1 /*
2 * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #ifndef CPU_X86_VM_ASSEMBLER_X86_HPP
26 #define CPU_X86_VM_ASSEMBLER_X86_HPP
27
28 #include "asm/register.hpp"
29 #include "vm_version_x86.hpp"
30
31 class BiasedLockingCounters;
32
33 // Contains all the definitions needed for x86 assembly code generation.
34
35 // Calling convention
36 class Argument {
37 public:
38 enum {
39 #ifdef _LP64
40 #ifdef _WIN64
41 n_int_register_parameters_c = 4, // rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
42 n_float_register_parameters_c = 4, // xmm0 - xmm3 (c_farg0, c_farg1, ... )
43 #else
44 n_int_register_parameters_c = 6, // rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
45 n_float_register_parameters_c = 8, // xmm0 - xmm7 (c_farg0, c_farg1, ... )
46 #endif // _WIN64
47 n_int_register_parameters_j = 6, // j_rarg0, j_rarg1, ...
48 n_float_register_parameters_j = 8 // j_farg0, j_farg1, ...
49 #else
50 n_register_parameters = 0 // 0 registers used to pass arguments
51 #endif // _LP64
52 };
53 };
54
55
56 #ifdef _LP64
57 // Symbolically name the register arguments used by the c calling convention.
58 // Windows is different from linux/solaris. So much for standards...
59
60 #ifdef _WIN64
61
62 REGISTER_DECLARATION(Register, c_rarg0, rcx);
63 REGISTER_DECLARATION(Register, c_rarg1, rdx);
64 REGISTER_DECLARATION(Register, c_rarg2, r8);
65 REGISTER_DECLARATION(Register, c_rarg3, r9);
66
67 REGISTER_DECLARATION(XMMRegister, c_farg0, xmm0);
68 REGISTER_DECLARATION(XMMRegister, c_farg1, xmm1);
69 REGISTER_DECLARATION(XMMRegister, c_farg2, xmm2);
70 REGISTER_DECLARATION(XMMRegister, c_farg3, xmm3);
71
72 #else
73
74 REGISTER_DECLARATION(Register, c_rarg0, rdi);
75 REGISTER_DECLARATION(Register, c_rarg1, rsi);
76 REGISTER_DECLARATION(Register, c_rarg2, rdx);
77 REGISTER_DECLARATION(Register, c_rarg3, rcx);
78 REGISTER_DECLARATION(Register, c_rarg4, r8);
79 REGISTER_DECLARATION(Register, c_rarg5, r9);
80
81 REGISTER_DECLARATION(XMMRegister, c_farg0, xmm0);
82 REGISTER_DECLARATION(XMMRegister, c_farg1, xmm1);
83 REGISTER_DECLARATION(XMMRegister, c_farg2, xmm2);
84 REGISTER_DECLARATION(XMMRegister, c_farg3, xmm3);
85 REGISTER_DECLARATION(XMMRegister, c_farg4, xmm4);
86 REGISTER_DECLARATION(XMMRegister, c_farg5, xmm5);
87 REGISTER_DECLARATION(XMMRegister, c_farg6, xmm6);
88 REGISTER_DECLARATION(XMMRegister, c_farg7, xmm7);
89
90 #endif // _WIN64
91
92 // Symbolically name the register arguments used by the Java calling convention.
93 // We have control over the convention for java so we can do what we please.
94 // What pleases us is to offset the java calling convention so that when
95 // we call a suitable jni method the arguments are lined up and we don't
96 // have to do little shuffling. A suitable jni method is non-static and a
97 // small number of arguments (two fewer args on windows)
98 //
99 // |-------------------------------------------------------|
100 // | c_rarg0 c_rarg1 c_rarg2 c_rarg3 c_rarg4 c_rarg5 |
101 // |-------------------------------------------------------|
102 // | rcx rdx r8 r9 rdi* rsi* | windows (* not a c_rarg)
103 // | rdi rsi rdx rcx r8 r9 | solaris/linux
104 // |-------------------------------------------------------|
105 // | j_rarg5 j_rarg0 j_rarg1 j_rarg2 j_rarg3 j_rarg4 |
106 // |-------------------------------------------------------|
107
108 REGISTER_DECLARATION(Register, j_rarg0, c_rarg1);
109 REGISTER_DECLARATION(Register, j_rarg1, c_rarg2);
110 REGISTER_DECLARATION(Register, j_rarg2, c_rarg3);
111 // Windows runs out of register args here
112 #ifdef _WIN64
113 REGISTER_DECLARATION(Register, j_rarg3, rdi);
114 REGISTER_DECLARATION(Register, j_rarg4, rsi);
115 #else
116 REGISTER_DECLARATION(Register, j_rarg3, c_rarg4);
117 REGISTER_DECLARATION(Register, j_rarg4, c_rarg5);
118 #endif /* _WIN64 */
119 REGISTER_DECLARATION(Register, j_rarg5, c_rarg0);
120
121 REGISTER_DECLARATION(XMMRegister, j_farg0, xmm0);
122 REGISTER_DECLARATION(XMMRegister, j_farg1, xmm1);
123 REGISTER_DECLARATION(XMMRegister, j_farg2, xmm2);
124 REGISTER_DECLARATION(XMMRegister, j_farg3, xmm3);
125 REGISTER_DECLARATION(XMMRegister, j_farg4, xmm4);
126 REGISTER_DECLARATION(XMMRegister, j_farg5, xmm5);
127 REGISTER_DECLARATION(XMMRegister, j_farg6, xmm6);
128 REGISTER_DECLARATION(XMMRegister, j_farg7, xmm7);
129
130 REGISTER_DECLARATION(Register, rscratch1, r10); // volatile
131 REGISTER_DECLARATION(Register, rscratch2, r11); // volatile
132
133 REGISTER_DECLARATION(Register, r12_heapbase, r12); // callee-saved
134 REGISTER_DECLARATION(Register, r15_thread, r15); // callee-saved
135
136 #else
137 // rscratch1 will apear in 32bit code that is dead but of course must compile
138 // Using noreg ensures if the dead code is incorrectly live and executed it
139 // will cause an assertion failure
140 #define rscratch1 noreg
141 #define rscratch2 noreg
142
143 #endif // _LP64
144
145 // JSR 292
146 // On x86, the SP does not have to be saved when invoking method handle intrinsics
147 // or compiled lambda forms. We indicate that by setting rbp_mh_SP_save to noreg.
148 REGISTER_DECLARATION(Register, rbp_mh_SP_save, noreg);
149
150 // Address is an abstraction used to represent a memory location
151 // using any of the amd64 addressing modes with one object.
152 //
153 // Note: A register location is represented via a Register, not
154 // via an address for efficiency & simplicity reasons.
155
156 class ArrayAddress;
157
158 class Address {
159 public:
160 enum ScaleFactor {
161 no_scale = -1,
162 times_1 = 0,
163 times_2 = 1,
164 times_4 = 2,
165 times_8 = 3,
166 times_ptr = LP64_ONLY(times_8) NOT_LP64(times_4)
167 };
168 static ScaleFactor times(int size) {
169 assert(size >= 1 && size <= 8 && is_power_of_2(size), "bad scale size");
170 if (size == 8) return times_8;
171 if (size == 4) return times_4;
172 if (size == 2) return times_2;
173 return times_1;
174 }
175 static int scale_size(ScaleFactor scale) {
176 assert(scale != no_scale, "");
177 assert(((1 << (int)times_1) == 1 &&
178 (1 << (int)times_2) == 2 &&
179 (1 << (int)times_4) == 4 &&
180 (1 << (int)times_8) == 8), "");
181 return (1 << (int)scale);
182 }
183
184 private:
185 Register _base;
186 Register _index;
187 ScaleFactor _scale;
188 int _disp;
189 RelocationHolder _rspec;
190
191 // Easily misused constructors make them private
192 // %%% can we make these go away?
193 NOT_LP64(Address(address loc, RelocationHolder spec);)
194 Address(int disp, address loc, relocInfo::relocType rtype);
195 Address(int disp, address loc, RelocationHolder spec);
196
197 public:
198
199 int disp() { return _disp; }
200 // creation
201 Address()
202 : _base(noreg),
203 _index(noreg),
204 _scale(no_scale),
205 _disp(0) {
206 }
207
208 // No default displacement otherwise Register can be implicitly
209 // converted to 0(Register) which is quite a different animal.
210
211 Address(Register base, int disp)
212 : _base(base),
213 _index(noreg),
214 _scale(no_scale),
215 _disp(disp) {
216 }
217
218 Address(Register base, Register index, ScaleFactor scale, int disp = 0)
219 : _base (base),
220 _index(index),
221 _scale(scale),
222 _disp (disp) {
223 assert(!index->is_valid() == (scale == Address::no_scale),
224 "inconsistent address");
225 }
226
227 Address(Register base, RegisterOrConstant index, ScaleFactor scale = times_1, int disp = 0)
228 : _base (base),
229 _index(index.register_or_noreg()),
230 _scale(scale),
231 _disp (disp + (index.constant_or_zero() * scale_size(scale))) {
232 if (!index.is_register()) scale = Address::no_scale;
233 assert(!_index->is_valid() == (scale == Address::no_scale),
234 "inconsistent address");
235 }
236
237 Address plus_disp(int disp) const {
238 Address a = (*this);
239 a._disp += disp;
240 return a;
241 }
242 Address plus_disp(RegisterOrConstant disp, ScaleFactor scale = times_1) const {
243 Address a = (*this);
244 a._disp += disp.constant_or_zero() * scale_size(scale);
245 if (disp.is_register()) {
246 assert(!a.index()->is_valid(), "competing indexes");
247 a._index = disp.as_register();
248 a._scale = scale;
249 }
250 return a;
251 }
252 bool is_same_address(Address a) const {
253 // disregard _rspec
254 return _base == a._base && _disp == a._disp && _index == a._index && _scale == a._scale;
255 }
256
257 // The following two overloads are used in connection with the
258 // ByteSize type (see sizes.hpp). They simplify the use of
259 // ByteSize'd arguments in assembly code. Note that their equivalent
260 // for the optimized build are the member functions with int disp
261 // argument since ByteSize is mapped to an int type in that case.
262 //
263 // Note: DO NOT introduce similar overloaded functions for WordSize
264 // arguments as in the optimized mode, both ByteSize and WordSize
265 // are mapped to the same type and thus the compiler cannot make a
266 // distinction anymore (=> compiler errors).
267
268 #ifdef ASSERT
269 Address(Register base, ByteSize disp)
270 : _base(base),
271 _index(noreg),
272 _scale(no_scale),
273 _disp(in_bytes(disp)) {
274 }
275
276 Address(Register base, Register index, ScaleFactor scale, ByteSize disp)
277 : _base(base),
278 _index(index),
279 _scale(scale),
280 _disp(in_bytes(disp)) {
281 assert(!index->is_valid() == (scale == Address::no_scale),
282 "inconsistent address");
283 }
284
285 Address(Register base, RegisterOrConstant index, ScaleFactor scale, ByteSize disp)
286 : _base (base),
287 _index(index.register_or_noreg()),
288 _scale(scale),
289 _disp (in_bytes(disp) + (index.constant_or_zero() * scale_size(scale))) {
290 if (!index.is_register()) scale = Address::no_scale;
291 assert(!_index->is_valid() == (scale == Address::no_scale),
292 "inconsistent address");
293 }
294
295 #endif // ASSERT
296
297 // accessors
298 bool uses(Register reg) const { return _base == reg || _index == reg; }
299 Register base() const { return _base; }
300 Register index() const { return _index; }
301 ScaleFactor scale() const { return _scale; }
302 int disp() const { return _disp; }
303
304 // Convert the raw encoding form into the form expected by the constructor for
305 // Address. An index of 4 (rsp) corresponds to having no index, so convert
306 // that to noreg for the Address constructor.
