacroAssembler::string_compare(Register str1, Register str2,
8612 Register cnt1, Register cnt2, Register result,
8613 XMMRegister vec1, XMMRegister vec2) {
8614 Label LENGTH_DIFF_LABEL, POP_LABEL, DONE_LABEL, WHILE_HEAD_LABEL;
8616 // Compute the minimum of the string lengths and the
8617 // difference of the string lengths (stack).
8618 // Do the conditional move stuff
8619 movl(result, cnt1);
8620 subl(cnt1, cnt2);
8621 push(cnt1);
8622 if (VM_Version::supports_cmov()) {
8623 cmovl(Assembler::lessEqual, cnt2, result);
8624 } else {
8625 Label GT_LABEL;
8626 jccb(Assembler::greater, GT_LABEL);
8627 movl(cnt2, result);
8628 bind(GT_LABEL);
8629 }
8631 // Is the minimum length zero?
8632 testl(cnt2, cnt2);
8633 jcc(Assembler::zero, LENGTH_DIFF_LABEL);
8635 // Load first characters
8636 load_unsigned_short(result, Address(str1, 0));
8637 load_unsigned_short(cnt1, Address(str2, 0));
8639 // Compare first characters
8640 subl(result, cnt1);
8641 jcc(Assembler::notZero, POP_LABEL);
8642 decrementl(cnt2);
8643 jcc(Assembler::zero, LENGTH_DIFF_LABEL);
8645 {
8646 // Check after comparing first character to see if strings are equivalent
8647 Label LSkip2;
8648 // Check if the strings start at same location
8649 cmpptr(str1, str2);
8650 jccb(Assembler::notEqual, LSkip2);
8652 // Check if the length difference is zero (from stack)
8653 cmpl(Address(rsp, 0), 0x0);
8654 jcc(Assembler::equal, LENGTH_DIFF_LABEL);
8656 // Strings might not be equivalent
8657 bind(LSkip2);
8658 }
8660 // Advance to next character
8661 addptr(str1, 2);
8662 addptr(str2, 2);
8664 if (UseSSE42Intrinsics) {
8665 // With SSE4.2, use double quad vector compare
8666 Label COMPARE_VECTORS, VECTOR_NOT_EQUAL, COMPARE_TAIL;
8667 // Setup to compare 16-byte vectors
8668 movl(cnt1, cnt2);
8669 andl(cnt2, 0xfffffff8); // cnt2 holds the vector count
8670 andl(cnt1, 0x00000007); // cnt1 holds the tail count
8671 testl(cnt2, cnt2);
8672 jccb(Assembler::zero, COMPARE_TAIL);
8674 lea(str2, Address(str2, cnt2, Address::times_2));
8675 lea(str1, Address(str1, cnt2, Address::times_2));
8676 negptr(cnt2);
8678 bind(COMPARE_VECTORS);
8679 movdqu(vec1, Address(str1, cnt2, Address::times_2));
8680 movdqu(vec2, Address(str2, cnt2, Address::times_2));
8681 pxor(vec1, vec2);
8682 ptest(vec1, vec1);
8683 jccb(Assembler::notZero, VECTOR_NOT_EQUAL);
8684 addptr(cnt2, 8);
8685 jcc(Assembler::notZero, COMPARE_VECTORS);
8686 jmpb(COMPARE_TAIL);
8688 // Mismatched characters in the vectors
8689 bind(VECTOR_NOT_EQUAL);
8690 lea(str1, Address(str1, cnt2, Address::times_2));
8691 lea(str2, Address(str2, cnt2, Address::times_2));
8692 movl(cnt1, 8);
8694 // Compare tail (< 8 chars), or rescan last vectors to
8695 // find 1st mismatched characters
8696 bind(COMPARE_TAIL);
8697 testl(cnt1, cnt1);
8698 jccb(Assembler::zero, LENGTH_DIFF_LABEL);
8699 movl(cnt2, cnt1);
8700 // Fallthru to tail compare
8701 }
8703 // Shift str2 and str1 to the end of the arrays, negate min
8704 lea(str1, Address(str1, cnt2, Address::times_2, 0));
8705 lea(str2, Address(str2, cnt2, Address::times_2, 0));
8706 negptr(cnt2);
8708 // Compare the rest of the characters
8709 bind(WHILE_HEAD_LABEL);
8710 load_unsigned_short(result, Address(str1, cnt2, Address::times_2, 0));
8711 load_unsigned_short(cnt1, Address(str2, cnt2, Address::times_2, 0));
8712 subl(result, cnt1);
8713 jccb(Assembler::notZero, POP_LABEL);
8714 increment(cnt2);
8715 jcc(Assembler::notZero, WHILE_HEAD_LABEL);
8717 // Strings are equal up to min length. Return the length difference.
