1 /*
2 * MMX/3DNow!/SSE/SSE2/SSE3/SSSE3/SSE4/PNI support
3 *
4 * Copyright (c) 2005 Fabrice Bellard
5 * Copyright (c) 2008 Intel Corporation <andrew.zaborowski@intel.com>
6 *
7 * This library is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * This library is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
19 */
20
21 #include "crypto/aes.h"
22 #include "crypto/aes-round.h"
23 #include "crypto/clmul.h"
24
25 #if SHIFT == 0
26 #define Reg MMXReg
27 #define XMM_ONLY(...)
28 #define B(n) MMX_B(n)
29 #define W(n) MMX_W(n)
30 #define L(n) MMX_L(n)
31 #define Q(n) MMX_Q(n)
32 #define SUFFIX _mmx
33 #else
34 #define Reg ZMMReg
35 #define XMM_ONLY(...) __VA_ARGS__
36 #define B(n) ZMM_B(n)
37 #define W(n) ZMM_W(n)
38 #define L(n) ZMM_L(n)
39 #define Q(n) ZMM_Q(n)
40 #if SHIFT == 1
41 #define SUFFIX _xmm
42 #else
43 #define SUFFIX _ymm
44 #endif
45 #endif
46
47 #define LANE_WIDTH (SHIFT ? 16 : 8)
48 #define PACK_WIDTH (LANE_WIDTH / 2)
49
50 #if SHIFT == 0
51 #define FPSRL(x, c) ((x) >> shift)
52 #define FPSRAW(x, c) ((int16_t)(x) >> shift)
53 #define FPSRAL(x, c) ((int32_t)(x) >> shift)
54 #define FPSLL(x, c) ((x) << shift)
55 #endif
56
glue(helper_psrlw,SUFFIX)57 void glue(helper_psrlw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
58 {
59 int shift;
60 if (c->Q(0) > 15) {
61 for (int i = 0; i < 1 << SHIFT; i++) {
62 d->Q(i) = 0;
63 }
64 } else {
65 shift = c->B(0);
66 for (int i = 0; i < 4 << SHIFT; i++) {
67 d->W(i) = FPSRL(s->W(i), shift);
68 }
69 }
70 }
71
glue(helper_psllw,SUFFIX)72 void glue(helper_psllw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
73 {
74 int shift;
75 if (c->Q(0) > 15) {
76 for (int i = 0; i < 1 << SHIFT; i++) {
77 d->Q(i) = 0;
78 }
79 } else {
80 shift = c->B(0);
81 for (int i = 0; i < 4 << SHIFT; i++) {
82 d->W(i) = FPSLL(s->W(i), shift);
83 }
84 }
85 }
86
glue(helper_psraw,SUFFIX)87 void glue(helper_psraw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
88 {
89 int shift;
90 if (c->Q(0) > 15) {
91 shift = 15;
92 } else {
93 shift = c->B(0);
94 }
95 for (int i = 0; i < 4 << SHIFT; i++) {
96 d->W(i) = FPSRAW(s->W(i), shift);
97 }
98 }
99
glue(helper_psrld,SUFFIX)100 void glue(helper_psrld, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
101 {
102 int shift;
103 if (c->Q(0) > 31) {
104 for (int i = 0; i < 1 << SHIFT; i++) {
105 d->Q(i) = 0;
106 }
107 } else {
108 shift = c->B(0);
109 for (int i = 0; i < 2 << SHIFT; i++) {
110 d->L(i) = FPSRL(s->L(i), shift);
111 }
112 }
113 }
114
glue(helper_pslld,SUFFIX)115 void glue(helper_pslld, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
116 {
117 int shift;
118 if (c->Q(0) > 31) {
119 for (int i = 0; i < 1 << SHIFT; i++) {
120 d->Q(i) = 0;
121 }
122 } else {
123 shift = c->B(0);
124 for (int i = 0; i < 2 << SHIFT; i++) {
125 d->L(i) = FPSLL(s->L(i), shift);
126 }
127 }
128 }
129
glue(helper_psrad,SUFFIX)130 void glue(helper_psrad, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
131 {
132 int shift;
133 if (c->Q(0) > 31) {
134 shift = 31;
135 } else {
136 shift = c->B(0);
137 }
138 for (int i = 0; i < 2 << SHIFT; i++) {
139 d->L(i) = FPSRAL(s->L(i), shift);
140 }
141 }
142
glue(helper_psrlq,SUFFIX)143 void glue(helper_psrlq, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
144 {
145 int shift;
146 if (c->Q(0) > 63) {
147 for (int i = 0; i < 1 << SHIFT; i++) {
148 d->Q(i) = 0;
149 }
150 } else {
151 shift = c->B(0);
152 for (int i = 0; i < 1 << SHIFT; i++) {
153 d->Q(i) = FPSRL(s->Q(i), shift);
154 }
155 }
156 }
157
glue(helper_psllq,SUFFIX)158 void glue(helper_psllq, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
159 {
160 int shift;
161 if (c->Q(0) > 63) {
162 for (int i = 0; i < 1 << SHIFT; i++) {
163 d->Q(i) = 0;
164 }
165 } else {
166 shift = c->B(0);
167 for (int i = 0; i < 1 << SHIFT; i++) {
168 d->Q(i) = FPSLL(s->Q(i), shift);
169 }
170 }
171 }
172
173 #if SHIFT >= 1
glue(helper_psrldq,SUFFIX)174 void glue(helper_psrldq, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
175 {
176 int shift, i, j;
177
178 shift = c->L(0);
179 if (shift > 16) {
180 shift = 16;
181 }
182 for (j = 0; j < 8 << SHIFT; j += LANE_WIDTH) {
183 for (i = 0; i < 16 - shift; i++) {
184 d->B(j + i) = s->B(j + i + shift);
185 }
186 for (i = 16 - shift; i < 16; i++) {
187 d->B(j + i) = 0;
188 }
189 }
190 }
191
glue(helper_pslldq,SUFFIX)192 void glue(helper_pslldq, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
193 {
194 int shift, i, j;
195
196 shift = c->L(0);
197 if (shift > 16) {
198 shift = 16;
199 }
200 for (j = 0; j < 8 << SHIFT; j += LANE_WIDTH) {
201 for (i = 15; i >= shift; i--) {
202 d->B(j + i) = s->B(j + i - shift);
203 }
204 for (i = 0; i < shift; i++) {
205 d->B(j + i) = 0;
206 }
207 }
208 }
209 #endif
210
211 #define SSE_HELPER_1(name, elem, num, F) \
212 void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s) \
213 { \
214 int n = num; \
215 for (int i = 0; i < n; i++) { \
216 d->elem(i) = F(s->elem(i)); \
217 } \
218 }
219
220 #define SSE_HELPER_2(name, elem, num, F) \
221 void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s) \
222 { \
223 int n = num; \
224 for (int i = 0; i < n; i++) { \
225 d->elem(i) = F(v->elem(i), s->elem(i)); \
226 } \
227 }
228
229 #define SSE_HELPER_B(name, F) \
230 SSE_HELPER_2(name, B, 8 << SHIFT, F)
231
232 #define SSE_HELPER_W(name, F) \
233 SSE_HELPER_2(name, W, 4 << SHIFT, F)
234
235 #define SSE_HELPER_L(name, F) \
236 SSE_HELPER_2(name, L, 2 << SHIFT, F)
237
238 #define SSE_HELPER_Q(name, F) \
239 SSE_HELPER_2(name, Q, 1 << SHIFT, F)
240
241 #if SHIFT == 0
satub(int x)242 static inline int satub(int x)
243 {
244 if (x < 0) {
245 return 0;
246 } else if (x > 255) {
247 return 255;
248 } else {
249 return x;
250 }
251 }
252
satuw(int x)253 static inline int satuw(int x)
254 {
255 if (x < 0) {
256 return 0;
257 } else if (x > 65535) {
258 return 65535;
259 } else {
260 return x;
261 }
262 }
263
satsb(int x)264 static inline int satsb(int x)
265 {
266 if (x < -128) {
267 return -128;
268 } else if (x > 127) {
269 return 127;
270 } else {
271 return x;
272 }
273 }
274
satsw(int x)275 static inline int satsw(int x)
276 {
277 if (x < -32768) {
278 return -32768;
279 } else if (x > 32767) {
280 return 32767;
281 } else {
282 return x;
283 }
284 }
285
286 #define FADD(a, b) ((a) + (b))
287 #define FADDUB(a, b) satub((a) + (b))
288 #define FADDUW(a, b) satuw((a) + (b))
289 #define FADDSB(a, b) satsb((int8_t)(a) + (int8_t)(b))
290 #define FADDSW(a, b) satsw((int16_t)(a) + (int16_t)(b))
291
292 #define FSUB(a, b) ((a) - (b))
293 #define FSUBUB(a, b) satub((a) - (b))
294 #define FSUBUW(a, b) satuw((a) - (b))
295 #define FSUBSB(a, b) satsb((int8_t)(a) - (int8_t)(b))
296 #define FSUBSW(a, b) satsw((int16_t)(a) - (int16_t)(b))
297 #define FMINUB(a, b) ((a) < (b)) ? (a) : (b)
298 #define FMINSW(a, b) ((int16_t)(a) < (int16_t)(b)) ? (a) : (b)
299 #define FMAXUB(a, b) ((a) > (b)) ? (a) : (b)
300 #define FMAXSW(a, b) ((int16_t)(a) > (int16_t)(b)) ? (a) : (b)
301
302 #define FMULHRW(a, b) (((int16_t)(a) * (int16_t)(b) + 0x8000) >> 16)
303 #define FMULHUW(a, b) ((a) * (b) >> 16)
304 #define FMULHW(a, b) ((int16_t)(a) * (int16_t)(b) >> 16)
305
306 #define FAVG(a, b) (((a) + (b) + 1) >> 1)
307 #endif
308
SSE_HELPER_W(helper_pmulhuw,FMULHUW)309 SSE_HELPER_W(helper_pmulhuw, FMULHUW)
310 SSE_HELPER_W(helper_pmulhw, FMULHW)
311
312 #if SHIFT == 0
313 void glue(helper_pmulhrw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
314 {
315 d->W(0) = FMULHRW(d->W(0), s->W(0));
316 d->W(1) = FMULHRW(d->W(1), s->W(1));
317 d->W(2) = FMULHRW(d->W(2), s->W(2));
318 d->W(3) = FMULHRW(d->W(3), s->W(3));
319 }
320 #endif
321
SSE_HELPER_B(helper_pavgb,FAVG)322 SSE_HELPER_B(helper_pavgb, FAVG)
323 SSE_HELPER_W(helper_pavgw, FAVG)
324
325 void glue(helper_pmuludq, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
326 {
327 int i;
328
329 for (i = 0; i < (1 << SHIFT); i++) {
330 d->Q(i) = (uint64_t)s->L(i * 2) * (uint64_t)v->L(i * 2);
331 }
332 }
333
glue(helper_pmaddwd,SUFFIX)334 void glue(helper_pmaddwd, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
335 {
336 int i;
337
338 for (i = 0; i < (2 << SHIFT); i++) {
339 d->L(i) = (int16_t)s->W(2 * i) * (int16_t)v->W(2 * i) +
340 (int16_t)s->W(2 * i + 1) * (int16_t)v->W(2 * i + 1);
341 }
342 }
343
344 #if SHIFT == 0
abs1(int a)345 static inline int abs1(int a)
346 {
347 if (a < 0) {
348 return -a;
349 } else {
350 return a;
351 }
352 }
353 #endif
glue(helper_psadbw,SUFFIX)354 void glue(helper_psadbw, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
355 {
356 int i;
357
358 for (i = 0; i < (1 << SHIFT); i++) {
359 unsigned int val = 0;
360 val += abs1(v->B(8 * i + 0) - s->B(8 * i + 0));
361 val += abs1(v->B(8 * i + 1) - s->B(8 * i + 1));
362 val += abs1(v->B(8 * i + 2) - s->B(8 * i + 2));
363 val += abs1(v->B(8 * i + 3) - s->B(8 * i + 3));
364 val += abs1(v->B(8 * i + 4) - s->B(8 * i + 4));
365 val += abs1(v->B(8 * i + 5) - s->B(8 * i + 5));
366 val += abs1(v->B(8 * i + 6) - s->B(8 * i + 6));
367 val += abs1(v->B(8 * i + 7) - s->B(8 * i + 7));
368 d->Q(i) = val;
369 }
370 }
371
372 #if SHIFT < 2
glue(helper_maskmov,SUFFIX)373 void glue(helper_maskmov, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
374 target_ulong a0)
375 {
376 int i;
377
378 for (i = 0; i < (8 << SHIFT); i++) {
379 if (s->B(i) & 0x80) {
380 cpu_stb_data_ra(env, a0 + i, d->B(i), GETPC());
381 }
382 }
383 }
384 #endif
385
386 #define SHUFFLE4(F, a, b, offset) do { \