307 static Address make_raw(int base, int index, int scale, int disp, relocInfo::relocType disp_reloc);
308
309 static Address make_array(ArrayAddress);
310
311 private:
312 bool base_needs_rex() const {
313 return _base != noreg && _base->encoding() >= 8;
314 }
315
316 bool index_needs_rex() const {
317 return _index != noreg &&_index->encoding() >= 8;
318 }
319
320 relocInfo::relocType reloc() const { return _rspec.type(); }
321
322 friend class Assembler;
323 friend class MacroAssembler;
324 friend class LIR_Assembler; // base/index/scale/disp
325 };
326
327 //
328 // AddressLiteral has been split out from Address because operands of this type
329 // need to be treated specially on 32bit vs. 64bit platforms. By splitting it out
330 // the few instructions that need to deal with address literals are unique and the
331 // MacroAssembler does not have to implement every instruction in the Assembler
332 // in order to search for address literals that may need special handling depending
333 // on the instruction and the platform. As small step on the way to merging i486/amd64
334 // directories.
335 //
336 class AddressLiteral {
337 friend class ArrayAddress;
338 RelocationHolder _rspec;
339 // Typically we use AddressLiterals we want to use their rval
340 // However in some situations we want the lval (effect address) of the item.
341 // We provide a special factory for making those lvals.
342 bool _is_lval;
343
344 // If the target is far we'll need to load the ea of this to
345 // a register to reach it. Otherwise if near we can do rip
346 // relative addressing.
347
348 address _target;
349
350 protected:
351 // creation
352 AddressLiteral()
353 : _is_lval(false),
354 _target(NULL)
355 {}
356
357 public:
358
359
360 AddressLiteral(address target, relocInfo::relocType rtype);
361
362 AddressLiteral(address target, RelocationHolder const& rspec)
363 : _rspec(rspec),
364 _is_lval(false),
365 _target(target)
366 {}
367
368 AddressLiteral addr() {
369 AddressLiteral ret = *this;
370 ret._is_lval = true;
371 return ret;
372 }
373
374
375 private:
376
377 address target() { return _target; }
378 bool is_lval() { return _is_lval; }
379
380 relocInfo::relocType reloc() const { return _rspec.type(); }
381 const RelocationHolder& rspec() const { return _rspec; }
382
383 friend class Assembler;
384 friend class MacroAssembler;
385 friend class Address;
386 friend class LIR_Assembler;
387 };
388
389 // Convience classes
390 class RuntimeAddress: public AddressLiteral {
391
392 public:
393
394 RuntimeAddress(address target) : AddressLiteral(target, relocInfo::runtime_call_type) {}
395
396 };
397
398 class ExternalAddress: public AddressLiteral {
399 private:
400 static relocInfo::relocType reloc_for_target(address target) {
401 // Sometimes ExternalAddress is used for values which aren't
402 // exactly addresses, like the card table base.
403 // external_word_type can't be used for values in the first page
404 // so just skip the reloc in that case.
405 return external_word_Relocation::can_be_relocated(target) ? relocInfo::external_word_type : relocInfo::none;
406 }
407
408 public:
409
410 ExternalAddress(address target) : AddressLiteral(target, reloc_for_target(target)) {}
411
412 };
413
414 class InternalAddress: public AddressLiteral {
415
416 public:
417
418 InternalAddress(address target) : AddressLiteral(target, relocInfo::internal_word_type) {}
419
420 };
421
422 // x86 can do array addressing as a single operation since disp can be an absolute
423 // address amd64 can't. We create a class that expresses the concept but does extra
424 // magic on amd64 to get the final result
425
426 class ArrayAddress {
427 private:
428
429 AddressLiteral _base;
430 Address _index;
431
432 public:
433
434 ArrayAddress() {};
435 ArrayAddress(AddressLiteral base, Address index): _base(base), _index(index) {};
436 AddressLiteral base() { return _base; }
437 Address index() { return _index; }
438
439 };
440
441 class InstructionAttr;
442
443 // 64-bit refect the fxsave size which is 512 bytes and the new xsave area on EVEX which is another 2176 bytes
444 // See fxsave and xsave(EVEX enabled) documentation for layout
445 const int FPUStateSizeInWords = NOT_LP64(27) LP64_ONLY(2688 / wordSize);
446
447 // The Intel x86/Amd64 Assembler: Pure assembler doing NO optimizations on the instruction
448 // level (e.g. mov rax, 0 is not translated into xor rax, rax!); i.e., what you write
449 // is what you get. The Assembler is generating code into a CodeBuffer.
450
451 class Assembler : public AbstractAssembler {
452 friend class AbstractAssembler; // for the non-virtual hack
453 friend class LIR_Assembler; // as_Address()
454 friend class StubGenerator;
455
456 public:
457 enum Condition { // The x86 condition codes used for conditional jumps/moves.
458 zero = 0x4,
459 notZero = 0x5,
460 equal = 0x4,
461 notEqual = 0x5,
462 less = 0xc,
463 lessEqual = 0xe,
464 greater = 0xf,
465 greaterEqual = 0xd,
466 below = 0x2,
467 belowEqual = 0x6,
468 above = 0x7,
469 aboveEqual = 0x3,
470 overflow = 0x0,
471 noOverflow = 0x1,
472 carrySet = 0x2,
473 carryClear = 0x3,
474 negative = 0x8,
475 positive = 0x9,
476 parity = 0xa,
477 noParity = 0xb
478 };
479
480 enum Prefix {
481 // segment overrides
482 CS_segment = 0x2e,
483 SS_segment = 0x36,
484 DS_segment = 0x3e,
485 ES_segment = 0x26,
486 FS_segment = 0x64,
487 GS_segment = 0x65,
488
489 REX = 0x40,
490
491 REX_B = 0x41,
492 REX_X = 0x42,
493 REX_XB = 0x43,
494 REX_R = 0x44,
495 REX_RB = 0x45,
496 REX_RX = 0x46,
497 REX_RXB = 0x47,
498
499 REX_W = 0x48,
500
501 REX_WB = 0x49,
502 REX_WX = 0x4A,
503 REX_WXB = 0x4B,
504 REX_WR = 0x4C,
505 REX_WRB = 0x4D,
506 REX_WRX = 0x4E,
507 REX_WRXB = 0x4F,
508
509 VEX_3bytes = 0xC4,
510 VEX_2bytes = 0xC5,
511 EVEX_4bytes = 0x62,
512 Prefix_EMPTY = 0x0
513 };
514
515 enum VexPrefix {
516 VEX_B = 0x20,
517 VEX_X = 0x40,
518 VEX_R = 0x80,
519 VEX_W = 0x80
520 };
521
522 enum ExexPrefix {
523 EVEX_F = 0x04,
524 EVEX_V = 0x08,
525 EVEX_Rb = 0x10,
526 EVEX_X = 0x40,
527 EVEX_Z = 0x80
528 };
529
530 enum VexSimdPrefix {
531 VEX_SIMD_NONE = 0x0,
532 VEX_SIMD_66 = 0x1,
533 VEX_SIMD_F3 = 0x2,
534 VEX_SIMD_F2 = 0x3
535 };
536
537 enum VexOpcode {
538 VEX_OPCODE_NONE = 0x0,
539 VEX_OPCODE_0F = 0x1,
540 VEX_OPCODE_0F_38 = 0x2,
541 VEX_OPCODE_0F_3A = 0x3,
542 VEX_OPCODE_MASK = 0x1F
543 };
544
545 enum AvxVectorLen {
546 AVX_128bit = 0x0,
547 AVX_256bit = 0x1,
548 AVX_512bit = 0x2,
549 AVX_NoVec = 0x4
550 };
551
552 enum EvexTupleType {
553 EVEX_FV = 0,
554 EVEX_HV = 4,
555 EVEX_FVM = 6,
556 EVEX_T1S = 7,
557 EVEX_T1F = 11,
558 EVEX_T2 = 13,
559 EVEX_T4 = 15,
560 EVEX_T8 = 17,
561 EVEX_HVM = 18,
562 EVEX_QVM = 19,
563 EVEX_OVM = 20,
564 EVEX_M128 = 21,
565 EVEX_DUP = 22,
566 EVEX_ETUP = 23
567 };
568
569 enum EvexInputSizeInBits {
570 EVEX_8bit = 0,
571 EVEX_16bit = 1,
572 EVEX_32bit = 2,
573 EVEX_64bit = 3,
574 EVEX_NObit = 4
575 };
576
577 enum WhichOperand {
578 // input to locate_operand, and format code for relocations
579 imm_operand = 0, // embedded 32-bit|64-bit immediate operand
580 disp32_operand = 1, // embedded 32-bit displacement or address
581 call32_operand = 2, // embedded 32-bit self-relative displacement
582 #ifndef _LP64
583 _WhichOperand_limit = 3
584 #else
585 narrow_oop_operand = 3, // embedded 32-bit immediate narrow oop
586 _WhichOperand_limit = 4
587 #endif
588 };
589
590 // Comparison predicates for integral types & FP types when using SSE
591 enum ComparisonPredicate {
592 eq = 0,
593 lt = 1,
594 le = 2,
595 _false = 3,
596 neq = 4,
597 nlt = 5,
598 nle = 6,
599 _true = 7
600 };
601
602 // Comparison predicates for FP types when using AVX
603 // O means ordered. U is unordered. When using ordered, any NaN comparison is false. Otherwise, it is true.
604 // S means signaling. Q means non-signaling. When signaling is true, instruction signals #IA on NaN.
605 enum ComparisonPredicateFP {
606 EQ_OQ = 0,
607 LT_OS = 1,
608 LE_OS = 2,
609 UNORD_Q = 3,
610 NEQ_UQ = 4,
611 NLT_US = 5,
612 NLE_US = 6,
613 ORD_Q = 7,
614 EQ_UQ = 8,
615 NGE_US = 9,
616 NGT_US = 0xA,
617 FALSE_OQ = 0XB,
618 NEQ_OQ = 0xC,
619 GE_OS = 0xD,
620 GT_OS = 0xE,
621 TRUE_UQ = 0xF,
622 EQ_OS = 0x10,
623 LT_OQ = 0x11,
624 LE_OQ = 0x12,
625 UNORD_S = 0x13,
626 NEQ_US = 0x14,
627 NLT_UQ = 0x15,
628 NLE_UQ = 0x16,
629 ORD_S = 0x17,
630 EQ_US = 0x18,
631 NGE_UQ = 0x19,
632 NGT_UQ = 0x1A,
633 FALSE_OS = 0x1B,
634 NEQ_OS = 0x1C,
635 GE_OQ = 0x1D,
636 GT_OQ = 0x1E,
637 TRUE_US =0x1F
638 };
639
640
641 // NOTE: The general philopsophy of the declarations here is that 64bit versions
642 // of instructions are freely declared without the need for wrapping them an ifdef.
643 // (Some dangerous instructions are ifdef's out of inappropriate jvm's.)
644 // In the .cpp file the implementations are wrapped so that they are dropped out
645 // of the resulting jvm. This is done mostly to keep the footprint of MINIMAL
646 // to the size it was prior to merging up the 32bit and 64bit assemblers.
647 //
648 // This does mean you'll get a linker/runtime error if you use a 64bit only instruction
649 // in a 32bit vm. This is somewhat unfortunate but keeps the ifdef noise down.