8718 bind(LENGTH_DIFF_LABEL);
8719 pop(result);
8720 jmpb(DONE_LABEL);
8722 // Discard the stored length difference
8723 bind(POP_LABEL);
8724 addptr(rsp, wordSize);
8726 // That's it
8727 bind(DONE_LABEL);
8728 }
8730 // Compare char[] arrays aligned to 4 bytes or substrings.
8731 void MacroAssembler::char_arrays_equals(bool is_array_equ, Register ary1, Register ary2,
8732 Register limit, Register result, Register chr,
8733 XMMRegister vec1, XMMRegister vec2) {
8734 Label TRUE_LABEL, FALSE_LABEL, DONE, COMPARE_VECTORS, COMPARE_CHAR;
8736 int length_offset = arrayOopDesc::length_offset_in_bytes();
8737 int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR);
8739 // Check the input args
8740 cmpptr(ary1, ary2);
8741 jcc(Assembler::equal, TRUE_LABEL);
8743 if (is_array_equ) {
8744 // Need additional checks for arrays_equals.
8745 testptr(ary1, ary1);
8746 jcc(Assembler::zero, FALSE_LABEL);
8747 testptr(ary2, ary2);
8748 jcc(Assembler::zero, FALSE_LABEL);
8750 // Check the lengths
8751 movl(limit, Address(ary1, length_offset));
8752 cmpl(limit, Address(ary2, length_offset));
8753 jcc(Assembler::notEqual, FALSE_LABEL);
8754 }
8756 // count == 0
8757 testl(limit, limit);
8758 jcc(Assembler::zero, TRUE_LABEL);
8760 if (is_array_equ) {
8761 // Load array address
8762 lea(ary1, Address(ary1, base_offset));
8763 lea(ary2, Address(ary2, base_offset));
8764 }
8766 shll(limit, 1); // byte count != 0
8767 movl(result, limit); // copy
8769 if (UseSSE42Intrinsics) {
8770 // With SSE4.2, use double quad vector compare
8771 Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
8772 // Compare 16-byte vectors
8773 andl(result, 0x0000000e); // tail count (in bytes)
8774 andl(limit, 0xfffffff0); // vector count (in bytes)
8775 jccb(Assembler::zero, COMPARE_TAIL);
8777 lea(ary1, Address(ary1, limit, Address::times_1));
8778 lea(ary2, Address(ary2, limit, Address::times_1));
8779 negptr(limit);
8781 bind(COMPARE_WIDE_VECTORS);
8782 movdqu(vec1, Address(ary1, limit, Address::times_1));
8783 movdqu(vec2, Address(ary2, limit, Address::times_1));
8784 pxor(vec1, vec2);
8785 ptest(vec1, vec1);
8786 jccb(Assembler::notZero, FALSE_LABEL);
8787 addptr(limit, 16);
8788 jcc(Assembler::notZero, COMPARE_WIDE_VECTORS);
8790 bind(COMPARE_TAIL); // limit is zero
8791 movl(limit, result);
8792 // Fallthru to tail compare
8793 }
8795 // Compare 4-byte vectors
8796 andl(limit, 0xfffffffc); // vector count (in bytes)
8797 jccb(Assembler::zero, COMPARE_CHAR);
8799 lea(ary1, Address(ary1, limit, Address::times_1));
8800 lea(ary2, Address(ary2, limit, Address::times_1));
8801 negptr(limit);
8803 bind(COMPARE_VECTORS);
8804 movl(chr, Address(ary1, limit, Address::times_1));
8805 cmpl(chr, Address(ary2, limit, Address::times_1));
8806 jccb(Assembler::notEqual, FALSE_LABEL);
8807 addptr(limit, 4);
8808 jcc(Assembler::notZero, COMPARE_VECTORS);
8810 // Compare trailing char (final 2 bytes), if any
8811 bind(COMPARE_CHAR);
8812 testl(result, 0x2); // tail char
8813 jccb(Assembler::zero, TRUE_LABEL);
8814 load_unsigned_short(chr, Address(ary1, 0));
8815 load_unsigned_short(limit, Address(ary2, 0));
8816 cmpl(chr, limit);
8817 jccb(Assembler::notEqual, FALSE_LABEL);
8819 bind(TRUE_LABEL);