387 r0 = a->F((order & 3) + offset); \
388 r1 = a->F(((order >> 2) & 3) + offset); \
389 r2 = b->F(((order >> 4) & 3) + offset); \
390 r3 = b->F(((order >> 6) & 3) + offset); \
391 d->F(offset) = r0; \
392 d->F(offset + 1) = r1; \
393 d->F(offset + 2) = r2; \
394 d->F(offset + 3) = r3; \
395 } while (0)
396
397 #if SHIFT == 0
glue(helper_pshufw,SUFFIX)398 void glue(helper_pshufw, SUFFIX)(Reg *d, Reg *s, int order)
399 {
400 uint16_t r0, r1, r2, r3;
401
402 SHUFFLE4(W, s, s, 0);
403 }
404 #else
glue(helper_shufps,SUFFIX)405 void glue(helper_shufps, SUFFIX)(Reg *d, Reg *v, Reg *s, int order)
406 {
407 uint32_t r0, r1, r2, r3;
408 int i;
409
410 for (i = 0; i < 2 << SHIFT; i += 4) {
411 SHUFFLE4(L, v, s, i);
412 }
413 }
414
glue(helper_shufpd,SUFFIX)415 void glue(helper_shufpd, SUFFIX)(Reg *d, Reg *v, Reg *s, int order)
416 {
417 uint64_t r0, r1;
418 int i;
419
420 for (i = 0; i < 1 << SHIFT; i += 2) {
421 r0 = v->Q(((order & 1) & 1) + i);
422 r1 = s->Q(((order >> 1) & 1) + i);
423 d->Q(i) = r0;
424 d->Q(i + 1) = r1;
425 order >>= 2;
426 }
427 }
428
glue(helper_pshufd,SUFFIX)429 void glue(helper_pshufd, SUFFIX)(Reg *d, Reg *s, int order)
430 {
431 uint32_t r0, r1, r2, r3;
432 int i;
433
434 for (i = 0; i < 2 << SHIFT; i += 4) {
435 SHUFFLE4(L, s, s, i);
436 }
437 }
438
glue(helper_pshuflw,SUFFIX)439 void glue(helper_pshuflw, SUFFIX)(Reg *d, Reg *s, int order)
440 {
441 uint16_t r0, r1, r2, r3;
442 int i, j;
443
444 for (i = 0, j = 1; j < 1 << SHIFT; i += 8, j += 2) {
445 SHUFFLE4(W, s, s, i);
446 d->Q(j) = s->Q(j);
447 }
448 }
449
glue(helper_pshufhw,SUFFIX)450 void glue(helper_pshufhw, SUFFIX)(Reg *d, Reg *s, int order)
451 {
452 uint16_t r0, r1, r2, r3;
453 int i, j;
454
455 for (i = 4, j = 0; j < 1 << SHIFT; i += 8, j += 2) {
456 d->Q(j) = s->Q(j);
457 SHUFFLE4(W, s, s, i);
458 }
459 }
460 #endif
461
462 #if SHIFT >= 1
463 /* FPU ops */
464 /* XXX: not accurate */
465
466 #define SSE_HELPER_P(name, F) \
467 void glue(helper_ ## name ## ps, SUFFIX)(CPUX86State *env, \
468 Reg *d, Reg *v, Reg *s) \
469 { \
470 int i; \
471 for (i = 0; i < 2 << SHIFT; i++) { \
472 d->ZMM_S(i) = F(32, v->ZMM_S(i), s->ZMM_S(i)); \
473 } \
474 } \
475 \
476 void glue(helper_ ## name ## pd, SUFFIX)(CPUX86State *env, \
477 Reg *d, Reg *v, Reg *s) \
478 { \
479 int i; \
480 for (i = 0; i < 1 << SHIFT; i++) { \
481 d->ZMM_D(i) = F(64, v->ZMM_D(i), s->ZMM_D(i)); \
482 } \
483 }
484
485 #if SHIFT == 1
486
487 #define SSE_HELPER_S(name, F) \
488 SSE_HELPER_P(name, F) \
489 \
490 void helper_ ## name ## ss(CPUX86State *env, Reg *d, Reg *v, Reg *s)\
491 { \
492 int i; \
493 d->ZMM_S(0) = F(32, v->ZMM_S(0), s->ZMM_S(0)); \
494 for (i = 1; i < 2 << SHIFT; i++) { \
495 d->ZMM_L(i) = v->ZMM_L(i); \
496 } \
497 } \
498 \
499 void helper_ ## name ## sd(CPUX86State *env, Reg *d, Reg *v, Reg *s)\
500 { \
501 int i; \
502 d->ZMM_D(0) = F(64, v->ZMM_D(0), s->ZMM_D(0)); \
503 for (i = 1; i < 1 << SHIFT; i++) { \
504 d->ZMM_Q(i) = v->ZMM_Q(i); \
505 } \
506 }
507
508 #else
509
510 #define SSE_HELPER_S(name, F) SSE_HELPER_P(name, F)
511
512 #endif
513
514 #define FPU_ADD(size, a, b) float ## size ## _add(a, b, &env->sse_status)
515 #define FPU_SUB(size, a, b) float ## size ## _sub(a, b, &env->sse_status)
516 #define FPU_MUL(size, a, b) float ## size ## _mul(a, b, &env->sse_status)
517 #define FPU_DIV(size, a, b) float ## size ## _div(a, b, &env->sse_status)
518
519 /* Note that the choice of comparison op here is important to get the
520 * special cases right: for min and max Intel specifies that (-0,0),
521 * (NaN, anything) and (anything, NaN) return the second argument.
522 */
523 #define FPU_MIN(size, a, b) \
524 (float ## size ## _lt(a, b, &env->sse_status) ? (a) : (b))
525 #define FPU_MAX(size, a, b) \
526 (float ## size ## _lt(b, a, &env->sse_status) ? (a) : (b))
527
SSE_HELPER_S(add,FPU_ADD)528 SSE_HELPER_S(add, FPU_ADD)
529 SSE_HELPER_S(sub, FPU_SUB)
530 SSE_HELPER_S(mul, FPU_MUL)
531 SSE_HELPER_S(div, FPU_DIV)
532 SSE_HELPER_S(min, FPU_MIN)
533 SSE_HELPER_S(max, FPU_MAX)
534
535 void glue(helper_sqrtps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
536 {
537 int i;
538 for (i = 0; i < 2 << SHIFT; i++) {
539 d->ZMM_S(i) = float32_sqrt(s->ZMM_S(i), &env->sse_status);
540 }
541 }
542
glue(helper_sqrtpd,SUFFIX)543 void glue(helper_sqrtpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
544 {
545 int i;
546 for (i = 0; i < 1 << SHIFT; i++) {
547 d->ZMM_D(i) = float64_sqrt(s->ZMM_D(i), &env->sse_status);
548 }
549 }
550
551 #if SHIFT == 1
helper_sqrtss(CPUX86State * env,Reg * d,Reg * v,Reg * s)552 void helper_sqrtss(CPUX86State *env, Reg *d, Reg *v, Reg *s)
553 {
554 int i;
555 d->ZMM_S(0) = float32_sqrt(s->ZMM_S(0), &env->sse_status);
556 for (i = 1; i < 2 << SHIFT; i++) {
557 d->ZMM_L(i) = v->ZMM_L(i);
558 }
559 }
560
helper_sqrtsd(CPUX86State * env,Reg * d,Reg * v,Reg * s)561 void helper_sqrtsd(CPUX86State *env, Reg *d, Reg *v, Reg *s)
562 {
563 int i;
564 d->ZMM_D(0) = float64_sqrt(s->ZMM_D(0), &env->sse_status);
565 for (i = 1; i < 1 << SHIFT; i++) {
566 d->ZMM_Q(i) = v->ZMM_Q(i);
567 }
568 }
569 #endif
570
571 /* float to float conversions */
glue(helper_cvtps2pd,SUFFIX)572 void glue(helper_cvtps2pd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
573 {
574 int i;
575 for (i = 1 << SHIFT; --i >= 0; ) {
576 d->ZMM_D(i) = float32_to_float64(s->ZMM_S(i), &env->sse_status);
577 }
578 }
579
glue(helper_cvtpd2ps,SUFFIX)580 void glue(helper_cvtpd2ps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
581 {
582 int i;
583 for (i = 0; i < 1 << SHIFT; i++) {
584 d->ZMM_S(i) = float64_to_float32(s->ZMM_D(i), &env->sse_status);
585 }
586 for (i >>= 1; i < 1 << SHIFT; i++) {
587 d->Q(i) = 0;
588 }
589 }
590
591 #if SHIFT >= 1
glue(helper_cvtph2ps,SUFFIX)592 void glue(helper_cvtph2ps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
593 {
594 int i;
595
596 for (i = 2 << SHIFT; --i >= 0; ) {
597 d->ZMM_S(i) = float16_to_float32(s->ZMM_H(i), true, &env->sse_status);
598 }
599 }
600
glue(helper_cvtps2ph,SUFFIX)601 void glue(helper_cvtps2ph, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, int mode)
602 {
603 int i;
604 FloatRoundMode prev_rounding_mode = env->sse_status.float_rounding_mode;
605 if (!(mode & (1 << 2))) {
606 set_x86_rounding_mode(mode & 3, &env->sse_status);
607 }
608
609 for (i = 0; i < 2 << SHIFT; i++) {
610 d->ZMM_H(i) = float32_to_float16(s->ZMM_S(i), true, &env->sse_status);
611 }
612 for (i >>= 2; i < 1 << SHIFT; i++) {
613 d->Q(i) = 0;
614 }
615
616 env->sse_status.float_rounding_mode = prev_rounding_mode;
617 }
618 #endif
619
620 #if SHIFT == 1
helper_cvtss2sd(CPUX86State * env,Reg * d,Reg * v,Reg * s)621 void helper_cvtss2sd(CPUX86State *env, Reg *d, Reg *v, Reg *s)
622 {
623 int i;
624 d->ZMM_D(0) = float32_to_float64(s->ZMM_S(0), &env->sse_status);
625 for (i = 1; i < 1 << SHIFT; i++) {
626 d->ZMM_Q(i) = v->ZMM_Q(i);
627 }
628 }
629
helper_cvtsd2ss(CPUX86State * env,Reg * d,Reg * v,Reg * s)630 void helper_cvtsd2ss(CPUX86State *env, Reg *d, Reg *v, Reg *s)
631 {
632 int i;
633 d->ZMM_S(0) = float64_to_float32(s->ZMM_D(0), &env->sse_status);
634 for (i = 1; i < 2 << SHIFT; i++) {
635 d->ZMM_L(i) = v->ZMM_L(i);
636 }
637 }
638 #endif
639
640 /* integer to float */
glue(helper_cvtdq2ps,SUFFIX)641 void glue(helper_cvtdq2ps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
642 {
643 int i;
644 for (i = 0; i < 2 << SHIFT; i++) {
645 d->ZMM_S(i) = int32_to_float32(s->ZMM_L(i), &env->sse_status);
646 }
647 }
648
glue(helper_cvtdq2pd,SUFFIX)649 void glue(helper_cvtdq2pd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
650 {
651 int i;
652 for (i = 1 << SHIFT; --i >= 0; ) {
653 int32_t l = s->ZMM_L(i);
654 d->ZMM_D(i) = int32_to_float64(l, &env->sse_status);
655 }
656 }
657
658 #if SHIFT == 1
helper_cvtpi2ps(CPUX86State * env,ZMMReg * d,MMXReg * s)659 void helper_cvtpi2ps(CPUX86State *env, ZMMReg *d, MMXReg *s)
660 {
661 d->ZMM_S(0) = int32_to_float32(s->MMX_L(0), &env->sse_status);
662 d->ZMM_S(1) = int32_to_float32(s->MMX_L(1), &env->sse_status);
663 }
664
helper_cvtpi2pd(CPUX86State * env,ZMMReg * d,MMXReg * s)665 void helper_cvtpi2pd(CPUX86State *env, ZMMReg *d, MMXReg *s)
666 {
667 d->ZMM_D(0) = int32_to_float64(s->MMX_L(0), &env->sse_status);
668 d->ZMM_D(1) = int32_to_float64(s->MMX_L(1), &env->sse_status);
669 }
670
helper_cvtsi2ss(CPUX86State * env,ZMMReg * d,uint32_t val)671 void helper_cvtsi2ss(CPUX86State *env, ZMMReg *d, uint32_t val)
672 {
673 d->ZMM_S(0) = int32_to_float32(val, &env->sse_status);
674 }
675
helper_cvtsi2sd(CPUX86State * env,ZMMReg * d,uint32_t val)676 void helper_cvtsi2sd(CPUX86State *env, ZMMReg *d, uint32_t val)
677 {
678 d->ZMM_D(0) = int32_to_float64(val, &env->sse_status);
679 }
680
681 #ifdef TARGET_X86_64
helper_cvtsq2ss(CPUX86State * env,ZMMReg * d,uint64_t val)682 void helper_cvtsq2ss(CPUX86State *env, ZMMReg *d, uint64_t val)
683 {
684 d->ZMM_S(0) = int64_to_float32(val, &env->sse_status);
685 }
686
helper_cvtsq2sd(CPUX86State * env,ZMMReg * d,uint64_t val)687 void helper_cvtsq2sd(CPUX86State *env, ZMMReg *d, uint64_t val)
688 {
689 d->ZMM_D(0) = int64_to_float64(val, &env->sse_status);
690 }
691 #endif
692
693 #endif
694
695 /* float to integer */
696
697 #if SHIFT == 1
698 /*
699 * x86 mandates that we return the indefinite integer value for the result
700 * of any float-to-integer conversion that raises the 'invalid' exception.