650
651 private:
652
653 bool _legacy_mode_bw;
654 bool _legacy_mode_dq;
655 bool _legacy_mode_vl;
656 bool _legacy_mode_vlbw;
657 bool _is_managed;
658 bool _vector_masking; // For stub code use only
659
660 class InstructionAttr *_attributes;
661
662 // 64bit prefixes
663 int prefix_and_encode(int reg_enc, bool byteinst = false);
664 int prefixq_and_encode(int reg_enc);
665
666 int prefix_and_encode(int dst_enc, int src_enc) {
667 return prefix_and_encode(dst_enc, false, src_enc, false);
668 }
669 int prefix_and_encode(int dst_enc, bool dst_is_byte, int src_enc, bool src_is_byte);
670 int prefixq_and_encode(int dst_enc, int src_enc);
671
672 void prefix(Register reg);
673 void prefix(Register dst, Register src, Prefix p);
674 void prefix(Register dst, Address adr, Prefix p);
675 void prefix(Address adr);
676 void prefixq(Address adr);
677
678 void prefix(Address adr, Register reg, bool byteinst = false);
679 void prefix(Address adr, XMMRegister reg);
680 void prefixq(Address adr, Register reg);
681 void prefixq(Address adr, XMMRegister reg);
682
683 void prefetch_prefix(Address src);
684
685 void rex_prefix(Address adr, XMMRegister xreg,
686 VexSimdPrefix pre, VexOpcode opc, bool rex_w);
687 int rex_prefix_and_encode(int dst_enc, int src_enc,
688 VexSimdPrefix pre, VexOpcode opc, bool rex_w);
689
690 void vex_prefix(bool vex_r, bool vex_b, bool vex_x, int nds_enc, VexSimdPrefix pre, VexOpcode opc);
691
692 void evex_prefix(bool vex_r, bool vex_b, bool vex_x, bool evex_r, bool evex_v,
693 int nds_enc, VexSimdPrefix pre, VexOpcode opc);
694
695 void vex_prefix(Address adr, int nds_enc, int xreg_enc,
696 VexSimdPrefix pre, VexOpcode opc,
697 InstructionAttr *attributes);
698
699 int vex_prefix_and_encode(int dst_enc, int nds_enc, int src_enc,
700 VexSimdPrefix pre, VexOpcode opc,
701 InstructionAttr *attributes);
702
703 void simd_prefix(XMMRegister xreg, XMMRegister nds, Address adr, VexSimdPrefix pre,
704 VexOpcode opc, InstructionAttr *attributes);
705
706 int simd_prefix_and_encode(XMMRegister dst, XMMRegister nds, XMMRegister src, VexSimdPrefix pre,
707 VexOpcode opc, InstructionAttr *attributes);
708
709 // Helper functions for groups of instructions
710 void emit_arith_b(int op1, int op2, Register dst, int imm8);
711
712 void emit_arith(int op1, int op2, Register dst, int32_t imm32);
713 // Force generation of a 4 byte immediate value even if it fits into 8bit
714 void emit_arith_imm32(int op1, int op2, Register dst, int32_t imm32);
715 void emit_arith(int op1, int op2, Register dst, Register src);
716
717 bool emit_compressed_disp_byte(int &disp);
718
719 void emit_operand(Register reg,
720 Register base, Register index, Address::ScaleFactor scale,
721 int disp,
722 RelocationHolder const& rspec,
723 int rip_relative_correction = 0);
724
725 void emit_operand(Register reg, Address adr, int rip_relative_correction = 0);
726
727 // operands that only take the original 32bit registers
728 void emit_operand32(Register reg, Address adr);
729
730 void emit_operand(XMMRegister reg,
731 Register base, Register index, Address::ScaleFactor scale,
732 int disp,
733 RelocationHolder const& rspec);
734
735 void emit_operand(XMMRegister reg, Address adr);
736
737 void emit_operand(MMXRegister reg, Address adr);
738
739 // workaround gcc (3.2.1-7) bug
740 void emit_operand(Address adr, MMXRegister reg);
741
742
743 // Immediate-to-memory forms
744 void emit_arith_operand(int op1, Register rm, Address adr, int32_t imm32);
745
746 void emit_farith(int b1, int b2, int i);
747
748
749 protected:
750 #ifdef ASSERT
751 void check_relocation(RelocationHolder const& rspec, int format);
752 #endif
753
754 void emit_data(jint data, relocInfo::relocType rtype, int format);
755 void emit_data(jint data, RelocationHolder const& rspec, int format);
756 void emit_data64(jlong data, relocInfo::relocType rtype, int format = 0);
757 void emit_data64(jlong data, RelocationHolder const& rspec, int format = 0);
758
759 bool reachable(AddressLiteral adr) NOT_LP64({ return true;});
760
761 // These are all easily abused and hence protected
762
763 // 32BIT ONLY SECTION
764 #ifndef _LP64
765 // Make these disappear in 64bit mode since they would never be correct
766 void cmp_literal32(Register src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY
767 void cmp_literal32(Address src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY
768
769 void mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY
770 void mov_literal32(Address dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY
771
772 void push_literal32(int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY
773 #else
774 // 64BIT ONLY SECTION
775 void mov_literal64(Register dst, intptr_t imm64, RelocationHolder const& rspec); // 64BIT ONLY
776
777 void cmp_narrow_oop(Register src1, int32_t imm32, RelocationHolder const& rspec);
778 void cmp_narrow_oop(Address src1, int32_t imm32, RelocationHolder const& rspec);
779
780 void mov_narrow_oop(Register dst, int32_t imm32, RelocationHolder const& rspec);
781 void mov_narrow_oop(Address dst, int32_t imm32, RelocationHolder const& rspec);
782 #endif // _LP64
783
784 // These are unique in that we are ensured by the caller that the 32bit
785 // relative in these instructions will always be able to reach the potentially
786 // 64bit address described by entry. Since they can take a 64bit address they
787 // don't have the 32 suffix like the other instructions in this class.
788
789 void call_literal(address entry, RelocationHolder const& rspec);
790 void jmp_literal(address entry, RelocationHolder const& rspec);
791
792 // Avoid using directly section
793 // Instructions in this section are actually usable by anyone without danger
794 // of failure but have performance issues that are addressed my enhanced
795 // instructions which will do the proper thing base on the particular cpu.
796 // We protect them because we don't trust you...
797
798 // Don't use next inc() and dec() methods directly. INC & DEC instructions
799 // could cause a partial flag stall since they don't set CF flag.
800 // Use MacroAssembler::decrement() & MacroAssembler::increment() methods
801 // which call inc() & dec() or add() & sub() in accordance with
802 // the product flag UseIncDec value.
803
804 void decl(Register dst);
805 void decl(Address dst);
806 void decq(Register dst);
807 void decq(Address dst);
808
809 void incl(Register dst);
810 void incl(Address dst);
811 void incq(Register dst);
812 void incq(Address dst);
813
814 // New cpus require use of movsd and movss to avoid partial register stall
815 // when loading from memory. But for old Opteron use movlpd instead of movsd.
816 // The selection is done in MacroAssembler::movdbl() and movflt().
817
818 // Move Scalar Single-Precision Floating-Point Values
819 void movss(XMMRegister dst, Address src);
820 void movss(XMMRegister dst, XMMRegister src);
821 void movss(Address dst, XMMRegister src);
822
823 // Move Scalar Double-Precision Floating-Point Values
824 void movsd(XMMRegister dst, Address src);
825 void movsd(XMMRegister dst, XMMRegister src);
826 void movsd(Address dst, XMMRegister src);
827 void movlpd(XMMRegister dst, Address src);
828
829 // New cpus require use of movaps and movapd to avoid partial register stall
830 // when moving between registers.
831 void movaps(XMMRegister dst, XMMRegister src);
832 void movapd(XMMRegister dst, XMMRegister src);
833
834 // End avoid using directly
835
836
837 // Instruction prefixes
838 void prefix(Prefix p);
839
840 public:
841
842 // Creation
843 Assembler(CodeBuffer* code) : AbstractAssembler(code) {
844 init_attributes();
845 }
846
847 // Decoding
848 static address locate_operand(address inst, WhichOperand which);
849 static address locate_next_instruction(address inst);
850
851 // Utilities
852 static bool is_polling_page_far() NOT_LP64({ return false;});
853 static bool query_compressed_disp_byte(int disp, bool is_evex_inst, int vector_len,
854 int cur_tuple_type, int in_size_in_bits, int cur_encoding);
855
856 // Generic instructions
857 // Does 32bit or 64bit as needed for the platform. In some sense these
858 // belong in macro assembler but there is no need for both varieties to exist
859
860 void init_attributes(void) {
861 _legacy_mode_bw = (VM_Version::supports_avx512bw() == false);
862 _legacy_mode_dq = (VM_Version::supports_avx512dq() == false);
863 _legacy_mode_vl = (VM_Version::supports_avx512vl() == false);
864 _legacy_mode_vlbw = (VM_Version::supports_avx512vlbw() == false);
865 _is_managed = false;
866 _vector_masking = false;
867 _attributes = NULL;
868 }
869
870 void set_attributes(InstructionAttr *attributes) { _attributes = attributes; }
871 void clear_attributes(void) { _attributes = NULL; }
872
873 void set_managed(void) { _is_managed = true; }
874 void clear_managed(void) { _is_managed = false; }
875 bool is_managed(void) { return _is_managed; }
876
877 // Following functions are for stub code use only
878 void set_vector_masking(void) { _vector_masking = true; }
879 void clear_vector_masking(void) { _vector_masking = false; }
880 bool is_vector_masking(void) { return _vector_masking; }
881
882 void lea(Register dst, Address src);
883
884 void mov(Register dst, Register src);
885
886 void pusha();
887 void popa();
888
889 void pushf();
890 void popf();
891
892 void push(int32_t imm32);
893
894 void push(Register src);
895
896 void pop(Register dst);
897
898 // These are dummies to prevent surprise implicit conversions to Register
899 void push(void* v);
900 void pop(void* v);
901
902 // These do register sized moves/scans
903 void rep_mov();
904 void rep_stos();
905 void rep_stosb();
906 void repne_scan();
907 #ifdef _LP64
908 void repne_scanl();
909 #endif
910
911 // Vanilla instructions in lexical order
912
913 void adcl(Address dst, int32_t imm32);
914 void adcl(Address dst, Register src);
915 void adcl(Register dst, int32_t imm32);
916 void adcl(Register dst, Address src);
917 void adcl(Register dst, Register src);
918
919 void adcq(Register dst, int32_t imm32);
920 void adcq(Register dst, Address src);
921 void adcq(Register dst, Register src);
922
923 void addb(Register dst, Register src);
924 void addb(Address dst, int imm8);
925 void addw(Register dst, Register src);
926 void addw(Address dst, int imm16);
927
928 void addl(Address dst, int32_t imm32);
929 void addl(Address dst, Register src);
930 void addl(Register dst, int32_t imm32);
931 void addl(Register dst, Address src);
932 void addl(Register dst, Register src);
933
934 void addq(Address dst, int32_t imm32);
935 void addq(Address dst, Register src);
936 void addq(Register dst, int32_t imm32);
937 void addq(Register dst, Address src);
938 void addq(Register dst, Register src);
939
940 #ifdef _LP64
941 //Add Unsigned Integers with Carry Flag
942 void adcxq(Register dst, Register src);
943
944 //Add Unsigned Integers with Overflow Flag
945 void adoxq(Register dst, Register src);
946 #endif
947
948 void addr_nop_4();
949 void addr_nop_5();
950 void addr_nop_7();
951 void addr_nop_8();
952
953 // Add Scalar Double-Precision Floating-Point Values
954 void addsd(XMMRegister dst, Address src);
955 void addsd(XMMRegister dst, XMMRegister src);
956
957 // Add Scalar Single-Precision Floating-Point Values
958 void addss(XMMRegister dst, Address src);
959 void addss(XMMRegister dst, XMMRegister src);
960
961 // AES instructions
962 void aesdec(XMMRegister dst, Address src);
963 void aesdec(XMMRegister dst, XMMRegister src);
964 void aesdeclast(XMMRegister dst, Address src);
965 void aesdeclast(XMMRegister dst, XMMRegister src);
966 void aesenc(XMMRegister dst, Address src);
967 void aesenc(XMMRegister dst, XMMRegister src);
968 void aesenclast(XMMRegister dst, Address src);
969 void aesenclast(XMMRegister dst, XMMRegister src);
970
971 void andb(Register dst, Register src);
972 void andw(Register dst, Register src);
973
974 void andl(Address dst, int32_t imm32);