8820 movl(result, 1); // return true
8821 jmpb(DONE);
8823 bind(FALSE_LABEL);
8824 xorl(result, result); // return false
8826 // That's it
8827 bind(DONE);
8828 }
8830 #ifdef PRODUCT
8831 #define BLOCK_COMMENT(str) /* nothing */
8832 #else
8833 #define BLOCK_COMMENT(str) block_comment(str)
8834 #endif
8836 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
8837 void MacroAssembler::generate_fill(BasicType t, bool aligned,
8838 Register to, Register value, Register count,
8839 Register rtmp, XMMRegister xtmp) {
8840 assert_different_registers(to, value, count, rtmp);
8841 Label L_exit, L_skip_align1, L_skip_align2, L_fill_byte;
8842 Label L_fill_2_bytes, L_fill_4_bytes;
8844 int shift = -1;
8845 switch (t) {
8846 case T_BYTE:
8847 shift = 2;
8848 break;
8849 case T_SHORT:
8850 shift = 1;
8851 break;
8852 case T_INT:
8853 shift = 0;
8854 break;
8855 default: ShouldNotReachHere();
8856 }
8858 if (t == T_BYTE) {
8859 andl(value, 0xff);
8860 movl(rtmp, value);
8861 shll(rtmp, 8);
8862 orl(value, rtmp);
8863 }
8864 if (t == T_SHORT) {
8865 andl(value, 0xffff);
8866 }
8867 if (t == T_BYTE || t == T_SHORT) {
8868 movl(rtmp, value);
8869 shll(rtmp, 16);
8870 orl(value, rtmp);
8871 }
8873 cmpl(count, 2<<shift); // Short arrays (< 8 bytes) fill by element
8874 jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp
8875 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
8876 // align source address at 4 bytes address boundary
8877 if (t == T_BYTE) {
8878 // One byte misalignment happens only for byte arrays
8879 testptr(to, 1);
8880 jccb(Assembler::zero, L_skip_align1);
8881 movb(Address(to, 0), value);
8882 increment(to);
8883 decrement(count);
8884 BIND(L_skip_align1);
8885 }
8886 // Two bytes misalignment happens only for byte and short (char) arrays
8887 testptr(to, 2);
8888 jccb(Assembler::zero, L_skip_align2);
8889 movw(Address(to, 0), value);
8890 addptr(to, 2);
8891 subl(count, 1<<(shift-1));
8892 BIND(L_skip_align2);
8893 }
8894 if (UseSSE < 2) {
8895 Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
8896 // Fill 32-byte chunks
8897 subl(count, 8 << shift);
8898 jcc(Assembler::less, L_check_fill_8_bytes);
8899 align(16);
8901 BIND(L_fill_32_bytes_loop);
8903 for (int i = 0; i < 32; i += 4) {
8904 movl(Address(to, i), value);
8905 }
8907 addptr(to, 32);
8908 subl(count, 8 << shift);
8909 jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
8910 BIND(L_check_fill_8_bytes);
8911 addl(count, 8 << shift);
8912 jccb(Assembler::zero, L_exit);
8913 jmpb(L_fill_8_bytes);
8915 //
8916 // length is too short, just fill qwords
8917 //
8918 BIND(L_fill_8_bytes_loop);
8919 movl(Address(to, 0), value);
8920 movl(Address(to, 4), value);
8921 addptr(to, 8);
8922 BIND(L_fill_8_bytes);
8923 subl(count, 1 << (shift + 1));
8924 jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
8925 // fall through to fill 4 bytes
8926 } else {
8927 Label L_fill_32_bytes;
8928 if (!UseUnalignedLoadStores) {
8929 // align to 8 bytes, we know we are 4 byte aligned to start
8930 testptr(to, 4);
8931 jccb(Assembler::zero, L_fill_32_bytes);
8932 movl(Address(to, 0), value);
8933 addptr(to, 4);
8934 subl(count, 1<<shift);
8935 }
8936 BIND(L_fill_32_bytes);
8937 {
8938 assert( UseSSE >= 2, "supported cpu only" );