701 * Wrap the softfloat functions to get this behaviour.
702 */
703 #define WRAP_FLOATCONV(RETTYPE, FN, FLOATTYPE, INDEFVALUE) \
704 static inline RETTYPE x86_##FN(FLOATTYPE a, float_status *s) \
705 { \
706 int oldflags, newflags; \
707 RETTYPE r; \
708 \
709 oldflags = get_float_exception_flags(s); \
710 set_float_exception_flags(0, s); \
711 r = FN(a, s); \
712 newflags = get_float_exception_flags(s); \
713 if (newflags & float_flag_invalid) { \
714 r = INDEFVALUE; \
715 } \
716 set_float_exception_flags(newflags | oldflags, s); \
717 return r; \
718 }
719
WRAP_FLOATCONV(int32_t,float32_to_int32,float32,INT32_MIN)720 WRAP_FLOATCONV(int32_t, float32_to_int32, float32, INT32_MIN)
721 WRAP_FLOATCONV(int32_t, float32_to_int32_round_to_zero, float32, INT32_MIN)
722 WRAP_FLOATCONV(int32_t, float64_to_int32, float64, INT32_MIN)
723 WRAP_FLOATCONV(int32_t, float64_to_int32_round_to_zero, float64, INT32_MIN)
724 WRAP_FLOATCONV(int64_t, float32_to_int64, float32, INT64_MIN)
725 WRAP_FLOATCONV(int64_t, float32_to_int64_round_to_zero, float32, INT64_MIN)
726 WRAP_FLOATCONV(int64_t, float64_to_int64, float64, INT64_MIN)
727 WRAP_FLOATCONV(int64_t, float64_to_int64_round_to_zero, float64, INT64_MIN)
728 #endif
729
730 void glue(helper_cvtps2dq, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
731 {
732 int i;
733 for (i = 0; i < 2 << SHIFT; i++) {
734 d->ZMM_L(i) = x86_float32_to_int32(s->ZMM_S(i), &env->sse_status);
735 }
736 }
737
glue(helper_cvtpd2dq,SUFFIX)738 void glue(helper_cvtpd2dq, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
739 {
740 int i;
741 for (i = 0; i < 1 << SHIFT; i++) {
742 d->ZMM_L(i) = x86_float64_to_int32(s->ZMM_D(i), &env->sse_status);
743 }
744 for (i >>= 1; i < 1 << SHIFT; i++) {
745 d->Q(i) = 0;
746 }
747 }
748
749 #if SHIFT == 1
helper_cvtps2pi(CPUX86State * env,MMXReg * d,ZMMReg * s)750 void helper_cvtps2pi(CPUX86State *env, MMXReg *d, ZMMReg *s)
751 {
752 d->MMX_L(0) = x86_float32_to_int32(s->ZMM_S(0), &env->sse_status);
753 d->MMX_L(1) = x86_float32_to_int32(s->ZMM_S(1), &env->sse_status);
754 }
755
helper_cvtpd2pi(CPUX86State * env,MMXReg * d,ZMMReg * s)756 void helper_cvtpd2pi(CPUX86State *env, MMXReg *d, ZMMReg *s)
757 {
758 d->MMX_L(0) = x86_float64_to_int32(s->ZMM_D(0), &env->sse_status);
759 d->MMX_L(1) = x86_float64_to_int32(s->ZMM_D(1), &env->sse_status);
760 }
761
helper_cvtss2si(CPUX86State * env,ZMMReg * s)762 int32_t helper_cvtss2si(CPUX86State *env, ZMMReg *s)
763 {
764 return x86_float32_to_int32(s->ZMM_S(0), &env->sse_status);
765 }
766
helper_cvtsd2si(CPUX86State * env,ZMMReg * s)767 int32_t helper_cvtsd2si(CPUX86State *env, ZMMReg *s)
768 {
769 return x86_float64_to_int32(s->ZMM_D(0), &env->sse_status);
770 }
771
772 #ifdef TARGET_X86_64
helper_cvtss2sq(CPUX86State * env,ZMMReg * s)773 int64_t helper_cvtss2sq(CPUX86State *env, ZMMReg *s)
774 {
775 return x86_float32_to_int64(s->ZMM_S(0), &env->sse_status);
776 }
777
helper_cvtsd2sq(CPUX86State * env,ZMMReg * s)778 int64_t helper_cvtsd2sq(CPUX86State *env, ZMMReg *s)
779 {
780 return x86_float64_to_int64(s->ZMM_D(0), &env->sse_status);
781 }
782 #endif
783 #endif
784
785 /* float to integer truncated */
glue(helper_cvttps2dq,SUFFIX)786 void glue(helper_cvttps2dq, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
787 {
788 int i;
789 for (i = 0; i < 2 << SHIFT; i++) {
790 d->ZMM_L(i) = x86_float32_to_int32_round_to_zero(s->ZMM_S(i),
791 &env->sse_status);
792 }
793 }
794
glue(helper_cvttpd2dq,SUFFIX)795 void glue(helper_cvttpd2dq, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
796 {
797 int i;
798 for (i = 0; i < 1 << SHIFT; i++) {
799 d->ZMM_L(i) = x86_float64_to_int32_round_to_zero(s->ZMM_D(i),
800 &env->sse_status);
801 }
802 for (i >>= 1; i < 1 << SHIFT; i++) {
803 d->Q(i) = 0;
804 }
805 }
806
807 #if SHIFT == 1
helper_cvttps2pi(CPUX86State * env,MMXReg * d,ZMMReg * s)808 void helper_cvttps2pi(CPUX86State *env, MMXReg *d, ZMMReg *s)
809 {
810 d->MMX_L(0) = x86_float32_to_int32_round_to_zero(s->ZMM_S(0), &env->sse_status);
811 d->MMX_L(1) = x86_float32_to_int32_round_to_zero(s->ZMM_S(1), &env->sse_status);
812 }
813
helper_cvttpd2pi(CPUX86State * env,MMXReg * d,ZMMReg * s)814 void helper_cvttpd2pi(CPUX86State *env, MMXReg *d, ZMMReg *s)
815 {
816 d->MMX_L(0) = x86_float64_to_int32_round_to_zero(s->ZMM_D(0), &env->sse_status);
817 d->MMX_L(1) = x86_float64_to_int32_round_to_zero(s->ZMM_D(1), &env->sse_status);
818 }
819
helper_cvttss2si(CPUX86State * env,ZMMReg * s)820 int32_t helper_cvttss2si(CPUX86State *env, ZMMReg *s)
821 {
822 return x86_float32_to_int32_round_to_zero(s->ZMM_S(0), &env->sse_status);
823 }
824
helper_cvttsd2si(CPUX86State * env,ZMMReg * s)825 int32_t helper_cvttsd2si(CPUX86State *env, ZMMReg *s)
826 {
827 return x86_float64_to_int32_round_to_zero(s->ZMM_D(0), &env->sse_status);
828 }
829
830 #ifdef TARGET_X86_64
helper_cvttss2sq(CPUX86State * env,ZMMReg * s)831 int64_t helper_cvttss2sq(CPUX86State *env, ZMMReg *s)
832 {
833 return x86_float32_to_int64_round_to_zero(s->ZMM_S(0), &env->sse_status);
834 }
835
helper_cvttsd2sq(CPUX86State * env,ZMMReg * s)836 int64_t helper_cvttsd2sq(CPUX86State *env, ZMMReg *s)
837 {
838 return x86_float64_to_int64_round_to_zero(s->ZMM_D(0), &env->sse_status);
839 }
840 #endif
841 #endif
842
glue(helper_rsqrtps,SUFFIX)843 void glue(helper_rsqrtps, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
844 {
845 uint8_t old_flags = get_float_exception_flags(&env->sse_status);
846 int i;
847 for (i = 0; i < 2 << SHIFT; i++) {
848 d->ZMM_S(i) = float32_div(float32_one,
849 float32_sqrt(s->ZMM_S(i), &env->sse_status),
850 &env->sse_status);
851 }
852 set_float_exception_flags(old_flags, &env->sse_status);
853 }
854
855 #if SHIFT == 1
helper_rsqrtss(CPUX86State * env,ZMMReg * d,ZMMReg * v,ZMMReg * s)856 void helper_rsqrtss(CPUX86State *env, ZMMReg *d, ZMMReg *v, ZMMReg *s)
857 {
858 uint8_t old_flags = get_float_exception_flags(&env->sse_status);
859 int i;
860 d->ZMM_S(0) = float32_div(float32_one,
861 float32_sqrt(s->ZMM_S(0), &env->sse_status),
862 &env->sse_status);
863 set_float_exception_flags(old_flags, &env->sse_status);
864 for (i = 1; i < 2 << SHIFT; i++) {
865 d->ZMM_L(i) = v->ZMM_L(i);
866 }
867 }
868 #endif
869
glue(helper_rcpps,SUFFIX)870 void glue(helper_rcpps, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
871 {
872 uint8_t old_flags = get_float_exception_flags(&env->sse_status);
873 int i;
874 for (i = 0; i < 2 << SHIFT; i++) {
875 d->ZMM_S(i) = float32_div(float32_one, s->ZMM_S(i), &env->sse_status);
876 }
877 set_float_exception_flags(old_flags, &env->sse_status);
878 }
879
880 #if SHIFT == 1
helper_rcpss(CPUX86State * env,ZMMReg * d,ZMMReg * v,ZMMReg * s)881 void helper_rcpss(CPUX86State *env, ZMMReg *d, ZMMReg *v, ZMMReg *s)
882 {
883 uint8_t old_flags = get_float_exception_flags(&env->sse_status);
884 int i;
885 d->ZMM_S(0) = float32_div(float32_one, s->ZMM_S(0), &env->sse_status);
886 for (i = 1; i < 2 << SHIFT; i++) {
887 d->ZMM_L(i) = v->ZMM_L(i);
888 }
889 set_float_exception_flags(old_flags, &env->sse_status);
890 }
891 #endif
892
893 #if SHIFT == 1
helper_extrq(uint64_t src,int shift,int len)894 static inline uint64_t helper_extrq(uint64_t src, int shift, int len)
895 {
896 uint64_t mask;
897
898 if (len == 0) {
899 mask = ~0LL;
900 } else {
901 mask = (1ULL << len) - 1;
902 }
903 return (src >> shift) & mask;
904 }
905
helper_extrq_r(CPUX86State * env,ZMMReg * d,ZMMReg * s)906 void helper_extrq_r(CPUX86State *env, ZMMReg *d, ZMMReg *s)
907 {
908 d->ZMM_Q(0) = helper_extrq(d->ZMM_Q(0), s->ZMM_B(1) & 63, s->ZMM_B(0) & 63);
909 }
910