975 void andl(Register dst, int32_t imm32);
976 void andl(Register dst, Address src);
977 void andl(Register dst, Register src);
978
979 void andq(Address dst, int32_t imm32);
980 void andq(Register dst, int32_t imm32);
981 void andq(Register dst, Address src);
982 void andq(Register dst, Register src);
983
984 // BMI instructions
985 void andnl(Register dst, Register src1, Register src2);
986 void andnl(Register dst, Register src1, Address src2);
987 void andnq(Register dst, Register src1, Register src2);
988 void andnq(Register dst, Register src1, Address src2);
989
990 void blsil(Register dst, Register src);
991 void blsil(Register dst, Address src);
992 void blsiq(Register dst, Register src);
993 void blsiq(Register dst, Address src);
994
995 void blsmskl(Register dst, Register src);
996 void blsmskl(Register dst, Address src);
997 void blsmskq(Register dst, Register src);
998 void blsmskq(Register dst, Address src);
999
1000 void blsrl(Register dst, Register src);
1001 void blsrl(Register dst, Address src);
1002 void blsrq(Register dst, Register src);
1003 void blsrq(Register dst, Address src);
1004
1005 void bsfl(Register dst, Register src);
1006 void bsrl(Register dst, Register src);
1007
1008 #ifdef _LP64
1009 void bsfq(Register dst, Register src);
1010 void bsrq(Register dst, Register src);
1011 #endif
1012
1013 void bswapl(Register reg);
1014
1015 void bswapq(Register reg);
1016
1017 void call(Label& L, relocInfo::relocType rtype);
1018 void call(Register reg); // push pc; pc <- reg
1019 void call(Address adr); // push pc; pc <- adr
1020
1021 void cdql();
1022
1023 void cdqq();
1024
1025 void cld();
1026
1027 void clflush(Address adr);
1028
1029 void cmovl(Condition cc, Register dst, Register src);
1030 void cmovl(Condition cc, Register dst, Address src);
1031
1032 void cmovq(Condition cc, Register dst, Register src);
1033 void cmovq(Condition cc, Register dst, Address src);
1034
1035
1036 void cmpb(Address dst, int imm8);
1037
1038 void cmpl(Address dst, int32_t imm32);
1039
1040 void cmpl(Register dst, int32_t imm32);
1041 void cmpl(Register dst, Register src);
1042 void cmpl(Register dst, Address src);
1043
1044 void cmpq(Address dst, int32_t imm32);
1045 void cmpq(Address dst, Register src);
1046
1047 void cmpq(Register dst, int32_t imm32);
1048 void cmpq(Register dst, Register src);
1049 void cmpq(Register dst, Address src);
1050
1051 // these are dummies used to catch attempting to convert NULL to Register
1052 void cmpl(Register dst, void* junk); // dummy
1053 void cmpq(Register dst, void* junk); // dummy
1054
1055 void cmpw(Address dst, int imm16);
1056
1057 void cmpxchg8 (Address adr);
1058
1059 void cmpxchgb(Register reg, Address adr);
1060 void cmpxchgl(Register reg, Address adr);
1061
1062 void cmpxchgq(Register reg, Address adr);
1063
1064 // Ordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS
1065 void comisd(XMMRegister dst, Address src);
1066 void comisd(XMMRegister dst, XMMRegister src);
1067
1068 // Ordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS
1069 void comiss(XMMRegister dst, Address src);
1070 void comiss(XMMRegister dst, XMMRegister src);
1071
1072 // Identify processor type and features
1073 void cpuid();
1074
1075 // CRC32C
1076 void crc32(Register crc, Register v, int8_t sizeInBytes);
1077 void crc32(Register crc, Address adr, int8_t sizeInBytes);
1078
1079 // Convert Scalar Double-Precision Floating-Point Value to Scalar Single-Precision Floating-Point Value
1080 void cvtsd2ss(XMMRegister dst, XMMRegister src);
1081 void cvtsd2ss(XMMRegister dst, Address src);
1082
1083 // Convert Doubleword Integer to Scalar Double-Precision Floating-Point Value
1084 void cvtsi2sdl(XMMRegister dst, Register src);
1085 void cvtsi2sdl(XMMRegister dst, Address src);
1086 void cvtsi2sdq(XMMRegister dst, Register src);
1087 void cvtsi2sdq(XMMRegister dst, Address src);
1088
1089 // Convert Doubleword Integer to Scalar Single-Precision Floating-Point Value
1090 void cvtsi2ssl(XMMRegister dst, Register src);
1091 void cvtsi2ssl(XMMRegister dst, Address src);
1092 void cvtsi2ssq(XMMRegister dst, Register src);
1093 void cvtsi2ssq(XMMRegister dst, Address src);
1094
1095 // Convert Packed Signed Doubleword Integers to Packed Double-Precision Floating-Point Value
1096 void cvtdq2pd(XMMRegister dst, XMMRegister src);
1097
1098 // Convert Packed Signed Doubleword Integers to Packed Single-Precision Floating-Point Value
1099 void cvtdq2ps(XMMRegister dst, XMMRegister src);
1100
1101 // Convert Scalar Single-Precision Floating-Point Value to Scalar Double-Precision Floating-Point Value
1102 void cvtss2sd(XMMRegister dst, XMMRegister src);
1103 void cvtss2sd(XMMRegister dst, Address src);
1104
1105 // Convert with Truncation Scalar Double-Precision Floating-Point Value to Doubleword Integer
1106 void cvttsd2sil(Register dst, Address src);
1107 void cvttsd2sil(Register dst, XMMRegister src);
1108 void cvttsd2siq(Register dst, XMMRegister src);
1109
1110 // Convert with Truncation Scalar Single-Precision Floating-Point Value to Doubleword Integer
1111 void cvttss2sil(Register dst, XMMRegister src);
1112 void cvttss2siq(Register dst, XMMRegister src);
1113
1114 void cvttpd2dq(XMMRegister dst, XMMRegister src);
1115
1116 // Convert vector float and double
1117 void vcvtps2pd(XMMRegister dst, XMMRegister src, int vector_len);
1118 void evcvtps2pd(XMMRegister dst, XMMRegister src, int vector_len);
1119
1120 //Abs of packed Integer values
1121 void pabsd(XMMRegister dst, XMMRegister src);
1122 void vpabsb(XMMRegister dst, XMMRegister src, int vector_len);
1123 void vpabsw(XMMRegister dst, XMMRegister src, int vector_len);
1124 void vpabsd(XMMRegister dst, XMMRegister src, int vector_len);
1125 void evpabsd(XMMRegister dst, XMMRegister src, int vector_len);
1126
1127 // Divide Scalar Double-Precision Floating-Point Values
1128 void divsd(XMMRegister dst, Address src);
1129 void divsd(XMMRegister dst, XMMRegister src);
1130
1131 // Divide Scalar Single-Precision Floating-Point Values
1132 void divss(XMMRegister dst, Address src);
1133 void divss(XMMRegister dst, XMMRegister src);
1134
1135 void emms();
1136
1137 void fabs();
1138
1139 void fadd(int i);
1140
1141 void fadd_d(Address src);
1142 void fadd_s(Address src);
1143
1144 // "Alternate" versions of x87 instructions place result down in FPU
1145 // stack instead of on TOS
1146
1147 void fadda(int i); // "alternate" fadd
1148 void faddp(int i = 1);
1149
1150 void fchs();
1151
1152 void fcom(int i);
1153
1154 void fcomp(int i = 1);
1155 void fcomp_d(Address src);
1156 void fcomp_s(Address src);
1157
1158 void fcompp();
1159
1160 void fcos();
1161
1162 void fdecstp();
1163
1164 void fdiv(int i);
1165 void fdiv_d(Address src);
1166 void fdivr_s(Address src);
1167 void fdiva(int i); // "alternate" fdiv
1168 void fdivp(int i = 1);
1169
1170 void fdivr(int i);
1171 void fdivr_d(Address src);
1172 void fdiv_s(Address src);
1173
1174 void fdivra(int i); // "alternate" reversed fdiv
1175
1176 void fdivrp(int i = 1);
1177
1178 void ffree(int i = 0);
1179
1180 void fild_d(Address adr);
1181 void fild_s(Address adr);
1182
1183 void fincstp();
1184
1185 void finit();
1186
1187 void fist_s (Address adr);
1188 void fistp_d(Address adr);
1189 void fistp_s(Address adr);
1190
1191 void fld1();
1192
1193 void fld_d(Address adr);
1194 void fld_s(Address adr);
1195 void fld_s(int index);
1196 void fld_x(Address adr); // extended-precision (80-bit) format
1197
1198 void fldcw(Address src);
1199
1200 void fldenv(Address src);
1201
1202 void fldlg2();
1203
1204 void fldln2();
1205
1206 void fldz();
1207
1208 void flog();
1209 void flog10();
1210
1211 void fmul(int i);
1212
1213 void fmul_d(Address src);
1214 void fmul_s(Address src);
1215
1216 void fmula(int i); // "alternate" fmul
1217
1218 void fmulp(int i = 1);
1219
1220 void fnsave(Address dst);
1221
1222 void fnstcw(Address src);
1223
1224 void fnstsw_ax();
1225
1226 void fprem();
1227 void fprem1();
1228
1229 void frstor(Address src);
1230
1231 void fsin();
1232
1233 void fsqrt();
1234
1235 void fst_d(Address adr);
1236 void fst_s(Address adr);
1237
1238 void fstp_d(Address adr);
1239 void fstp_d(int index);
1240 void fstp_s(Address adr);
1241 void fstp_x(Address adr); // extended-precision (80-bit) format
1242
1243 void fsub(int i);
1244 void fsub_d(Address src);
1245 void fsub_s(Address src);
1246
1247 void fsuba(int i); // "alternate" fsub
1248
1249 void fsubp(int i = 1);
1250
1251 void fsubr(int i);
1252 void fsubr_d(Address src);
1253 void fsubr_s(Address src);
1254
1255 void fsubra(int i); // "alternate" reversed fsub
1256
1257 void fsubrp(int i = 1);
1258
1259 void ftan();
1260
1261 void ftst();
1262
1263 void fucomi(int i = 1);
1264 void fucomip(int i = 1);
1265
1266 void fwait();
1267
1268 void fxch(int i = 1);
1269
1270 void fxrstor(Address src);
1271 void xrstor(Address src);
1272
1273 void fxsave(Address dst);
1274 void xsave(Address dst);
1275
1276 void fyl2x();
1277 void frndint();
1278 void f2xm1();
1279 void fldl2e();
1280
1281 void hlt();
1282
1283 void idivl(Register src);
1284 void divl(Register src); // Unsigned division
1285
1286 #ifdef _LP64
1287 void idivq(Register src);
1288 #endif
1289
1290 void imull(Register src);
1291 void imull(Register dst, Register src);
1292 void imull(Register dst, Register src, int value);
1293 void imull(Register dst, Address src);
1294
1295 #ifdef _LP64
1296 void imulq(Register dst, Register src);
1297 void imulq(Register dst, Register src, int value);
1298 void imulq(Register dst, Address src);
1299 #endif
1300
1301 // jcc is the generic conditional branch generator to run-
1302 // time routines, jcc is used for branches to labels. jcc
1303 // takes a branch opcode (cc) and a label (L) and generates
1304 // either a backward branch or a forward branch and links it
1305 // to the label fixup chain. Usage:
1306 //
1307 // Label L; // unbound label
1308 // jcc(cc, L); // forward branch to unbound label
1309 // bind(L); // bind label to the current pc
1310 // jcc(cc, L); // backward branch to bound label
1311 // bind(L); // illegal: a label may be bound only once
1312 //
1313 // Note: The same Label can be used for forward and backward branches
1314 // but it may be bound only once.
1315
1316 void jcc(Condition cc, Label& L, bool maybe_short = true);
1317
1318 // Conditional jump to a 8-bit offset to L.
1319 // WARNING: be very careful using this for forward jumps. If the label is
1320 // not bound within an 8-bit offset of this instruction, a run-time error
1321 // will occur.
1322 void jccb(Condition cc, Label& L);
1323
1324 void jmp(Address entry); // pc <- entry
1325
1326 // Label operations & relative jumps (PPUM Appendix D)
1327 void jmp(Label& L, bool maybe_short = true); // unconditional jump to L
1328
1329 void jmp(Register entry); // pc <- entry
1330
1331 // Unconditional 8-bit offset jump to L.
1332 // WARNING: be very careful using this for forward jumps. If the label is
1333 // not bound within an 8-bit offset of this instruction, a run-time error
1334 // will occur.
1335 void jmpb(Label& L);
1336
1337 void ldmxcsr( Address src );
1338
1339 void leal(Register dst, Address src);
1340
1341 void leaq(Register dst, Address src);
1342
1343 void lfence();
1344
1345 void lock();
1346
1347 void lzcntl(Register dst, Register src);
1348
1349 #ifdef _LP64
1350 void lzcntq(Register dst, Register src);
1351 #endif
1352
1353 enum Membar_mask_bits {
1354 StoreStore = 1 << 3,
1355 LoadStore = 1 << 2,
1356 StoreLoad = 1 << 1,
1357 LoadLoad = 1 << 0
1358 };
1359
1360 // Serializes memory and blows flags
1361 void membar(Membar_mask_bits order_constraint) {
1362 if (os::is_MP()) {
1363 // We only have to handle StoreLoad
1364 if (order_constraint & StoreLoad) {
1365 // All usable chips support "locked" instructions which suffice
1366 // as barriers, and are much faster than the alternative of
1367 // using cpuid instruction. We use here a locked add [esp-C],0.