8939 Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
8940 // Fill 32-byte chunks
8941 movdl(xtmp, value);
8942 pshufd(xtmp, xtmp, 0);
8944 subl(count, 8 << shift);
8945 jcc(Assembler::less, L_check_fill_8_bytes);
8946 align(16);
8948 BIND(L_fill_32_bytes_loop);
8950 if (UseUnalignedLoadStores) {
8951 movdqu(Address(to, 0), xtmp);
8952 movdqu(Address(to, 16), xtmp);
8953 } else {
8954 movq(Address(to, 0), xtmp);
8955 movq(Address(to, 8), xtmp);
8956 movq(Address(to, 16), xtmp);
8957 movq(Address(to, 24), xtmp);
8958 }
8960 addptr(to, 32);
8961 subl(count, 8 << shift);
8962 jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
8963 BIND(L_check_fill_8_bytes);
8964 addl(count, 8 << shift);
8965 jccb(Assembler::zero, L_exit);
8966 jmpb(L_fill_8_bytes);
8968 //
8969 // length is too short, just fill qwords
8970 //
8971 BIND(L_fill_8_bytes_loop);
8972 movq(Address(to, 0), xtmp);
8973 addptr(to, 8);
8974 BIND(L_fill_8_bytes);
8975 subl(count, 1 << (shift + 1));
8976 jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
8977 }
8978 }
8979 // fill trailing 4 bytes
8980 BIND(L_fill_4_bytes);
8981 testl(count, 1<<shift);
8982 jccb(Assembler::zero, L_fill_2_bytes);
8983 movl(Address(to, 0), value);
8984 if (t == T_BYTE || t == T_SHORT) {
8985 addptr(to, 4);
8986 BIND(L_fill_2_bytes);
8987 // fill trailing 2 bytes
8988 testl(count, 1<<(shift-1));
8989 jccb(Assembler::zero, L_fill_byte);
8990 movw(Address(to, 0), value);
8991 if (t == T_BYTE) {
8992 addptr(to, 2);
8993 BIND(L_fill_byte);
8994 // fill trailing byte
8995 testl(count, 1);
8996 jccb(Assembler::zero, L_exit);
8997 movb(Address(to, 0), value);
8998 } else {
8999 BIND(L_fill_byte);
9000 }
9001 } else {
9002 BIND(L_fill_2_bytes);
9003 }
9004 BIND(L_exit);
9005 }
9006 #undef BIND
9007 #undef BLOCK_COMMENT
9010 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
9011 switch (cond) {
9012 // Note some conditions are synonyms for others
9013 case Assembler::zero: return Assembler::notZero;
9014 case Assembler::notZero: return Assembler::zero;
9015 case Assembler::less: return Assembler::greaterEqual;
9016 case Assembler::lessEqual: return Assembler::greater;
9017 case Assembler::greater: return Assembler::lessEqual;
9018 case Assembler::greaterEqual: return Assembler::less;
9019 case Assembler::below: return Assembler::aboveEqual;
9020 case Assembler::belowEqual: return Assembler::above;
9021 case Assembler::above: return Assembler::belowEqual;
9022 case Assembler::aboveEqual: return Assembler::below;
9023 case Assembler::overflow: return Assembler::noOverflow;
9024 case Assembler::noOverflow: return Assembler::overflow;
9025 case Assembler::negative: return Assembler::positive;
9026 case Assembler::positive: return Assembler::negative;
9027 case Assembler::parity: return Assembler::noParity;
9028 case Assembler::noParity: return Assembler::parity;
9029 }
9030 ShouldNotReachHere(); return Assembler::overflow;
9031 }
9033 SkipIfEqual::SkipIfEqual(
9034 MacroAssembler* masm, const bool* flag_addr, bool value) {
9035 _masm = masm;
9036 _masm->cmp8(ExternalAddress((address)flag_addr), value);
9037 _masm->jcc(Assembler::equal, _label);
9038 }
9040 SkipIfEqual::~SkipIfEqual() {
9041 _masm->bind(_label);
9042 }