helper_extrq_i(CPUX86State * env,ZMMReg * d,int index,int length)911 void helper_extrq_i(CPUX86State *env, ZMMReg *d, int index, int length)
912 {
913 d->ZMM_Q(0) = helper_extrq(d->ZMM_Q(0), index, length);
914 }
915
helper_insertq(uint64_t dest,uint64_t src,int shift,int len)916 static inline uint64_t helper_insertq(uint64_t dest, uint64_t src, int shift, int len)
917 {
918 uint64_t mask;
919
920 if (len == 0) {
921 mask = ~0ULL;
922 } else {
923 mask = (1ULL << len) - 1;
924 }
925 return (dest & ~(mask << shift)) | ((src & mask) << shift);
926 }
927
helper_insertq_r(CPUX86State * env,ZMMReg * d,ZMMReg * s)928 void helper_insertq_r(CPUX86State *env, ZMMReg *d, ZMMReg *s)
929 {
930 d->ZMM_Q(0) = helper_insertq(d->ZMM_Q(0), s->ZMM_Q(0), s->ZMM_B(9) & 63, s->ZMM_B(8) & 63);
931 }
932
helper_insertq_i(CPUX86State * env,ZMMReg * d,ZMMReg * s,int index,int length)933 void helper_insertq_i(CPUX86State *env, ZMMReg *d, ZMMReg *s, int index, int length)
934 {
935 d->ZMM_Q(0) = helper_insertq(d->ZMM_Q(0), s->ZMM_Q(0), index, length);
936 }
937 #endif
938
939 #define SSE_HELPER_HPS(name, F) \
940 void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s) \
941 { \
942 float32 r[2 << SHIFT]; \
943 int i, j, k; \
944 for (k = 0; k < 2 << SHIFT; k += LANE_WIDTH / 4) { \
945 for (i = j = 0; j < 4; i++, j += 2) { \
946 r[i + k] = F(v->ZMM_S(j + k), v->ZMM_S(j + k + 1), &env->sse_status); \
947 } \
948 for (j = 0; j < 4; i++, j += 2) { \
949 r[i + k] = F(s->ZMM_S(j + k), s->ZMM_S(j + k + 1), &env->sse_status); \
950 } \
951 } \
952 for (i = 0; i < 2 << SHIFT; i++) { \
953 d->ZMM_S(i) = r[i]; \
954 } \
955 }
956
SSE_HELPER_HPS(haddps,float32_add)957 SSE_HELPER_HPS(haddps, float32_add)
958 SSE_HELPER_HPS(hsubps, float32_sub)
959
960 #define SSE_HELPER_HPD(name, F) \
961 void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s) \
962 { \
963 float64 r[1 << SHIFT]; \
964 int i, j, k; \
965 for (k = 0; k < 1 << SHIFT; k += LANE_WIDTH / 8) { \
966 for (i = j = 0; j < 2; i++, j += 2) { \
967 r[i + k] = F(v->ZMM_D(j + k), v->ZMM_D(j + k + 1), &env->sse_status); \
968 } \
969 for (j = 0; j < 2; i++, j += 2) { \
970 r[i + k] = F(s->ZMM_D(j + k), s->ZMM_D(j + k + 1), &env->sse_status); \
971 } \
972 } \
973 for (i = 0; i < 1 << SHIFT; i++) { \
974 d->ZMM_D(i) = r[i]; \
975 } \
976 }
977
978 SSE_HELPER_HPD(haddpd, float64_add)
979 SSE_HELPER_HPD(hsubpd, float64_sub)
980
981 void glue(helper_addsubps, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
982 {
983 int i;
984 for (i = 0; i < 2 << SHIFT; i += 2) {
985 d->ZMM_S(i) = float32_sub(v->ZMM_S(i), s->ZMM_S(i), &env->sse_status);
986 d->ZMM_S(i+1) = float32_add(v->ZMM_S(i+1), s->ZMM_S(i+1), &env->sse_status);
987 }
988 }
989
glue(helper_addsubpd,SUFFIX)990 void glue(helper_addsubpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
991 {
992 int i;
993 for (i = 0; i < 1 << SHIFT; i += 2) {
994 d->ZMM_D(i) = float64_sub(v->ZMM_D(i), s->ZMM_D(i), &env->sse_status);
995 d->ZMM_D(i+1) = float64_add(v->ZMM_D(i+1), s->ZMM_D(i+1), &env->sse_status);
996 }
997 }
998
999 #define SSE_HELPER_CMP_P(name, F, C) \
1000 void glue(helper_ ## name ## ps, SUFFIX)(CPUX86State *env, \
1001 Reg *d, Reg *v, Reg *s) \
1002 { \
1003 int i; \
1004 for (i = 0; i < 2 << SHIFT; i++) { \
1005 d->ZMM_L(i) = C(F(32, v->ZMM_S(i), s->ZMM_S(i))) ? -1 : 0; \
1006 } \
1007 } \
1008 \
1009 void glue(helper_ ## name ## pd, SUFFIX)(CPUX86State *env, \
1010 Reg *d, Reg *v, Reg *s) \
1011 { \
1012 int i; \
1013 for (i = 0; i < 1 << SHIFT; i++) { \
1014 d->ZMM_Q(i) = C(F(64, v->ZMM_D(i), s->ZMM_D(i))) ? -1 : 0; \
1015 } \
1016 }
1017
1018 #if SHIFT == 1
1019 #define SSE_HELPER_CMP(name, F, C) \
1020 SSE_HELPER_CMP_P(name, F, C) \
1021 void helper_ ## name ## ss(CPUX86State *env, Reg *d, Reg *v, Reg *s) \
1022 { \
1023 int i; \
1024 d->ZMM_L(0) = C(F(32, v->ZMM_S(0), s->ZMM_S(0))) ? -1 : 0; \
1025 for (i = 1; i < 2 << SHIFT; i++) { \
1026 d->ZMM_L(i) = v->ZMM_L(i); \
1027 } \
1028 } \
1029 \
1030 void helper_ ## name ## sd(CPUX86State *env, Reg *d, Reg *v, Reg *s) \
1031 { \
1032 int i; \
1033 d->ZMM_Q(0) = C(F(64, v->ZMM_D(0), s->ZMM_D(0))) ? -1 : 0; \
1034 for (i = 1; i < 1 << SHIFT; i++) { \
1035 d->ZMM_Q(i) = v->ZMM_Q(i); \
1036 } \
1037 }
1038
FPU_EQU(FloatRelation x)1039 static inline bool FPU_EQU(FloatRelation x)
1040 {
1041 return (x == float_relation_equal || x == float_relation_unordered);
1042 }
FPU_GE(FloatRelation x)1043 static inline bool FPU_GE(FloatRelation x)
1044 {
1045 return (x == float_relation_equal || x == float_relation_greater);
1046 }
1047 #define FPU_EQ(x) (x == float_relation_equal)
1048 #define FPU_LT(x) (x == float_relation_less)
1049 #define FPU_LE(x) (x <= float_relation_equal)
1050 #define FPU_GT(x) (x == float_relation_greater)
1051 #define FPU_UNORD(x) (x == float_relation_unordered)
1052 /* We must make sure we evaluate the argument in case it is a signalling NAN */
1053 #define FPU_FALSE(x) (x == float_relation_equal && 0)
1054
1055 #define FPU_CMPQ(size, a, b) \
1056 float ## size ## _compare_quiet(a, b, &env->sse_status)
1057 #define FPU_CMPS(size, a, b) \
1058 float ## size ## _compare(a, b, &env->sse_status)
1059
1060 #else
1061 #define SSE_HELPER_CMP(name, F, C) SSE_HELPER_CMP_P(name, F, C)
1062 #endif
1063
1064 SSE_HELPER_CMP(cmpeq, FPU_CMPQ, FPU_EQ)
1065 SSE_HELPER_CMP(cmplt, FPU_CMPS, FPU_LT)
1066 SSE_HELPER_CMP(cmple, FPU_CMPS, FPU_LE)
1067 SSE_HELPER_CMP(cmpunord, FPU_CMPQ, FPU_UNORD)
1068 SSE_HELPER_CMP(cmpneq, FPU_CMPQ, !FPU_EQ)
1069 SSE_HELPER_CMP(cmpnlt, FPU_CMPS, !FPU_LT)
1070 SSE_HELPER_CMP(cmpnle, FPU_CMPS, !FPU_LE)
1071 SSE_HELPER_CMP(cmpord, FPU_CMPQ, !FPU_UNORD)
1072
1073 SSE_HELPER_CMP(cmpequ, FPU_CMPQ, FPU_EQU)
1074 SSE_HELPER_CMP(cmpnge, FPU_CMPS, !FPU_GE)
1075 SSE_HELPER_CMP(cmpngt, FPU_CMPS, !FPU_GT)
1076 SSE_HELPER_CMP(cmpfalse, FPU_CMPQ, FPU_FALSE)
1077 SSE_HELPER_CMP(cmpnequ, FPU_CMPQ, !FPU_EQU)
1078 SSE_HELPER_CMP(cmpge, FPU_CMPS, FPU_GE)
1079 SSE_HELPER_CMP(cmpgt, FPU_CMPS, FPU_GT)
1080 SSE_HELPER_CMP(cmptrue, FPU_CMPQ, !FPU_FALSE)
1081
1082 SSE_HELPER_CMP(cmpeqs, FPU_CMPS, FPU_EQ)
1083 SSE_HELPER_CMP(cmpltq, FPU_CMPQ, FPU_LT)
1084 SSE_HELPER_CMP(cmpleq, FPU_CMPQ, FPU_LE)
1085 SSE_HELPER_CMP(cmpunords, FPU_CMPS, FPU_UNORD)
1086 SSE_HELPER_CMP(cmpneqq, FPU_CMPS, !FPU_EQ)
1087 SSE_HELPER_CMP(cmpnltq, FPU_CMPQ, !FPU_LT)
1088 SSE_HELPER_CMP(cmpnleq, FPU_CMPQ, !FPU_LE)
1089 SSE_HELPER_CMP(cmpords, FPU_CMPS, !FPU_UNORD)
1090
1091 SSE_HELPER_CMP(cmpequs, FPU_CMPS, FPU_EQU)
1092 SSE_HELPER_CMP(cmpngeq, FPU_CMPQ, !FPU_GE)
1093 SSE_HELPER_CMP(cmpngtq, FPU_CMPQ, !FPU_GT)
1094 SSE_HELPER_CMP(cmpfalses, FPU_CMPS, FPU_FALSE)
1095 SSE_HELPER_CMP(cmpnequs, FPU_CMPS, !FPU_EQU)
1096 SSE_HELPER_CMP(cmpgeq, FPU_CMPQ, FPU_GE)
1097 SSE_HELPER_CMP(cmpgtq, FPU_CMPQ, FPU_GT)
1098 SSE_HELPER_CMP(cmptrues, FPU_CMPS, !FPU_FALSE)
1099
1100 #undef SSE_HELPER_CMP
1101
1102 #if SHIFT == 1
1103 static const int comis_eflags[4] = {CC_C, CC_Z, 0, CC_Z | CC_P | CC_C};
1104
helper_ucomiss(CPUX86State * env,Reg * d,Reg * s)1105 void helper_ucomiss(CPUX86State *env, Reg *d, Reg *s)
1106 {
1107 FloatRelation ret;
1108 float32 s0, s1;
1109
1110 s0 = d->ZMM_S(0);
1111 s1 = s->ZMM_S(0);
1112 ret = float32_compare_quiet(s0, s1, &env->sse_status);
1113 CC_SRC = comis_eflags[ret + 1];
1114 }
1115
helper_comiss(CPUX86State * env,Reg * d,Reg * s)1116 void helper_comiss(CPUX86State *env, Reg *d, Reg *s)
1117 {
1118 FloatRelation ret;
1119 float32 s0, s1;
1120
1121 s0 = d->ZMM_S(0);
1122 s1 = s->ZMM_S(0);
1123 ret = float32_compare(s0, s1, &env->sse_status);
1124 CC_SRC = comis_eflags[ret + 1];
1125 }
1126
helper_ucomisd(CPUX86State * env,Reg * d,Reg * s)1127 void helper_ucomisd(CPUX86State *env, Reg *d, Reg *s)
1128 {
1129 FloatRelation ret;