1368 // This is conveniently otherwise a no-op except for blowing
1369 // flags, and introducing a false dependency on target memory
1370 // location. We can't do anything with flags, but we can avoid
1371 // memory dependencies in the current method by locked-adding
1372 // somewhere else on the stack. Doing [esp+C] will collide with
1373 // something on stack in current method, hence we go for [esp-C].
1374 // It is convenient since it is almost always in data cache, for
1375 // any small C. We need to step back from SP to avoid data
1376 // dependencies with other things on below SP (callee-saves, for
1377 // example). Without a clear way to figure out the minimal safe
1378 // distance from SP, it makes sense to step back the complete
1379 // cache line, as this will also avoid possible second-order effects
1380 // with locked ops against the cache line. Our choice of offset
1381 // is bounded by x86 operand encoding, which should stay within
1382 // [-128; +127] to have the 8-byte displacement encoding.
1383 //
1384 // Any change to this code may need to revisit other places in
1385 // the code where this idiom is used, in particular the
1386 // orderAccess code.
1387
1388 int offset = -VM_Version::L1_line_size();
1389 if (offset < -128) {
1390 offset = -128;
1391 }
1392
1393 lock();
1394 addl(Address(rsp, offset), 0);// Assert the lock# signal here
1395 }
1396 }
1397 }
1398
1399 void mfence();
1400
1401 // Moves
1402
1403 void mov64(Register dst, int64_t imm64);
1404
1405 void movb(Address dst, Register src);
1406 void movb(Address dst, int imm8);
1407 void movb(Register dst, Address src);
1408
1409 void movddup(XMMRegister dst, XMMRegister src);
1410
1411 void kmovbl(KRegister dst, Register src);
1412 void kmovbl(Register dst, KRegister src);
1413 void kmovwl(KRegister dst, Register src);
1414 void kmovwl(KRegister dst, Address src);
1415 void kmovwl(Register dst, KRegister src);
1416 void kmovdl(KRegister dst, Register src);
1417 void kmovdl(Register dst, KRegister src);
1418 void kmovql(KRegister dst, KRegister src);
1419 void kmovql(Address dst, KRegister src);
1420 void kmovql(KRegister dst, Address src);
1421 void kmovql(KRegister dst, Register src);
1422 void kmovql(Register dst, KRegister src);
1423
1424 void knotwl(KRegister dst, KRegister src);
1425
1426 void kortestbl(KRegister dst, KRegister src);
1427 void kortestwl(KRegister dst, KRegister src);
1428 void kortestdl(KRegister dst, KRegister src);
1429 void kortestql(KRegister dst, KRegister src);
1430
1431 void ktestq(KRegister src1, KRegister src2);
1432 void ktestd(KRegister src1, KRegister src2);
1433
1434 void ktestql(KRegister dst, KRegister src);
1435
1436 void movdl(XMMRegister dst, Register src);
1437 void movdl(Register dst, XMMRegister src);
1438 void movdl(XMMRegister dst, Address src);
1439 void movdl(Address dst, XMMRegister src);
1440
1441 // Move Double Quadword
1442 void movdq(XMMRegister dst, Register src);
1443 void movdq(Register dst, XMMRegister src);
1444
1445 // Move Aligned Double Quadword
1446 void movdqa(XMMRegister dst, XMMRegister src);
1447 void movdqa(XMMRegister dst, Address src);
1448
1449 // Move Unaligned Double Quadword
1450 void movdqu(Address dst, XMMRegister src);
1451 void movdqu(XMMRegister dst, Address src);
1452 void movdqu(XMMRegister dst, XMMRegister src);
1453
1454 // Move Unaligned 256bit Vector
1455 void vmovdqu(Address dst, XMMRegister src);
1456 void vmovdqu(XMMRegister dst, Address src);
1457 void vmovdqu(XMMRegister dst, XMMRegister src);
1458
1459 // Move Unaligned 512bit Vector
1460 void evmovdqub(Address dst, XMMRegister src, bool merge, int vector_len);
1461 void evmovdqub(XMMRegister dst, Address src, bool merge, int vector_len);
1462 void evmovdqub(XMMRegister dst, XMMRegister src, bool merge, int vector_len);
1463 void evmovdqub(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len);
1464 void evmovdquw(Address dst, XMMRegister src, bool merge, int vector_len);
1465 void evmovdquw(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len);
1466 void evmovdquw(XMMRegister dst, Address src, bool merge, int vector_len);
1467 void evmovdquw(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len);
1468 void evmovdqul(Address dst, XMMRegister src, int vector_len);
1469 void evmovdqul(XMMRegister dst, Address src, int vector_len);
1470 void evmovdqul(XMMRegister dst, XMMRegister src, int vector_len);
1471 void evmovdqul(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len);
1472 void evmovdqul(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len);
1473 void evmovdqul(XMMRegister dst, KRegister mask, XMMRegister src, bool merge, int vector_len);
1474 void evmovdquq(Address dst, XMMRegister src, int vector_len);
1475 void evmovdquq(XMMRegister dst, Address src, int vector_len);
1476 void evmovdquq(XMMRegister dst, XMMRegister src, int vector_len);
1477 void evmovdquq(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len);
1478 void evmovdquq(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len);
1479 void evmovdquq(XMMRegister dst, KRegister mask, XMMRegister src, bool merge, int vector_len);
1480
1481 // Move lower 64bit to high 64bit in 128bit register
1482 void movlhps(XMMRegister dst, XMMRegister src);
1483
1484 void movl(Register dst, int32_t imm32);
1485 void movl(Address dst, int32_t imm32);
1486 void movl(Register dst, Register src);
1487 void movl(Register dst, Address src);
1488 void movl(Address dst, Register src);
1489
1490 // These dummies prevent using movl from converting a zero (like NULL) into Register
1491 // by giving the compiler two choices it can't resolve
1492
1493 void movl(Address dst, void* junk);
1494 void movl(Register dst, void* junk);
1495
1496 #ifdef _LP64
1497 void movq(Register dst, Register src);
1498 void movq(Register dst, Address src);
1499 void movq(Address dst, Register src);
1500 #endif
1501
1502 void movq(Address dst, MMXRegister src );
1503 void movq(MMXRegister dst, Address src );
1504
1505 #ifdef _LP64
1506 // These dummies prevent using movq from converting a zero (like NULL) into Register
1507 // by giving the compiler two choices it can't resolve
1508
1509 void movq(Address dst, void* dummy);
1510 void movq(Register dst, void* dummy);
1511 #endif
1512
1513 // Move Quadword
1514 void movq(Address dst, XMMRegister src);
1515 void movq(XMMRegister dst, Address src);
1516 void movq(Register dst, XMMRegister src);
1517 void movq(XMMRegister dst, Register src);
1518
1519 void movsbl(Register dst, Address src);
1520 void movsbl(Register dst, Register src);
1521
1522 #ifdef _LP64
1523 void movsbq(Register dst, Address src);
1524 void movsbq(Register dst, Register src);
1525
1526 // Move signed 32bit immediate to 64bit extending sign
1527 void movslq(Address dst, int32_t imm64);
1528 void movslq(Register dst, int32_t imm64);
1529
1530 void movslq(Register dst, Address src);
1531 void movslq(Register dst, Register src);
1532 void movslq(Register dst, void* src); // Dummy declaration to cause NULL to be ambiguous
1533 #endif
1534
1535 void movswl(Register dst, Address src);
1536 void movswl(Register dst, Register src);
1537
1538 #ifdef _LP64
1539 void movswq(Register dst, Address src);
1540 void movswq(Register dst, Register src);
1541 #endif
1542
1543 void movw(Address dst, int imm16);
1544 void movw(Register dst, Address src);
1545 void movw(Address dst, Register src);
1546
1547 void movzbl(Register dst, Address src);
1548 void movzbl(Register dst, Register src);
1549
1550 #ifdef _LP64
1551 void movzbq(Register dst, Address src);
1552 void movzbq(Register dst, Register src);
1553 #endif
1554
1555 void movzwl(Register dst, Address src);
1556 void movzwl(Register dst, Register src);
1557
1558 #ifdef _LP64
1559 void movzwq(Register dst, Address src);
1560 void movzwq(Register dst, Register src);
1561 #endif
1562
1563 // Unsigned multiply with RAX destination register
1564 void mull(Address src);
1565 void mull(Register src);
1566
1567 #ifdef _LP64
1568 void mulq(Address src);
1569 void mulq(Register src);
1570 void mulxq(Register dst1, Register dst2, Register src);
1571 #endif
1572
1573 // Multiply Scalar Double-Precision Floating-Point Values
1574 void mulsd(XMMRegister dst, Address src);
1575 void mulsd(XMMRegister dst, XMMRegister src);
1576
1577 // Multiply Scalar Single-Precision Floating-Point Values
1578 void mulss(XMMRegister dst, Address src);
1579 void mulss(XMMRegister dst, XMMRegister src);
1580
1581 void negl(Register dst);
1582
1583 #ifdef _LP64
1584 void negq(Register dst);
1585 #endif
1586
1587 void nop(int i = 1);
1588
1589 void notl(Register dst);
1590
1591 #ifdef _LP64
1592 void notq(Register dst);
1593 #endif
1594
1595 void orl(Address dst, int32_t imm32);
1596 void orl(Register dst, int32_t imm32);
1597 void orl(Register dst, Address src);
1598 void orl(Register dst, Register src);
1599 void orl(Address dst, Register src);
1600
1601 void orq(Address dst, int32_t imm32);
1602 void orq(Register dst, int32_t imm32);
1603 void orq(Register dst, Address src);
1604 void orq(Register dst, Register src);
1605
1606 // Pack with unsigned saturation
1607 void packuswb(XMMRegister dst, XMMRegister src);
1608 void packuswb(XMMRegister dst, Address src);
1609 void vpackuswb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1610 void vpackusdw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1611
1612 // Permutations
1613 void vpermq(XMMRegister dst, XMMRegister src, int imm8, int vector_len);
1614 void vpermq(XMMRegister dst, XMMRegister src, int imm8);