1130 float64 d0, d1;
1131
1132 d0 = d->ZMM_D(0);
1133 d1 = s->ZMM_D(0);
1134 ret = float64_compare_quiet(d0, d1, &env->sse_status);
1135 CC_SRC = comis_eflags[ret + 1];
1136 }
1137
helper_comisd(CPUX86State * env,Reg * d,Reg * s)1138 void helper_comisd(CPUX86State *env, Reg *d, Reg *s)
1139 {
1140 FloatRelation ret;
1141 float64 d0, d1;
1142
1143 d0 = d->ZMM_D(0);
1144 d1 = s->ZMM_D(0);
1145 ret = float64_compare(d0, d1, &env->sse_status);
1146 CC_SRC = comis_eflags[ret + 1];
1147 }
1148 #endif
1149
glue(helper_movmskps,SUFFIX)1150 uint32_t glue(helper_movmskps, SUFFIX)(CPUX86State *env, Reg *s)
1151 {
1152 uint32_t mask;
1153 int i;
1154
1155 mask = 0;
1156 for (i = 0; i < 2 << SHIFT; i++) {
1157 mask |= (s->ZMM_L(i) >> (31 - i)) & (1 << i);
1158 }
1159 return mask;
1160 }
1161
glue(helper_movmskpd,SUFFIX)1162 uint32_t glue(helper_movmskpd, SUFFIX)(CPUX86State *env, Reg *s)
1163 {
1164 uint32_t mask;
1165 int i;
1166
1167 mask = 0;
1168 for (i = 0; i < 1 << SHIFT; i++) {
1169 mask |= (s->ZMM_Q(i) >> (63 - i)) & (1 << i);
1170 }
1171 return mask;
1172 }
1173
1174 #endif
1175
1176 #define PACK_HELPER_B(name, F) \
1177 void glue(helper_pack ## name, SUFFIX)(CPUX86State *env, \
1178 Reg *d, Reg *v, Reg *s) \
1179 { \
1180 uint8_t r[PACK_WIDTH * 2]; \
1181 int j, k; \
1182 for (j = 0; j < 4 << SHIFT; j += PACK_WIDTH) { \
1183 for (k = 0; k < PACK_WIDTH; k++) { \
1184 r[k] = F((int16_t)v->W(j + k)); \
1185 } \
1186 for (k = 0; k < PACK_WIDTH; k++) { \
1187 r[PACK_WIDTH + k] = F((int16_t)s->W(j + k)); \
1188 } \
1189 for (k = 0; k < PACK_WIDTH * 2; k++) { \
1190 d->B(2 * j + k) = r[k]; \
1191 } \
1192 } \
1193 }
1194
PACK_HELPER_B(sswb,satsb)1195 PACK_HELPER_B(sswb, satsb)
1196 PACK_HELPER_B(uswb, satub)
1197
1198 void glue(helper_packssdw, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
1199 {
1200 uint16_t r[PACK_WIDTH];
1201 int j, k;
1202
1203 for (j = 0; j < 2 << SHIFT; j += PACK_WIDTH / 2) {
1204 for (k = 0; k < PACK_WIDTH / 2; k++) {
1205 r[k] = satsw(v->L(j + k));
1206 }
1207 for (k = 0; k < PACK_WIDTH / 2; k++) {
1208 r[PACK_WIDTH / 2 + k] = satsw(s->L(j + k));
1209 }
1210 for (k = 0; k < PACK_WIDTH; k++) {
1211 d->W(2 * j + k) = r[k];
1212 }
1213 }
1214 }
1215
1216 #define UNPCK_OP(base_name, base) \
1217 \
1218 void glue(helper_punpck ## base_name ## bw, SUFFIX)(CPUX86State *env,\
1219 Reg *d, Reg *v, Reg *s) \
1220 { \
1221 uint8_t r[PACK_WIDTH * 2]; \
1222 int j, i; \
1223 \
1224 for (j = 0; j < 8 << SHIFT; ) { \
1225 int k = j + base * PACK_WIDTH; \
1226 for (i = 0; i < PACK_WIDTH; i++) { \
1227 r[2 * i] = v->B(k + i); \
1228 r[2 * i + 1] = s->B(k + i); \
1229 } \
1230 for (i = 0; i < PACK_WIDTH * 2; i++, j++) { \
1231 d->B(j) = r[i]; \
1232 } \
1233 } \
1234 } \
1235 \
1236 void glue(helper_punpck ## base_name ## wd, SUFFIX)(CPUX86State *env,\
1237 Reg *d, Reg *v, Reg *s) \
1238 { \
1239 uint16_t r[PACK_WIDTH]; \
1240 int j, i; \
1241 \
1242 for (j = 0; j < 4 << SHIFT; ) { \
1243 int k = j + base * PACK_WIDTH / 2; \
1244 for (i = 0; i < PACK_WIDTH / 2; i++) { \
1245 r[2 * i] = v->W(k + i); \
1246 r[2 * i + 1] = s->W(k + i); \
1247 } \
1248 for (i = 0; i < PACK_WIDTH; i++, j++) { \
1249 d->W(j) = r[i]; \
1250 } \
1251 } \
1252 } \
1253 \
1254 void glue(helper_punpck ## base_name ## dq, SUFFIX)(CPUX86State *env,\
1255 Reg *d, Reg *v, Reg *s) \
1256 { \
1257 uint32_t r[PACK_WIDTH / 2]; \
1258 int j, i; \
1259 \
1260 for (j = 0; j < 2 << SHIFT; ) { \
1261 int k = j + base * PACK_WIDTH / 4; \
1262 for (i = 0; i < PACK_WIDTH / 4; i++) { \
1263 r[2 * i] = v->L(k + i); \
1264 r[2 * i + 1] = s->L(k + i); \
1265 } \
1266 for (i = 0; i < PACK_WIDTH / 2; i++, j++) { \
1267 d->L(j) = r[i]; \
1268 } \
1269 } \
1270 } \
1271 \
1272 XMM_ONLY( \
1273 void glue(helper_punpck ## base_name ## qdq, SUFFIX)( \
1274 CPUX86State *env, Reg *d, Reg *v, Reg *s) \
1275 { \
1276 uint64_t r[2]; \
1277 int i; \
1278 \
1279 for (i = 0; i < 1 << SHIFT; i += 2) { \
1280 r[0] = v->Q(base + i); \
1281 r[1] = s->Q(base + i); \
1282 d->Q(i) = r[0]; \
1283 d->Q(i + 1) = r[1]; \
1284 } \
1285 } \
1286 )
1287
1288 UNPCK_OP(l, 0)
1289 UNPCK_OP(h, 1)
1290
1291 #undef PACK_WIDTH
1292 #undef PACK_HELPER_B
1293 #undef UNPCK_OP
1294
1295
1296 /* 3DNow! float ops */
1297 #if SHIFT == 0
helper_pi2fd(CPUX86State * env,MMXReg * d,MMXReg * s)1298 void helper_pi2fd(CPUX86State *env, MMXReg *d, MMXReg *s)
1299 {
1300 d->MMX_S(0) = int32_to_float32(s->MMX_L(0), &env->mmx_status);
1301 d->MMX_S(1) = int32_to_float32(s->MMX_L(1), &env->mmx_status);
1302 }
1303
helper_pi2fw(CPUX86State * env,MMXReg * d,MMXReg * s)1304 void helper_pi2fw(CPUX86State *env, MMXReg *d, MMXReg *s)
1305 {
1306 d->MMX_S(0) = int32_to_float32((int16_t)s->MMX_W(0), &env->mmx_status);
1307 d->MMX_S(1) = int32_to_float32((int16_t)s->MMX_W(2), &env->mmx_status);
1308 }
1309
helper_pf2id(CPUX86State * env,MMXReg * d,MMXReg * s)1310 void helper_pf2id(CPUX86State *env, MMXReg *d, MMXReg *s)
1311 {
1312 d->MMX_L(0) = float32_to_int32_round_to_zero(s->MMX_S(0), &env->mmx_status);
1313 d->MMX_L(1) = float32_to_int32_round_to_zero(s->MMX_S(1), &env->mmx_status);
1314 }
1315
helper_pf2iw(CPUX86State * env,MMXReg * d,MMXReg * s)1316 void helper_pf2iw(CPUX86State *env, MMXReg *d, MMXReg *s)
1317 {
1318 d->MMX_L(0) = satsw(float32_to_int32_round_to_zero(s->MMX_S(0),
1319 &env->mmx_status));
1320 d->MMX_L(1) = satsw(float32_to_int32_round_to_zero(s->MMX_S(1),
1321 &env->mmx_status));
1322 }
1323
helper_pfacc(CPUX86State * env,MMXReg * d,MMXReg * s)1324 void helper_pfacc(CPUX86State *env, MMXReg *d, MMXReg *s)
1325 {
1326 float32 r;
1327
1328 r = float32_add(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1329 d->MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1330 d->MMX_S(0) = r;
1331 }
1332
helper_pfadd(CPUX86State * env,MMXReg * d,MMXReg * s)1333 void helper_pfadd(CPUX86State *env, MMXReg *d, MMXReg *s)
1334 {
1335 d->MMX_S(0) = float32_add(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1336 d->MMX_S(1) = float32_add(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1337 }
1338
helper_pfcmpeq(CPUX86State * env,MMXReg * d,MMXReg * s)1339 void helper_pfcmpeq(CPUX86State *env, MMXReg *d, MMXReg *s)
1340 {
1341 d->MMX_L(0) = float32_eq_quiet(d->MMX_S(0), s->MMX_S(0),
1342 &env->mmx_status) ? -1 : 0;
1343 d->MMX_L(1) = float32_eq_quiet(d->MMX_S(1), s->MMX_S(1),
1344 &env->mmx_status) ? -1 : 0;
1345 }
1346
helper_pfcmpge(CPUX86State * env,MMXReg * d,MMXReg * s)1347 void helper_pfcmpge(CPUX86State *env, MMXReg *d, MMXReg *s)
1348 {
1349 d->MMX_L(0) = float32_le(s->MMX_S(0), d->MMX_S(0),
1350 &env->mmx_status) ? -1 : 0;
1351 d->MMX_L(1) = float32_le(s->MMX_S(1), d->MMX_S(1),
1352 &env->mmx_status) ? -1 : 0;
1353 }
1354
helper_pfcmpgt(CPUX86State * env,MMXReg * d,MMXReg * s)1355 void helper_pfcmpgt(CPUX86State *env, MMXReg *d, MMXReg *s)
1356 {
1357 d->MMX_L(0) = float32_lt(s->MMX_S(0), d->MMX_S(0),
1358 &env->mmx_status) ? -1 : 0;
1359 d->MMX_L(1) = float32_lt(s->MMX_S(1), d->MMX_S(1),
1360 &env->mmx_status) ? -1 : 0;
1361 }
1362
helper_pfmax(CPUX86State * env,MMXReg * d,MMXReg * s)1363 void helper_pfmax(CPUX86State *env, MMXReg *d, MMXReg *s)
1364 {
1365 if (float32_lt(d->MMX_S(0), s->MMX_S(0), &env->mmx_status)) {
1366 d->MMX_S(0) = s->MMX_S(0);
1367 }
1368 if (float32_lt(d->MMX_S(1), s->MMX_S(1), &env->mmx_status)) {
1369 d->MMX_S(1) = s->MMX_S(1);
1370 }
1371 }
1372
helper_pfmin(CPUX86State * env,MMXReg * d,MMXReg * s)1373 void helper_pfmin(CPUX86State *env, MMXReg *d, MMXReg *s)
1374 {
1375 if (float32_lt(s->MMX_S(0), d->MMX_S(0), &env->mmx_status)) {
1376 d->MMX_S(0) = s->MMX_S(0);
1377 }
1378 if (float32_lt(s->MMX_S(1), d->MMX_S(1), &env->mmx_status)) {
1379 d->MMX_S(1) = s->MMX_S(1);
1380 }
1381 }
1382
helper_pfmul(CPUX86State * env,MMXReg * d,MMXReg * s)1383 void helper_pfmul(CPUX86State *env, MMXReg *d, MMXReg *s)
1384 {
1385 d->MMX_S(0) = float32_mul(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1386 d->MMX_S(1) = float32_mul(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1387 }
1388
helper_pfnacc(CPUX86State * env,MMXReg * d,MMXReg * s)1389 void helper_pfnacc(CPUX86State *env, MMXReg *d, MMXReg *s)
1390 {
1391 float32 r;
1392
1393 r = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1394 d->MMX_S(1) = float32_sub(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1395 d->MMX_S(0) = r;
1396 }
1397
helper_pfpnacc(CPUX86State * env,MMXReg * d,MMXReg * s)1398 void helper_pfpnacc(CPUX86State *env, MMXReg *d, MMXReg *s)
1399 {
1400 float32 r;
1401
1402 r = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1403 d->MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1404 d->MMX_S(0) = r;
1405 }
1406
helper_pfrcp(CPUX86State * env,MMXReg * d,MMXReg * s)1407 void helper_pfrcp(CPUX86State *env, MMXReg *d, MMXReg *s)
1408 {
1409 d->MMX_S(0) = float32_div(float32_one, s->MMX_S(0), &env->mmx_status);
1410 d->MMX_S(1) = d->MMX_S(0);
1411 }
1412
helper_pfrsqrt(CPUX86State * env,MMXReg * d,MMXReg * s)1413 void helper_pfrsqrt(CPUX86State *env, MMXReg *d, MMXReg *s)
1414 {
1415 d->MMX_L(1) = s->MMX_L(0) & 0x7fffffff;
1416 d->MMX_S(1) = float32_div(float32_one,
1417 float32_sqrt(d->MMX_S(1), &env->mmx_status),
1418 &env->mmx_status);
1419 d->MMX_L(1) |= s->MMX_L(0) & 0x80000000;
1420 d->MMX_L(0) = d->MMX_L(1);
1421 }
1422
helper_pfsub(CPUX86State * env,MMXReg * d,MMXReg * s)1423 void helper_pfsub(CPUX86State *env, MMXReg *d, MMXReg *s)
1424 {
1425 d->MMX_S(0) = float32_sub(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1426 d->MMX_S(1) = float32_sub(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1427 }
1428
helper_pfsubr(CPUX86State * env,MMXReg * d,MMXReg * s)1429 void helper_pfsubr(CPUX86State *env, MMXReg *d, MMXReg *s)
1430 {
1431 d->MMX_S(0) = float32_sub(s->MMX_S(0), d->MMX_S(0), &env->mmx_status);
1432 d->MMX_S(1) = float32_sub(s->MMX_S(1), d->MMX_S(1), &env->mmx_status);
1433 }
1434
helper_pswapd(CPUX86State * env,MMXReg * d,MMXReg * s)1435 void helper_pswapd(CPUX86State *env, MMXReg *d, MMXReg *s)
1436 {
1437 uint32_t r;
1438
1439 r = s->MMX_L(0);
1440 d->MMX_L(0) = s->MMX_L(1);
1441 d->MMX_L(1) = r;
1442 }
1443 #endif
1444
1445 /* SSSE3 op helpers */
glue(helper_pshufb,SUFFIX)1446 void glue(helper_pshufb, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
1447 {
1448 int i;
1449 #if SHIFT == 0
1450 uint8_t r[8];
1451
1452 for (i = 0; i < 8; i++) {
1453 r[i] = (s->B(i) & 0x80) ? 0 : (v->B(s->B(i) & 7));
1454 }
1455 for (i = 0; i < 8; i++) {
1456 d->B(i) = r[i];
1457 }
1458 #else
1459 uint8_t r[8 << SHIFT];
1460
1461 for (i = 0; i < 8 << SHIFT; i++) {
1462 int j = i & ~0xf;
1463 r[i] = (s->B(i) & 0x80) ? 0 : v->B(j | (s->B(i) & 0xf));
1464 }
1465 for (i = 0; i < 8 << SHIFT; i++) {
1466 d->B(i) = r[i];
1467 }
1468 #endif
1469 }
1470
1471 #define SSE_HELPER_HW(name, F) \
1472 void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s) \
1473 { \
1474 uint16_t r[4 << SHIFT]; \
1475 int i, j, k; \
1476 for (k = 0; k < 4 << SHIFT; k += LANE_WIDTH / 2) { \
1477 for (i = j = 0; j < LANE_WIDTH / 2; i++, j += 2) { \
1478 r[i + k] = F(v->W(j + k), v->W(j + k + 1)); \
1479 } \
1480 for (j = 0; j < LANE_WIDTH / 2; i++, j += 2) { \
1481 r[i + k] = F(s->W(j + k), s->W(j + k + 1)); \
1482 } \
1483 } \
1484 for (i = 0; i < 4 << SHIFT; i++) { \
1485 d->W(i) = r[i]; \
1486 } \
1487 }
1488
1489 #define SSE_HELPER_HL(name, F) \
1490 void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s) \
1491 { \
1492 uint32_t r[2 << SHIFT]; \
1493 int i, j, k; \
1494 for (k = 0; k < 2 << SHIFT; k += LANE_WIDTH / 4) { \
1495 for (i = j = 0; j < LANE_WIDTH / 4; i++, j += 2) { \
1496 r[i + k] = F(v->L(j + k), v->L(j + k + 1)); \
1497 } \
1498 for (j = 0; j < LANE_WIDTH / 4; i++, j += 2) { \
1499 r[i + k] = F(s->L(j + k), s->L(j + k + 1)); \
1500 } \
1501 } \
1502 for (i = 0; i < 2 << SHIFT; i++) { \
1503 d->L(i) = r[i]; \
1504 } \
1505 }
1506
SSE_HELPER_HW(phaddw,FADD)1507 SSE_HELPER_HW(phaddw, FADD)
1508 SSE_HELPER_HW(phsubw, FSUB)
1509 SSE_HELPER_HW(phaddsw, FADDSW)
1510 SSE_HELPER_HW(phsubsw, FSUBSW)
1511 SSE_HELPER_HL(phaddd, FADD)
1512 SSE_HELPER_HL(phsubd, FSUB)
1513
1514 #undef SSE_HELPER_HW
1515 #undef SSE_HELPER_HL
1516
1517 void glue(helper_pmaddubsw, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
1518 {
1519 int i;
1520 for (i = 0; i < 4 << SHIFT; i++) {
1521 d->W(i) = satsw((int8_t)s->B(i * 2) * (uint8_t)v->B(i * 2) +
1522 (int8_t)s->B(i * 2 + 1) * (uint8_t)v->B(i * 2 + 1));
1523 }
1524 }
1525
1526 #define FMULHRSW(d, s) (((int16_t) d * (int16_t)s + 0x4000) >> 15)
SSE_HELPER_W(helper_pmulhrsw,FMULHRSW)1527 SSE_HELPER_W(helper_pmulhrsw, FMULHRSW)
1528
1529 #define FSIGNB(d, s) (s <= INT8_MAX ? s ? d : 0 : -(int8_t)d)
1530 #define FSIGNW(d, s) (s <= INT16_MAX ? s ? d : 0 : -(int16_t)d)
1531 #define FSIGNL(d, s) (s <= INT32_MAX ? s ? d : 0 : -(int32_t)d)
1532 SSE_HELPER_B(helper_psignb, FSIGNB)
1533 SSE_HELPER_W(helper_psignw, FSIGNW)
1534 SSE_HELPER_L(helper_psignd, FSIGNL)
1535
1536 void glue(helper_palignr, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
1537 uint32_t imm)
1538 {
1539 int i;
1540
1541 /* XXX could be checked during translation */
1542 if (imm >= (SHIFT ? 32 : 16)) {
1543 for (i = 0; i < (1 << SHIFT); i++) {
1544 d->Q(i) = 0;
1545 }
1546 } else {
1547 int shift = imm * 8;
1548 #define SHR(v, i) (i < 64 && i > -64 ? i > 0 ? v >> (i) : (v << -(i)) : 0)
1549 #if SHIFT == 0
1550 d->Q(0) = SHR(s->Q(0), shift - 0) |
1551 SHR(v->Q(0), shift - 64);
1552 #else
1553 for (i = 0; i < (1 << SHIFT); i += 2) {
1554 uint64_t r0, r1;
1555
1556 r0 = SHR(s->Q(i), shift - 0) |
1557 SHR(s->Q(i + 1), shift - 64) |
1558 SHR(v->Q(i), shift - 128) |
1559 SHR(v->Q(i + 1), shift - 192);
1560 r1 = SHR(s->Q(i), shift + 64) |
1561 SHR(s->Q(i + 1), shift - 0) |
1562 SHR(v->Q(i), shift - 64) |
1563 SHR(v->Q(i + 1), shift - 128);
1564 d->Q(i) = r0;
1565 d->Q(i + 1) = r1;
1566 }
1567 #endif
1568 #undef SHR
1569 }
1570 }
1571
1572 #if SHIFT >= 1
1573
1574 #define SSE_HELPER_V(name, elem, num, F) \
1575 void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s, \
1576 Reg *m) \
1577 { \
1578 int i; \
1579 for (i = 0; i < num; i++) { \
1580 d->elem(i) = F(v->elem(i), s->elem(i), m->elem(i)); \
1581 } \
1582 }
1583
1584 #define SSE_HELPER_I(name, elem, num, F) \
1585 void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s, \
1586 uint32_t imm) \
1587 { \
1588 int i; \
1589 for (i = 0; i < num; i++) { \
1590 int j = i & 7; \
1591 d->elem(i) = F(v->elem(i), s->elem(i), (imm >> j) & 1); \
1592 } \
1593 }
1594
1595 /* SSE4.1 op helpers */
1596 #define FBLENDVB(v, s, m) ((m & 0x80) ? s : v)
1597 #define FBLENDVPS(v, s, m) ((m & 0x80000000) ? s : v)
1598 #define FBLENDVPD(v, s, m) ((m & 0x8000000000000000LL) ? s : v)
1599 SSE_HELPER_V(helper_pblendvb, B, 8 << SHIFT, FBLENDVB)
1600 SSE_HELPER_V(helper_blendvps, L, 2 << SHIFT, FBLENDVPS)
1601 SSE_HELPER_V(helper_blendvpd, Q, 1 << SHIFT, FBLENDVPD)
1602
glue(helper_ptest,SUFFIX)1603 void glue(helper_ptest, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
1604 {
1605 uint64_t zf = 0, cf = 0;
1606 int i;
1607
1608 for (i = 0; i < 1 << SHIFT; i++) {
1609 zf |= (s->Q(i) & d->Q(i));
1610 cf |= (s->Q(i) & ~d->Q(i));
1611 }
1612 CC_SRC = (zf ? 0 : CC_Z) | (cf ? 0 : CC_C);
1613 }
1614
1615 #define FMOVSLDUP(i) s->L((i) & ~1)
1616 #define FMOVSHDUP(i) s->L((i) | 1)