1615 void vpermd(XMMRegister dst, XMMRegister nds, XMMRegister src);
1616 void vpermd(XMMRegister dst, XMMRegister nds, Address src);
1617 void vperm2i128(XMMRegister dst, XMMRegister nds, XMMRegister src, int imm8);
1618 void vperm2f128(XMMRegister dst, XMMRegister nds, XMMRegister src, int imm8);
1619
1620 void pause();
1621
1622 // Undefined Instruction
1623 void ud2();
1624
1625 // SSE4.2 string instructions
1626 void pcmpestri(XMMRegister xmm1, XMMRegister xmm2, int imm8);
1627 void pcmpestri(XMMRegister xmm1, Address src, int imm8);
1628
1629 void pcmpeqb(XMMRegister dst, XMMRegister src);
1630 void vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1631 void evpcmpeqb(KRegister kdst, XMMRegister nds, XMMRegister src, int vector_len);
1632 void evpcmpeqb(KRegister kdst, XMMRegister nds, Address src, int vector_len);
1633 void evpcmpeqb(KRegister kdst, KRegister mask, XMMRegister nds, Address src, int vector_len);
1634
1635 void vpcmpgtb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1636 void evpcmpgtb(KRegister kdst, XMMRegister nds, Address src, int vector_len);
1637 void evpcmpgtb(KRegister kdst, KRegister mask, XMMRegister nds, Address src, int vector_len);
1638
1639 void evpcmpuw(KRegister kdst, XMMRegister nds, XMMRegister src, ComparisonPredicate vcc, int vector_len);
1640 void evpcmpuw(KRegister kdst, KRegister mask, XMMRegister nds, XMMRegister src, ComparisonPredicate of, int vector_len);
1641 void evpcmpuw(KRegister kdst, XMMRegister nds, Address src, ComparisonPredicate vcc, int vector_len);
1642
1643 void pcmpeqw(XMMRegister dst, XMMRegister src);
1644 void vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1645 void evpcmpeqw(KRegister kdst, XMMRegister nds, XMMRegister src, int vector_len);
1646 void evpcmpeqw(KRegister kdst, XMMRegister nds, Address src, int vector_len);
1647
1648 void vpcmpgtw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1649
1650 void pcmpeqd(XMMRegister dst, XMMRegister src);
1651 void vpcmpeqd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1652 void evpcmpeqd(KRegister kdst, KRegister mask, XMMRegister nds, XMMRegister src, int vector_len);
1653 void evpcmpeqd(KRegister kdst, KRegister mask, XMMRegister nds, Address src, int vector_len);
1654
1655 void pcmpeqq(XMMRegister dst, XMMRegister src);
1656 void vpcmpeqq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1657 void evpcmpeqq(KRegister kdst, XMMRegister nds, XMMRegister src, int vector_len);
1658 void evpcmpeqq(KRegister kdst, XMMRegister nds, Address src, int vector_len);
1659
1660 void pcmpgtq(XMMRegister dst, XMMRegister src);
1661 void vpcmpgtq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1662
1663 void pmovmskb(Register dst, XMMRegister src);
1664 void vpmovmskb(Register dst, XMMRegister src);
1665
1666 // SSE 4.1 extract
1667 void pextrd(Register dst, XMMRegister src, int imm8);
1668 void pextrq(Register dst, XMMRegister src, int imm8);
1669 void pextrd(Address dst, XMMRegister src, int imm8);
1670 void pextrq(Address dst, XMMRegister src, int imm8);
1671 void pextrb(Register dst, XMMRegister src, int imm8);
1672 void pextrb(Address dst, XMMRegister src, int imm8);
1673 // SSE 2 extract
1674 void pextrw(Register dst, XMMRegister src, int imm8);
1675 void pextrw(Address dst, XMMRegister src, int imm8);
1676
1677 // SSE 4.1 insert
1678 void pinsrd(XMMRegister dst, Register src, int imm8);
1679 void pinsrq(XMMRegister dst, Register src, int imm8);
1680 void pinsrd(XMMRegister dst, Address src, int imm8);
1681 void pinsrq(XMMRegister dst, Address src, int imm8);
1682 void pinsrb(XMMRegister dst, Address src, int imm8);
1683 // SSE 2 insert
1684 void pinsrw(XMMRegister dst, Register src, int imm8);
1685 void pinsrw(XMMRegister dst, Address src, int imm8);
1686
1687 // Zero extend moves
1688 void pmovzxbw(XMMRegister dst, XMMRegister src);
1689 void pmovzxbw(XMMRegister dst, Address src);
1690 void vpmovzxbw( XMMRegister dst, Address src, int vector_len);
1691 void pmovzxdq(XMMRegister dst, XMMRegister src);
1692 void vpmovzxdq(XMMRegister dst, XMMRegister src, int vector_len);
1693 void vpmovzxbd(XMMRegister dst, XMMRegister src, int vector_len);
1694 void vpmovzxbq(XMMRegister dst, XMMRegister src, int vector_len);
1695 void evpmovzxbw(XMMRegister dst, KRegister mask, Address src, int vector_len);
1696
1697 // Sign extend moves
1698 void pmovsxbw(XMMRegister dst, XMMRegister src);
1699 void pmovsxbd(XMMRegister dst, XMMRegister src);
1700 void pmovsxbq(XMMRegister dst, XMMRegister src);
1701 void vpmovsxbd(XMMRegister dst, XMMRegister src, int vector_len);
1702 void vpmovsxbq(XMMRegister dst, XMMRegister src, int vector_len);
1703 void vpmovsxbw(XMMRegister dst, XMMRegister src, int vector_len);
1704
1705 void evpmovwb(Address dst, XMMRegister src, int vector_len);
1706 void evpmovwb(Address dst, KRegister mask, XMMRegister src, int vector_len);
1707
1708 #ifndef _LP64 // no 32bit push/pop on amd64
1709 void popl(Address dst);
1710 #endif
1711
1712 #ifdef _LP64
1713 void popq(Address dst);
1714 #endif
1715
1716 void popcntl(Register dst, Address src);
1717 void popcntl(Register dst, Register src);
1718
1719 void vpopcntd(XMMRegister dst, XMMRegister src, int vector_len);
1720
1721 #ifdef _LP64
1722 void popcntq(Register dst, Address src);
1723 void popcntq(Register dst, Register src);
1724 #endif
1725
1726 // Prefetches (SSE, SSE2, 3DNOW only)
1727
1728 void prefetchnta(Address src);
1729 void prefetchr(Address src);
1730 void prefetcht0(Address src);
1731 void prefetcht1(Address src);
1732 void prefetcht2(Address src);
1733 void prefetchw(Address src);
1734
1735 // Shuffle Bytes
1736 void pshufb(XMMRegister dst, XMMRegister src);
1737 void pshufb(XMMRegister dst, Address src);
1738 void vpshufb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1739
1740 // Shuffle Packed Doublewords
1741 void pshufd(XMMRegister dst, XMMRegister src, int mode);
1742 void pshufd(XMMRegister dst, Address src, int mode);
1743 void vpshufd(XMMRegister dst, XMMRegister src, int mode, int vector_len);
1744
1745 // Shuffle Packed Low Words
1746 void pshuflw(XMMRegister dst, XMMRegister src, int mode);
1747 void pshuflw(XMMRegister dst, Address src, int mode);
1748
1749 // Shift Right by bytes Logical DoubleQuadword Immediate
1750 void psrldq(XMMRegister dst, int shift);
1751 // Shift Left by bytes Logical DoubleQuadword Immediate
1752 void pslldq(XMMRegister dst, int shift);
1753
1754 // Logical Compare 128bit
1755 void ptest(XMMRegister dst, XMMRegister src);
1756 void ptest(XMMRegister dst, Address src);
1757 // Logical Compare 256bit
1758 void vptest(XMMRegister dst, XMMRegister src);
1759 void vptest(XMMRegister dst, Address src);
1760
1761 // Vector compare
1762 void vptest(XMMRegister dst, XMMRegister src, int vector_len);
1763
1764 // Interleave Low Bytes
1765 void punpcklbw(XMMRegister dst, XMMRegister src);
1766 void punpcklbw(XMMRegister dst, Address src);
1767
1768 // Interleave Low Doublewords
1769 void punpckldq(XMMRegister dst, XMMRegister src);
1770 void punpckldq(XMMRegister dst, Address src);
1771
1772 // Interleave Low Quadwords
1773 void punpcklqdq(XMMRegister dst, XMMRegister src);
1774
1775 #ifndef _LP64 // no 32bit push/pop on amd64
1776 void pushl(Address src);
1777 #endif
1778
1779 void pushq(Address src);
1780
1781 void rcll(Register dst, int imm8);
1782
1783 void rclq(Register dst, int imm8);
1784
1785 void rcrq(Register dst, int imm8);
1786
1787 void rcpps(XMMRegister dst, XMMRegister src);
1788
1789 void rcpss(XMMRegister dst, XMMRegister src);
1790
1791 void rdtsc();
1792
1793 void ret(int imm16);
1794
1795 #ifdef _LP64
1796 void rorq(Register dst, int imm8);
1797 void rorxq(Register dst, Register src, int imm8);
1798 void rorxd(Register dst, Register src, int imm8);
1799 #endif
1800
1801 void sahf();
1802
1803 void sarl(Register dst, int imm8);
1804 void sarl(Register dst);
1805
1806 void sarq(Register dst, int imm8);
1807 void sarq(Register dst);
1808
1809 void sbbl(Address dst, int32_t imm32);
1810 void sbbl(Register dst, int32_t imm32);
1811 void sbbl(Register dst, Address src);
1812 void sbbl(Register dst, Register src);
1813
1814 void sbbq(Address dst, int32_t imm32);
1815 void sbbq(Register dst, int32_t imm32);
1816 void sbbq(Register dst, Address src);
1817 void sbbq(Register dst, Register src);
1818
1819 void setb(Condition cc, Register dst);
1820
1821 void palignr(XMMRegister dst, XMMRegister src, int imm8);
1822 void vpalignr(XMMRegister dst, XMMRegister src1, XMMRegister src2, int imm8, int vector_len);
1823
1824 void pblendw(XMMRegister dst, XMMRegister src, int imm8);
1825
1826 void sha1rnds4(XMMRegister dst, XMMRegister src, int imm8);
1827 void sha1nexte(XMMRegister dst, XMMRegister src);
1828 void sha1msg1(XMMRegister dst, XMMRegister src);
1829 void sha1msg2(XMMRegister dst, XMMRegister src);
1830 // xmm0 is implicit additional source to the following instruction.
1831 void sha256rnds2(XMMRegister dst, XMMRegister src);
1832 void sha256msg1(XMMRegister dst, XMMRegister src);
1833 void sha256msg2(XMMRegister dst, XMMRegister src);
1834
1835 void shldl(Register dst, Register src);
1836 void shldl(Register dst, Register src, int8_t imm8);
1837
1838 void shll(Register dst, int imm8);
1839 void shll(Register dst);
1840
1841 void shlq(Register dst, int imm8);
1842 void shlq(Register dst);
1843
1844 void shrdl(Register dst, Register src);
1845
1846 void shrl(Register dst, int imm8);
1847 void shrl(Register dst);
1848
1849 void shrq(Register dst, int imm8);
1850 void shrq(Register dst);
1851
1852 void smovl(); // QQQ generic?
1853
1854 // Compute Square Root of Scalar Double-Precision Floating-Point Value
1855 void sqrtsd(XMMRegister dst, Address src);
1856 void sqrtsd(XMMRegister dst, XMMRegister src);
1857
1858 // Compute Square Root of Scalar Single-Precision Floating-Point Value
1859 void sqrtss(XMMRegister dst, Address src);
1860 void sqrtss(XMMRegister dst, XMMRegister src);
1861
1862 void std();
1863
1864 void stmxcsr( Address dst );
1865
1866 void subl(Address dst, int32_t imm32);
1867 void subl(Address dst, Register src);
1868 void subl(Register dst, int32_t imm32);
1869 void subl(Register dst, Address src);
1870 void subl(Register dst, Register src);
1871
1872 void subq(Address dst, int32_t imm32);
1873 void subq(Address dst, Register src);
1874 void subq(Register dst, int32_t imm32);
1875 void subq(Register dst, Address src);
1876 void subq(Register dst, Register src);
1877
1878 // Force generation of a 4 byte immediate value even if it fits into 8bit
1879 void subl_imm32(Register dst, int32_t imm32);
1880 void subq_imm32(Register dst, int32_t imm32);
1881
1882 // Subtract Scalar Double-Precision Floating-Point Values
1883 void subsd(XMMRegister dst, Address src);
1884 void subsd(XMMRegister dst, XMMRegister src);
1885
1886 // Subtract Scalar Single-Precision Floating-Point Values
1887 void subss(XMMRegister dst, Address src);
1888 void subss(XMMRegister dst, XMMRegister src);
1889
1890 void testb(Register dst, int imm8);
1891 void testb(Address dst, int imm8);
1892
1893 void testl(Register dst, int32_t imm32);
1894 void testl(Register dst, Register src);
1895 void testl(Register dst, Address src);
1896
1897 void testq(Register dst, int32_t imm32);
1898 void testq(Register dst, Register src);
1899
1900 // BMI - count trailing zeros
1901 void tzcntl(Register dst, Register src);
1902 void tzcntq(Register dst, Register src);
1903
1904 // Unordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS
1905 void ucomisd(XMMRegister dst, Address src);
1906 void ucomisd(XMMRegister dst, XMMRegister src);
1907
1908 // Unordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS
1909 void ucomiss(XMMRegister dst, Address src);
1910 void ucomiss(XMMRegister dst, XMMRegister src);
1911
1912 void xabort(int8_t imm8);
1913
1914 void xaddb(Address dst, Register src);
1915 void xaddw(Address dst, Register src);
1916 void xaddl(Address dst, Register src);
1917 void xaddq(Address dst, Register src);
1918
1919 void xbegin(Label& abort, relocInfo::relocType rtype = relocInfo::none);
1920
1921 void xchgb(Register reg, Address adr);
1922 void xchgw(Register reg, Address adr);
1923 void xchgl(Register reg, Address adr);
1924 void xchgl(Register dst, Register src);
1925
1926 void xchgq(Register reg, Address adr);
1927 void xchgq(Register dst, Register src);
1928
1929 void xend();
1930
1931 // Get Value of Extended Control Register
1932 void xgetbv();
1933
1934 void xorl(Register dst, int32_t imm32);
1935 void xorl(Register dst, Address src);
1936 void xorl(Register dst, Register src);
1937
1938 void xorb(Register dst, Address src);
1939
1940 void xorq(Register dst, Address src);
1941 void xorq(Register dst, Register src);
1942
1943 void set_byte_if_not_zero(Register dst); // sets reg to 1 if not zero, otherwise 0
1944
1945 // AVX 3-operands scalar instructions (encoded with VEX prefix)
1946
1947 void vaddsd(XMMRegister dst, XMMRegister nds, Address src);
1948 void vaddsd(XMMRegister dst, XMMRegister nds, XMMRegister src);
1949 void vaddss(XMMRegister dst, XMMRegister nds, Address src);
1950 void vaddss(XMMRegister dst, XMMRegister nds, XMMRegister src);
1951 void vdivsd(XMMRegister dst, XMMRegister nds, Address src);
1952 void vdivsd(XMMRegister dst, XMMRegister nds, XMMRegister src);
1953 void vdivss(XMMRegister dst, XMMRegister nds, Address src);
1954 void vdivss(XMMRegister dst, XMMRegister nds, XMMRegister src);
1955 void vfmadd231sd(XMMRegister dst, XMMRegister nds, XMMRegister src);
1956 void vfmadd231ss(XMMRegister dst, XMMRegister nds, XMMRegister src);
1957 void vmulsd(XMMRegister dst, XMMRegister nds, Address src);
1958 void vmulsd(XMMRegister dst, XMMRegister nds, XMMRegister src);
1959 void vmulss(XMMRegister dst, XMMRegister nds, Address src);
1960 void vmulss(XMMRegister dst, XMMRegister nds, XMMRegister src);
1961 void vsubsd(XMMRegister dst, XMMRegister nds, Address src);
1962 void vsubsd(XMMRegister dst, XMMRegister nds, XMMRegister src);
1963 void vsubss(XMMRegister dst, XMMRegister nds, Address src);
1964 void vsubss(XMMRegister dst, XMMRegister nds, XMMRegister src);
1965
1966 void shlxl(Register dst, Register src1, Register src2);
1967 void shlxq(Register dst, Register src1, Register src2);
1968
1969 //====================VECTOR ARITHMETIC=====================================
1970
1971 // Add Packed Floating-Point Values
1972 void addpd(XMMRegister dst, XMMRegister src);
1973 void addpd(XMMRegister dst, Address src);
1974 void addps(XMMRegister dst, XMMRegister src);
1975 void vaddpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1976 void vaddps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1977 void vaddpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1978 void vaddps(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1979
1980 // Subtract Packed Floating-Point Values
1981 void subpd(XMMRegister dst, XMMRegister src);
1982 void subps(XMMRegister dst, XMMRegister src);
1983 void vsubpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1984 void vsubps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1985 void vsubpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1986 void vsubps(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1987
1988 // Multiply Packed Floating-Point Values
1989 void mulpd(XMMRegister dst, XMMRegister src);
1990 void mulpd(XMMRegister dst, Address src);
1991 void mulps(XMMRegister dst, XMMRegister src);
1992 void vmulpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1993 void vmulps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1994 void vmulpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1995 void vmulps(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1996
1997 void vfmadd231pd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1998 void vfmadd231ps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1999 void vfmadd231pd(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2000 void vfmadd231ps(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2001
2002 // Divide Packed Floating-Point Values
2003 void divpd(XMMRegister dst, XMMRegister src);
2004 void divps(XMMRegister dst, XMMRegister src);
2005 void vdivpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2006 void vdivps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2007 void vdivpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2008 void vdivps(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2009
2010 // Sqrt Packed Floating-Point Values
2011 void vsqrtpd(XMMRegister dst, XMMRegister src, int vector_len);
2012 void vsqrtpd(XMMRegister dst, Address src, int vector_len);
2013 void vsqrtps(XMMRegister dst, XMMRegister src, int vector_len);
2014 void vsqrtps(XMMRegister dst, Address src, int vector_len);
2015
2016 // Bitwise Logical AND of Packed Floating-Point Values
2017 void andpd(XMMRegister dst, XMMRegister src);
2018 void andps(XMMRegister dst, XMMRegister src);
2019 void vandpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2020 void vandps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2021 void vandpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2022 void vandps(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2023
2024 void unpckhpd(XMMRegister dst, XMMRegister src);
2025 void unpcklpd(XMMRegister dst, XMMRegister src);
2026
2027 // Bitwise Logical XOR of Packed Floating-Point Values
2028 void xorpd(XMMRegister dst, XMMRegister src);
2029 void xorps(XMMRegister dst, XMMRegister src);
2030 void vxorpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2031 void vxorps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2032 void vxorpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2033 void vxorps(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2034
2035 // Add horizontal packed integers
2036 void vphaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2037 void vphaddd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2038 void phaddw(XMMRegister dst, XMMRegister src);
2039 void phaddd(XMMRegister dst, XMMRegister src);
2040
2041 // Add packed integers
2042 void paddb(XMMRegister dst, XMMRegister src);
2043 void paddw(XMMRegister dst, XMMRegister src);
2044 void paddd(XMMRegister dst, XMMRegister src);
2045 void paddd(XMMRegister dst, Address src);
2046 void paddq(XMMRegister dst, XMMRegister src);
2047 void vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2048 void vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2049 void vpaddd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2050 void vpaddq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2051 void vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2052 void vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2053 void vpaddd(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2054 void vpaddq(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2055
2056 // Sub packed integers
2057 void psubb(XMMRegister dst, XMMRegister src);
2058 void psubw(XMMRegister dst, XMMRegister src);
2059 void psubd(XMMRegister dst, XMMRegister src);
2060 void psubq(XMMRegister dst, XMMRegister src);
2061 void vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2062 void vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2063 void vpsubd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2064 void vpsubq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2065 void vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2066 void vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2067 void vpsubd(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2068 void vpsubq(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2069
2070 // Multiply packed integers (only shorts and ints)
2071 void pmullw(XMMRegister dst, XMMRegister src);
2072 void pmulld(XMMRegister dst, XMMRegister src);
2073 void pmuludq(XMMRegister dst, XMMRegister src);
2074 void vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2075 void vpmulld(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2076 void vpmullq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2077 void vpmuludq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2078 void vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2079 void vpmulld(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2080 void vpmullq(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2081
2082 // Minimum of packed integers
2083 void pminsb(XMMRegister dst, XMMRegister src);
2084 void vpminsb(XMMRegister dst, XMMRegister src1, XMMRegister src2, int vector_len);
2085 void pminsw(XMMRegister dst, XMMRegister src);
2086 void vpminsw(XMMRegister dst, XMMRegister src1, XMMRegister src2, int vector_len);
2087 void pminsd(XMMRegister dst, XMMRegister src);
2088 void vpminsd(XMMRegister dst, XMMRegister src1, XMMRegister src2, int vector_len);
2089 void vpminsq(XMMRegister dst, XMMRegister src1, XMMRegister src2, int vector_len);
2090 void minps(XMMRegister dst, XMMRegister src);
2091 void vminps(XMMRegister dst, XMMRegister src1, XMMRegister src2, int vector_len);
2092 void minpd(XMMRegister dst, XMMRegister src);
2093 void vminpd(XMMRegister dst, XMMRegister src1, XMMRegister src2, int vector_len);
2094
2095 // Maximum of packed integers
2096 void pmaxsb(XMMRegister dst, XMMRegister src);
2097 void vpmaxsb(XMMRegister dst, XMMRegister src1, XMMRegister src2, int vector_len);
2098 void pmaxsw(XMMRegister dst, XMMRegister src);
2099 void vpmaxsw(XMMRegister dst, XMMRegister src1, XMMRegister src2, int vector_len);
2100 void pmaxsd(XMMRegister dst, XMMRegister src);
2101 void vpmaxsd(XMMRegister dst, XMMRegister src1, XMMRegister src2, int vector_len);
2102 void vpmaxsq(XMMRegister dst, XMMRegister src1, XMMRegister src2, int vector_len);
2103 void maxps(XMMRegister dst, XMMRegister src);
2104 void vmaxps(XMMRegister dst, XMMRegister src1, XMMRegister src2, int vector_len);
2105 void maxpd(XMMRegister dst, XMMRegister src);
2106 void vmaxpd(XMMRegister dst, XMMRegister src1, XMMRegister src2, int vector_len);
2107
2108 // Shift left packed integers
2109 void psllw(XMMRegister dst, int shift);
2110 void pslld(XMMRegister dst, int shift);
2111 void psllq(XMMRegister dst, int shift);
2112 void psllw(XMMRegister dst, XMMRegister shift);
2113 void pslld(XMMRegister dst, XMMRegister shift);
2114 void psllq(XMMRegister dst, XMMRegister shift);
2115 void vpsllw(XMMRegister dst, XMMRegister src, int shift, int vector_len);
2116 void vpslld(XMMRegister dst, XMMRegister src, int shift, int vector_len);
2117 void vpsllq(XMMRegister dst, XMMRegister src, int shift, int vector_len);
2118 void vpsllw(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len);
2119 void vpslld(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len);
2120 void vpsllq(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len);
2121
2122 // Logical shift right packed integers
2123 void psrlw(XMMRegister dst, int shift);
2124 void psrld(XMMRegister dst, int shift);
2125 void psrlq(XMMRegister dst, int shift);
2126 void psrlw(XMMRegister dst, XMMRegister shift);
2127 void psrld(XMMRegister dst, XMMRegister shift);
2128 void psrlq(XMMRegister dst, XMMRegister shift);
2129 void vpsrlw(XMMRegister dst, XMMRegister src, int shift, int vector_len);
2130 void vpsrld(XMMRegister dst, XMMRegister src, int shift, int vector_len);
2131 void vpsrlq(XMMRegister dst, XMMRegister src, int shift, int vector_len);
2132 void vpsrlw(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len);
2133 void vpsrld(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len);