1617 #define FMOVDLDUP(i) s->Q((i) & ~1)
1618
1619 #define SSE_HELPER_F(name, elem, num, F) \
1620 void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s) \
1621 { \
1622 int n = num; \
1623 for (int i = n; --i >= 0; ) { \
1624 d->elem(i) = F(i); \
1625 } \
1626 }
1627
1628 #if SHIFT > 0
1629 SSE_HELPER_F(helper_pmovsxbw, W, 4 << SHIFT, (int8_t) s->B)
1630 SSE_HELPER_F(helper_pmovsxbd, L, 2 << SHIFT, (int8_t) s->B)
1631 SSE_HELPER_F(helper_pmovsxbq, Q, 1 << SHIFT, (int8_t) s->B)
1632 SSE_HELPER_F(helper_pmovsxwd, L, 2 << SHIFT, (int16_t) s->W)
1633 SSE_HELPER_F(helper_pmovsxwq, Q, 1 << SHIFT, (int16_t) s->W)
1634 SSE_HELPER_F(helper_pmovsxdq, Q, 1 << SHIFT, (int32_t) s->L)
1635 SSE_HELPER_F(helper_pmovzxbw, W, 4 << SHIFT, s->B)
1636 SSE_HELPER_F(helper_pmovzxbd, L, 2 << SHIFT, s->B)
1637 SSE_HELPER_F(helper_pmovzxbq, Q, 1 << SHIFT, s->B)
1638 SSE_HELPER_F(helper_pmovzxwd, L, 2 << SHIFT, s->W)
1639 SSE_HELPER_F(helper_pmovzxwq, Q, 1 << SHIFT, s->W)
1640 SSE_HELPER_F(helper_pmovzxdq, Q, 1 << SHIFT, s->L)
1641 SSE_HELPER_F(helper_pmovsldup, L, 2 << SHIFT, FMOVSLDUP)
1642 SSE_HELPER_F(helper_pmovshdup, L, 2 << SHIFT, FMOVSHDUP)
1643 SSE_HELPER_F(helper_pmovdldup, Q, 1 << SHIFT, FMOVDLDUP)
1644 #endif
1645
glue(helper_pmuldq,SUFFIX)1646 void glue(helper_pmuldq, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
1647 {
1648 int i;
1649
1650 for (i = 0; i < 1 << SHIFT; i++) {
1651 d->Q(i) = (int64_t)(int32_t) v->L(2 * i) * (int32_t) s->L(2 * i);
1652 }
1653 }
1654
glue(helper_packusdw,SUFFIX)1655 void glue(helper_packusdw, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
1656 {
1657 uint16_t r[8];
1658 int i, j, k;
1659
1660 for (i = 0, j = 0; i <= 2 << SHIFT; i += 8, j += 4) {
1661 r[0] = satuw(v->L(j));
1662 r[1] = satuw(v->L(j + 1));
1663 r[2] = satuw(v->L(j + 2));
1664 r[3] = satuw(v->L(j + 3));
1665 r[4] = satuw(s->L(j));
1666 r[5] = satuw(s->L(j + 1));
1667 r[6] = satuw(s->L(j + 2));
1668 r[7] = satuw(s->L(j + 3));
1669 for (k = 0; k < 8; k++) {
1670 d->W(i + k) = r[k];
1671 }
1672 }
1673 }
1674
1675 #if SHIFT == 1
glue(helper_phminposuw,SUFFIX)1676 void glue(helper_phminposuw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
1677 {
1678 int idx = 0;
1679
1680 if (s->W(1) < s->W(idx)) {
1681 idx = 1;
1682 }
1683 if (s->W(2) < s->W(idx)) {
1684 idx = 2;
1685 }
1686 if (s->W(3) < s->W(idx)) {
1687 idx = 3;
1688 }
1689 if (s->W(4) < s->W(idx)) {
1690 idx = 4;
1691 }
1692 if (s->W(5) < s->W(idx)) {
1693 idx = 5;
1694 }
1695 if (s->W(6) < s->W(idx)) {
1696 idx = 6;
1697 }
1698 if (s->W(7) < s->W(idx)) {
1699 idx = 7;
1700 }
1701
1702 d->W(0) = s->W(idx);
1703 d->W(1) = idx;
1704 d->L(1) = 0;
1705 d->Q(1) = 0;
1706 }
1707 #endif
1708
glue(helper_roundps,SUFFIX)1709 void glue(helper_roundps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
1710 uint32_t mode)
1711 {
1712 uint8_t old_flags = get_float_exception_flags(&env->sse_status);
1713 signed char prev_rounding_mode;
1714 int i;
1715
1716 prev_rounding_mode = env->sse_status.float_rounding_mode;
1717 if (!(mode & (1 << 2))) {
1718 set_x86_rounding_mode(mode & 3, &env->sse_status);
1719 }
1720
1721 for (i = 0; i < 2 << SHIFT; i++) {
1722 d->ZMM_S(i) = float32_round_to_int(s->ZMM_S(i), &env->sse_status);
1723 }
1724
1725 if (mode & (1 << 3) && !(old_flags & float_flag_inexact)) {
1726 set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1727 ~float_flag_inexact,
1728 &env->sse_status);
1729 }
1730 env->sse_status.float_rounding_mode = prev_rounding_mode;
1731 }
1732
glue(helper_roundpd,SUFFIX)1733 void glue(helper_roundpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
1734 uint32_t mode)
1735 {
1736 uint8_t old_flags = get_float_exception_flags(&env->sse_status);
1737 signed char prev_rounding_mode;
1738 int i;
1739
1740 prev_rounding_mode = env->sse_status.float_rounding_mode;
1741 if (!(mode & (1 << 2))) {
1742 set_x86_rounding_mode(mode & 3, &env->sse_status);
1743 }
1744
1745 for (i = 0; i < 1 << SHIFT; i++) {
1746 d->ZMM_D(i) = float64_round_to_int(s->ZMM_D(i), &env->sse_status);
1747 }
1748
1749 if (mode & (1 << 3) && !(old_flags & float_flag_inexact)) {
1750 set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1751 ~float_flag_inexact,
1752 &env->sse_status);
1753 }
1754 env->sse_status.float_rounding_mode = prev_rounding_mode;
1755 }
1756
1757 #if SHIFT == 1
glue(helper_roundss,SUFFIX)1758 void glue(helper_roundss, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
1759 uint32_t mode)
1760 {
1761 uint8_t old_flags = get_float_exception_flags(&env->sse_status);
1762 signed char prev_rounding_mode;
1763 int i;
1764
1765 prev_rounding_mode = env->sse_status.float_rounding_mode;
1766 if (!(mode & (1 << 2))) {
1767 set_x86_rounding_mode(mode & 3, &env->sse_status);
1768 }
1769
1770 d->ZMM_S(0) = float32_round_to_int(s->ZMM_S(0), &env->sse_status);
1771 for (i = 1; i < 2 << SHIFT; i++) {
1772 d->ZMM_L(i) = v->ZMM_L(i);
1773 }
1774
1775 if (mode & (1 << 3) && !(old_flags & float_flag_inexact)) {
1776 set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1777 ~float_flag_inexact,
1778 &env->sse_status);
1779 }
1780 env->sse_status.float_rounding_mode = prev_rounding_mode;
1781 }
1782
glue(helper_roundsd,SUFFIX)1783 void glue(helper_roundsd, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
1784 uint32_t mode)
1785 {
1786 uint8_t old_flags = get_float_exception_flags(&env->sse_status);
1787 signed char prev_rounding_mode;
1788 int i;
1789
1790 prev_rounding_mode = env->sse_status.float_rounding_mode;
1791 if (!(mode & (1 << 2))) {
1792 set_x86_rounding_mode(mode & 3, &env->sse_status);
1793 }
1794
1795 d->ZMM_D(0) = float64_round_to_int(s->ZMM_D(0), &env->sse_status);
1796 for (i = 1; i < 1 << SHIFT; i++) {
1797 d->ZMM_Q(i) = v->ZMM_Q(i);
1798 }
1799
1800 if (mode & (1 << 3) && !(old_flags & float_flag_inexact)) {
1801 set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1802 ~float_flag_inexact,
1803 &env->sse_status);
1804 }
1805 env->sse_status.float_rounding_mode = prev_rounding_mode;
1806 }
1807 #endif
1808
1809 #define FBLENDP(v, s, m) (m ? s : v)
1810 SSE_HELPER_I(helper_blendps, L, 2 << SHIFT, FBLENDP)
1811 SSE_HELPER_I(helper_blendpd, Q, 1 << SHIFT, FBLENDP)
1812 SSE_HELPER_I(helper_pblendw, W, 4 << SHIFT, FBLENDP)
1813
glue(helper_dpps,SUFFIX)1814 void glue(helper_dpps, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
1815 uint32_t mask)
1816 {
1817 float32 prod1, prod2, temp2, temp3, temp4;
1818 int i;
1819
1820 for (i = 0; i < 2 << SHIFT; i += 4) {
1821 /*
1822 * We must evaluate (A+B)+(C+D), not ((A+B)+C)+D
1823 * to correctly round the intermediate results
1824 */
1825 if (mask & (1 << 4)) {
1826 prod1 = float32_mul(v->ZMM_S(i), s->ZMM_S(i), &env->sse_status);
1827 } else {
1828 prod1 = float32_zero;
1829 }
1830 if (mask & (1 << 5)) {
1831 prod2 = float32_mul(v->ZMM_S(i+1), s->ZMM_S(i+1), &env->sse_status);
1832 } else {
1833 prod2 = float32_zero;
1834 }
1835 temp2 = float32_add(prod1, prod2, &env->sse_status);
1836 if (mask & (1 << 6)) {
1837 prod1 = float32_mul(v->ZMM_S(i+2), s->ZMM_S(i+2), &env->sse_status);
1838 } else {
1839 prod1 = float32_zero;
1840 }
1841 if (mask & (1 << 7)) {
1842 prod2 = float32_mul(v->ZMM_S(i+3), s->ZMM_S(i+3), &env->sse_status);
1843 } else {
1844 prod2 = float32_zero;
1845 }
1846 temp3 = float32_add(prod1, prod2, &env->sse_status);
1847 temp4 = float32_add(temp2, temp3, &env->sse_status);
1848
1849 d->ZMM_S(i) = (mask & (1 << 0)) ? temp4 : float32_zero;
1850 d->ZMM_S(i+1) = (mask & (1 << 1)) ? temp4 : float32_zero;
1851 d->ZMM_S(i+2) = (mask & (1 << 2)) ? temp4 : float32_zero;
1852 d->ZMM_S(i+3) = (mask & (1 << 3)) ? temp4 : float32_zero;
1853 }
1854 }
1855