2134 void vpsrlq(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len);
2135
2136 // Arithmetic shift right packed integers (only shorts and ints, no instructions for longs)
2137 void psraw(XMMRegister dst, int shift);
2138 void psrad(XMMRegister dst, int shift);
2139 void psraw(XMMRegister dst, XMMRegister shift);
2140 void psrad(XMMRegister dst, XMMRegister shift);
2141 void vpsraw(XMMRegister dst, XMMRegister src, int shift, int vector_len);
2142 void vpsrad(XMMRegister dst, XMMRegister src, int shift, int vector_len);
2143 void vpsraw(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len);
2144 void vpsrad(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len);
2145
2146 // And packed integers
2147 void pand(XMMRegister dst, XMMRegister src);
2148 void vpand(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2149 void vpand(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2150 void evpandd(XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len);
2151 void vpandq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2152
2153 // Andn packed integers
2154 void pandn(XMMRegister dst, XMMRegister src);
2155
2156 // Or packed integers
2157 void por(XMMRegister dst, XMMRegister src);
2158 void vpor(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2159 void vpor(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2160 void vporq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2161
2162 void evpord(XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len);
2163 void evpord(XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len);
2164
2165 // Xor packed integers
2166 void pxor(XMMRegister dst, XMMRegister src);
2167 void vpxor(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2168 void vpxor(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
2169 void vpxorq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2170 void evpxord(XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len);
2171
2172 // vinserti forms
2173 void vinserti128(XMMRegister dst, XMMRegister nds, XMMRegister src, uint8_t imm8);
2174 void vinserti128(XMMRegister dst, XMMRegister nds, Address src, uint8_t imm8);
2175 void vinserti32x4(XMMRegister dst, XMMRegister nds, XMMRegister src, uint8_t imm8);
2176 void vinserti32x4(XMMRegister dst, XMMRegister nds, Address src, uint8_t imm8);
2177 void vinserti64x4(XMMRegister dst, XMMRegister nds, XMMRegister src, uint8_t imm8);
2178
2179 // vinsertf forms
2180 void vinsertf128(XMMRegister dst, XMMRegister nds, XMMRegister src, uint8_t imm8);
2181 void vinsertf128(XMMRegister dst, XMMRegister nds, Address src, uint8_t imm8);
2182 void vinsertf32x4(XMMRegister dst, XMMRegister nds, XMMRegister src, uint8_t imm8);
2183 void vinsertf32x4(XMMRegister dst, XMMRegister nds, Address src, uint8_t imm8);
2184 void vinsertf64x4(XMMRegister dst, XMMRegister nds, XMMRegister src, uint8_t imm8);
2185 void vinsertf64x4(XMMRegister dst, XMMRegister nds, Address src, uint8_t imm8);
2186
2187 // vextracti forms
2188 void vextracti128(XMMRegister dst, XMMRegister src, uint8_t imm8);
2189 void vextracti128(Address dst, XMMRegister src, uint8_t imm8);
2190 void vextracti32x4(XMMRegister dst, XMMRegister src, uint8_t imm8);
2191 void vextracti32x4(Address dst, XMMRegister src, uint8_t imm8);
2192 void vextracti64x2(XMMRegister dst, XMMRegister src, uint8_t imm8);
2193 void vextracti64x4(XMMRegister dst, XMMRegister src, uint8_t imm8);
2194
2195 // vextractf forms
2196 void vextractf128(XMMRegister dst, XMMRegister src, uint8_t imm8);
2197 void vextractf128(Address dst, XMMRegister src, uint8_t imm8);
2198 void vextractf32x4(XMMRegister dst, XMMRegister src, uint8_t imm8);
2199 void vextractf32x4(Address dst, XMMRegister src, uint8_t imm8);
2200 void vextractf64x2(XMMRegister dst, XMMRegister src, uint8_t imm8);
2201 void vextractf64x4(XMMRegister dst, XMMRegister src, uint8_t imm8);
2202 void vextractf64x4(Address dst, XMMRegister src, uint8_t imm8);
2203
2204 // legacy xmm sourced word/dword replicate
2205 void vpbroadcastw(XMMRegister dst, XMMRegister src);
2206 void vpbroadcastd(XMMRegister dst, XMMRegister src);
2207
2208 // xmm/mem sourced byte/word/dword/qword replicate
2209 void evpbroadcastb(XMMRegister dst, XMMRegister src, int vector_len);
2210 void evpbroadcastb(XMMRegister dst, Address src, int vector_len);
2211 void evpbroadcastw(XMMRegister dst, XMMRegister src, int vector_len);
2212 void evpbroadcastw(XMMRegister dst, Address src, int vector_len);
2213 void evpbroadcastd(XMMRegister dst, XMMRegister src, int vector_len);
2214 void evpbroadcastd(XMMRegister dst, Address src, int vector_len);
2215 void evpbroadcastq(XMMRegister dst, XMMRegister src, int vector_len);
2216 void evpbroadcastq(XMMRegister dst, Address src, int vector_len);
2217
2218 // scalar single/double precision replicate
2219 void evpbroadcastss(XMMRegister dst, XMMRegister src, int vector_len);
2220 void evpbroadcastss(XMMRegister dst, Address src, int vector_len);
2221 void evpbroadcastsd(XMMRegister dst, XMMRegister src, int vector_len);
2222 void evpbroadcastsd(XMMRegister dst, Address src, int vector_len);
2223
2224 // gpr sourced byte/word/dword/qword replicate
2225 void evpbroadcastb(XMMRegister dst, Register src, int vector_len);
2226 void evpbroadcastw(XMMRegister dst, Register src, int vector_len);
2227 void evpbroadcastd(XMMRegister dst, Register src, int vector_len);
2228 void evpbroadcastq(XMMRegister dst, Register src, int vector_len);
2229
2230 // Carry-Less Multiplication Quadword
2231 void pclmulqdq(XMMRegister dst, XMMRegister src, int mask);
2232 void vpclmulqdq(XMMRegister dst, XMMRegister nds, XMMRegister src, int mask);
2233
2234 // AVX instruction which is used to clear upper 128 bits of YMM registers and
2235 // to avoid transaction penalty between AVX and SSE states. There is no
2236 // penalty if legacy SSE instructions are encoded using VEX prefix because
2237 // they always clear upper 128 bits. It should be used before calling
2238 // runtime code and native libraries.
2239 void vzeroupper();
2240
2241 // Vector double compares
2242 void vcmppd(XMMRegister dst, XMMRegister nds, XMMRegister src, int cop, int vector_len);
2243 void evcmppd(KRegister kdst, KRegister mask, XMMRegister nds, XMMRegister src,
2244 ComparisonPredicateFP comparison, int vector_len);
2245
2246 // Vector float compares
2247 void vcmpps(XMMRegister dst, XMMRegister nds, XMMRegister src, int comparison, int vector_len);
2248 void evcmpps(KRegister kdst, KRegister mask, XMMRegister nds, XMMRegister src,
2249 ComparisonPredicateFP comparison, int vector_len);
2250
2251 // Vector integer compares
2252 void vpcmpgtd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
2253 void evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, XMMRegister src,
2254 int comparison, int vector_len);
2255 void evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, Address src,
2256 int comparison, int vector_len);
2257
2258 // Vector long compares
2259 void evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, XMMRegister src,
2260 int comparison, int vector_len);
2261 void evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, Address src,
2262 int comparison, int vector_len);
2263
2264 // Vector byte compares
2265 void evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, XMMRegister src,
2266 int comparison, int vector_len);
2267 void evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, Address src,
2268 int comparison, int vector_len);
2269
2270 // Vector short compares
2271 void evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, XMMRegister src,
2272 int comparison, int vector_len);
2273 void evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, Address src,
2274 int comparison, int vector_len);
2275
2276 // Vector blends
2277 void blendvps(XMMRegister dst, XMMRegister src);
2278 void blendvpd(XMMRegister dst, XMMRegister src);
2279 void pblendvb(XMMRegister dst, XMMRegister src);
2280 void vblendvps(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len);
2281 void vblendvpd(XMMRegister dst, XMMRegister nds, XMMRegister src1, XMMRegister src2, int vector_len);
2282 void vpblendvb(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len);
2283 void vpblendd(XMMRegister dst, XMMRegister nds, XMMRegister src, int imm8, int vector_len);
2284 void evblendmpd(XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len);
2285 void evblendmps(XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len);
2286 void evpblendmb(XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len);
2287 void evpblendmw(XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len);
2288 void evpblendmd(XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len);
2289 void evpblendmq(XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len);
2290 protected:
2291 // Next instructions require address alignment 16 bytes SSE mode.
2292 // They should be called only from corresponding MacroAssembler instructions.
2293 void andpd(XMMRegister dst, Address src);
2294 void andps(XMMRegister dst, Address src);
2295 void xorpd(XMMRegister dst, Address src);
2296 void xorps(XMMRegister dst, Address src);
2297
2298 };
2299
2300 // The Intel x86/Amd64 Assembler attributes: All fields enclosed here are to guide encoding level decisions.
2301 // Specific set functions are for specialized use, else defaults or whatever was supplied to object construction
2302 // are applied.
2303 class InstructionAttr {
2304 public:
2305 InstructionAttr(
2306 int vector_len, // The length of vector to be applied in encoding - for both AVX and EVEX
2307 bool rex_vex_w, // Width of data: if 32-bits or less, false, else if 64-bit or specially defined, true
2308 bool legacy_mode, // Details if either this instruction is conditionally encoded to AVX or earlier if true else possibly EVEX
2309 bool no_reg_mask, // when true, k0 is used when EVEX encoding is chosen, else k1 is used under the same condition
2310 bool uses_vl) // This instruction may have legacy constraints based on vector length for EVEX
2311 :
2312 _avx_vector_len(vector_len),
2313 _rex_vex_w(rex_vex_w),
2314 _rex_vex_w_reverted(false),
2315 _legacy_mode(legacy_mode),
2316 _no_reg_mask(no_reg_mask),
2317 _uses_vl(uses_vl),
2318 _tuple_type(Assembler::EVEX_ETUP),
2319 _input_size_in_bits(Assembler::EVEX_NObit),
2320 _is_evex_instruction(false),
2321 _evex_encoding(0),
2322 _is_clear_context(true),
2323 _is_extended_context(false),
2324 _current_assembler(NULL),
2325 _embedded_opmask_register_specifier(1) { // hard code k1, it will be initialized for now
2326 if (UseAVX < 3) _legacy_mode = true;
2327 }
2328
2329 ~InstructionAttr() {
2330 if (_current_assembler != NULL) {
2331 _current_assembler->clear_attributes();
2332 }
2333 _current_assembler = NULL;
2334 }
2335
2336 private:
2337 int _avx_vector_len;
2338 bool _rex_vex_w;
2339 bool _rex_vex_w_reverted;
2340 bool _legacy_mode;
2341 bool _no_reg_mask;
2342 bool _uses_vl;
2343 int _tuple_type;
2344 int _input_size_in_bits;
2345 bool _is_evex_instruction;
2346 int _evex_encoding;
2347 bool _is_clear_context;
2348 bool _is_extended_context;
2349 int _embedded_opmask_register_specifier;
2350
2351 Assembler *_current_assembler;
2352
2353 public:
2354 // query functions for field accessors
2355 int get_vector_len(void) const { return _avx_vector_len; }
2356 bool is_rex_vex_w(void) const { return _rex_vex_w; }
2357 bool is_rex_vex_w_reverted(void) { return _rex_vex_w_reverted; }
2358 bool is_legacy_mode(void) const { return _legacy_mode; }
2359 bool is_no_reg_mask(void) const { return _no_reg_mask; }
2360 bool uses_vl(void) const { return _uses_vl; }
2361 int get_tuple_type(void) const { return _tuple_type; }
2362 int get_input_size(void) const { return _input_size_in_bits; }
2363 int is_evex_instruction(void) const { return _is_evex_instruction; }
2364 int get_evex_encoding(void) const { return _evex_encoding; }
2365 bool is_clear_context(void) const { return _is_clear_context; }
2366 bool is_extended_context(void) const { return _is_extended_context; }
2367 int get_embedded_opmask_register_specifier(void) const { return _embedded_opmask_register_specifier; }
2368
2369 // Set the vector len manually
2370 void set_vector_len(int vector_len) { _avx_vector_len = vector_len; }
2371
2372 // Set revert rex_vex_w for avx encoding
2373 void set_rex_vex_w_reverted(void) { _rex_vex_w_reverted = true; }
2374
2375 // Set rex_vex_w based on state
2376 void set_rex_vex_w(bool state) { _rex_vex_w = state; }
2377
2378 // Set the instruction to be encoded in AVX mode
2379 void set_is_legacy_mode(void) { _legacy_mode = true; }
2380
2381 // Set the current instuction to be encoded as an EVEX instuction
2382 void set_is_evex_instruction(void) { _is_evex_instruction = true; }
2383
2384 // Internal encoding data used in compressed immediate offset programming
2385 void set_evex_encoding(int value) { _evex_encoding = value; }
2386
2387 // When the Evex.Z field is set (true), it is used to clear all non directed XMM/YMM/ZMM components.
2388 // This method unsets it so that merge semantics are used instead.
2389 void reset_is_clear_context(void) { _is_clear_context = false; }
2390
2391 // Map back to current asembler so that we can manage object level assocation
2392 void set_current_assembler(Assembler *current_assembler) { _current_assembler = current_assembler; }
2393
2394 // Address modifiers used for compressed displacement calculation
2395 void set_address_attributes(int tuple_type, int input_size_in_bits) {
2396 if (VM_Version::supports_evex()) {
2397 _tuple_type = tuple_type;
2398 _input_size_in_bits = input_size_in_bits;
2399 }
2400 }
2401
2402 // Set embedded opmask register specifier.
2403 void set_embedded_opmask_register_specifier(KRegister mask) {
2404 _embedded_opmask_register_specifier = (*mask).encoding() & 0x7;
2405 }
2406
2407 };
2408
2409 #endif // CPU_X86_VM_ASSEMBLER_X86_HPP