1856 #if SHIFT == 1
1857 /* Oddly, there is no ymm version of dppd */
glue(helper_dppd,SUFFIX)1858 void glue(helper_dppd, SUFFIX)(CPUX86State *env,
1859 Reg *d, Reg *v, Reg *s, uint32_t mask)
1860 {
1861 float64 prod1, prod2, temp2;
1862
1863 if (mask & (1 << 4)) {
1864 prod1 = float64_mul(v->ZMM_D(0), s->ZMM_D(0), &env->sse_status);
1865 } else {
1866 prod1 = float64_zero;
1867 }
1868 if (mask & (1 << 5)) {
1869 prod2 = float64_mul(v->ZMM_D(1), s->ZMM_D(1), &env->sse_status);
1870 } else {
1871 prod2 = float64_zero;
1872 }
1873 temp2 = float64_add(prod1, prod2, &env->sse_status);
1874 d->ZMM_D(0) = (mask & (1 << 0)) ? temp2 : float64_zero;
1875 d->ZMM_D(1) = (mask & (1 << 1)) ? temp2 : float64_zero;
1876 }
1877 #endif
1878
glue(helper_mpsadbw,SUFFIX)1879 void glue(helper_mpsadbw, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
1880 uint32_t offset)
1881 {
1882 int i, j;
1883 uint16_t r[8];
1884
1885 for (j = 0; j < 4 << SHIFT; ) {
1886 int s0 = (j * 2) + ((offset & 3) << 2);
1887 int d0 = (j * 2) + ((offset & 4) << 0);
1888 for (i = 0; i < LANE_WIDTH / 2; i++, d0++) {
1889 r[i] = 0;
1890 r[i] += abs1(v->B(d0 + 0) - s->B(s0 + 0));
1891 r[i] += abs1(v->B(d0 + 1) - s->B(s0 + 1));
1892 r[i] += abs1(v->B(d0 + 2) - s->B(s0 + 2));
1893 r[i] += abs1(v->B(d0 + 3) - s->B(s0 + 3));
1894 }
1895 for (i = 0; i < LANE_WIDTH / 2; i++, j++) {
1896 d->W(j) = r[i];
1897 }
1898 offset >>= 3;
1899 }
1900 }
1901
1902 /* SSE4.2 op helpers */
1903 #if SHIFT == 1
pcmp_elen(CPUX86State * env,int reg,uint32_t ctrl)1904 static inline int pcmp_elen(CPUX86State *env, int reg, uint32_t ctrl)
1905 {
1906 target_long val, limit;
1907
1908 /* Presence of REX.W is indicated by a bit higher than 7 set */
1909 if (ctrl >> 8) {
1910 val = (target_long)env->regs[reg];
1911 } else {
1912 val = (int32_t)env->regs[reg];
1913 }
1914 if (ctrl & 1) {
1915 limit = 8;
1916 } else {
1917 limit = 16;
1918 }
1919 if ((val > limit) || (val < -limit)) {
1920 return limit;
1921 }
1922 return abs1(val);
1923 }
1924
pcmp_ilen(Reg * r,uint8_t ctrl)1925 static inline int pcmp_ilen(Reg *r, uint8_t ctrl)
1926 {
1927 int val = 0;
1928
1929 if (ctrl & 1) {
1930 while (val < 8 && r->W(val)) {
1931 val++;
1932 }
1933 } else {
1934 while (val < 16 && r->B(val)) {
1935 val++;
1936 }
1937 }
1938
1939 return val;
1940 }
1941
pcmp_val(Reg * r,uint8_t ctrl,int i)1942 static inline int pcmp_val(Reg *r, uint8_t ctrl, int i)
1943 {
1944 switch ((ctrl >> 0) & 3) {
1945 case 0:
1946 return r->B(i);
1947 case 1:
1948 return r->W(i);
1949 case 2:
1950 return (int8_t)r->B(i);
1951 case 3:
1952 default:
1953 return (int16_t)r->W(i);
1954 }
1955 }
1956
pcmpxstrx(CPUX86State * env,Reg * d,Reg * s,uint8_t ctrl,int valids,int validd)1957 static inline unsigned pcmpxstrx(CPUX86State *env, Reg *d, Reg *s,
1958 uint8_t ctrl, int valids, int validd)
1959 {
1960 unsigned int res = 0;
1961 int v;
1962 int j, i;
1963 int upper = (ctrl & 1) ? 7 : 15;
1964
1965 valids--;
1966 validd--;
1967
1968 CC_SRC = (valids < upper ? CC_Z : 0) | (validd < upper ? CC_S : 0);
1969
1970 switch ((ctrl >> 2) & 3) {
1971 case 0:
1972 for (j = valids; j >= 0; j--) {
1973 res <<= 1;
1974 v = pcmp_val(s, ctrl, j);
1975 for (i = validd; i >= 0; i--) {
1976 res |= (v == pcmp_val(d, ctrl, i));
1977 }
1978 }
1979 break;
1980 case 1:
1981 for (j = valids; j >= 0; j--) {
1982 res <<= 1;
1983 v = pcmp_val(s, ctrl, j);
1984 for (i = ((validd - 1) | 1); i >= 0; i -= 2) {
1985 res |= (pcmp_val(d, ctrl, i - 0) >= v &&
1986 pcmp_val(d, ctrl, i - 1) <= v);
1987 }
1988 }
1989 break;
1990 case 2:
1991 res = (1 << (upper - MAX(valids, validd))) - 1;
1992 res <<= MAX(valids, validd) - MIN(valids, validd);
1993 for (i = MIN(valids, validd); i >= 0; i--) {
1994 res <<= 1;
1995 v = pcmp_val(s, ctrl, i);
1996 res |= (v == pcmp_val(d, ctrl, i));
1997 }
1998 break;
1999 case 3:
2000 if (validd == -1) {
2001 res = (2 << upper) - 1;
2002 break;
2003 }
2004 for (j = valids == upper ? valids : valids - validd; j >= 0; j--) {
2005 res <<= 1;
2006 v = 1;
2007 for (i = MIN(valids - j, validd); i >= 0; i--) {
2008 v &= (pcmp_val(s, ctrl, i + j) == pcmp_val(d, ctrl, i));
2009 }
2010 res |= v;
2011 }
2012 break;
2013 }
2014
2015 switch ((ctrl >> 4) & 3) {
2016 case 1:
2017 res ^= (2 << upper) - 1;
2018 break;
2019 case 3:
2020 res ^= (1 << (valids + 1)) - 1;
2021 break;
2022 }
2023
2024 if (res) {
2025 CC_SRC |= CC_C;
2026 }
2027 if (res & 1) {
2028 CC_SRC |= CC_O;
2029 }
2030
2031 return res;
2032 }
2033
glue(helper_pcmpestri,SUFFIX)2034 void glue(helper_pcmpestri, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2035 uint32_t ctrl)
2036 {
2037 unsigned int res = pcmpxstrx(env, d, s, ctrl,
2038 pcmp_elen(env, R_EDX, ctrl),
2039 pcmp_elen(env, R_EAX, ctrl));
2040
2041 if (res) {
2042 env->regs[R_ECX] = (ctrl & (1 << 6)) ? 31 - clz32(res) : ctz32(res);
2043 } else {
2044 env->regs[R_ECX] = 16 >> (ctrl & (1 << 0));
2045 }
2046 }
2047
glue(helper_pcmpestrm,SUFFIX)2048 void glue(helper_pcmpestrm, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2049 uint32_t ctrl)
2050 {
2051 int i;
2052 unsigned int res = pcmpxstrx(env, d, s, ctrl,
2053 pcmp_elen(env, R_EDX, ctrl),
2054 pcmp_elen(env, R_EAX, ctrl));
2055
2056 if ((ctrl >> 6) & 1) {
2057 if (ctrl & 1) {
2058 for (i = 0; i < 8; i++, res >>= 1) {
2059 env->xmm_regs[0].W(i) = (res & 1) ? ~0 : 0;
2060 }
2061 } else {
2062 for (i = 0; i < 16; i++, res >>= 1) {
2063 env->xmm_regs[0].B(i) = (res & 1) ? ~0 : 0;
2064 }
2065 }
2066 } else {
2067 env->xmm_regs[0].Q(1) = 0;
2068 env->xmm_regs[0].Q(0) = res;
2069 }
2070 }
2071
glue(helper_pcmpistri,SUFFIX)2072 void glue(helper_pcmpistri, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2073 uint32_t ctrl)
2074 {
2075 unsigned int res = pcmpxstrx(env, d, s, ctrl,
2076 pcmp_ilen(s, ctrl),
2077 pcmp_ilen(d, ctrl));
2078
2079 if (res) {
2080 env->regs[R_ECX] = (ctrl & (1 << 6)) ? 31 - clz32(res) : ctz32(res);
2081 } else {
2082 env->regs[R_ECX] = 16 >> (ctrl & (1 << 0));
2083 }
2084 }
2085
glue(helper_pcmpistrm,SUFFIX)2086 void glue(helper_pcmpistrm, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2087 uint32_t ctrl)
2088 {
2089 int i;
2090 unsigned int res = pcmpxstrx(env, d, s, ctrl,
2091 pcmp_ilen(s, ctrl),
2092 pcmp_ilen(d, ctrl));
2093
2094 if ((ctrl >> 6) & 1) {
2095 if (ctrl & 1) {
2096 for (i = 0; i < 8; i++, res >>= 1) {
2097 env->xmm_regs[0].W(i) = (res & 1) ? ~0 : 0;
2098 }
2099 } else {
2100 for (i = 0; i < 16; i++, res >>= 1) {
2101 env->xmm_regs[0].B(i) = (res & 1) ? ~0 : 0;
2102 }
2103 }
2104 } else {
2105 env->xmm_regs[0].Q(1) = 0;
2106 env->xmm_regs[0].Q(0) = res;
2107 }
2108 }
2109
2110 #define CRCPOLY 0x1edc6f41
2111 #define CRCPOLY_BITREV 0x82f63b78
helper_crc32(uint32_t crc1,target_ulong msg,uint32_t len)2112 target_ulong helper_crc32(uint32_t crc1, target_ulong msg, uint32_t len)
2113 {
2114 target_ulong crc = (msg & ((target_ulong) -1 >>
2115 (TARGET_LONG_BITS - len))) ^ crc1;
2116
2117 while (len--) {
2118 crc = (crc >> 1) ^ ((crc & 1) ? CRCPOLY_BITREV : 0);
2119 }
2120
2121 return crc;
2122 }
2123
2124 #endif
2125
glue(helper_pclmulqdq,SUFFIX)2126 void glue(helper_pclmulqdq, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
2127 uint32_t ctrl)
2128 {
2129 int a_idx = (ctrl & 1) != 0;
2130 int b_idx = (ctrl & 16) != 0;
2131
2132 for (int i = 0; i < SHIFT; i++) {
2133 uint64_t a = v->Q(2 * i + a_idx);
2134 uint64_t b = s->Q(2 * i + b_idx);
2135 Int128 *r = (Int128 *)&d->ZMM_X(i);
2136
2137 *r = clmul_64(a, b);
2138 }
2139 }
2140
glue(helper_aesdec,SUFFIX)2141 void glue(helper_aesdec, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2142 {
2143 for (int i = 0; i < SHIFT; i++) {
2144 AESState *ad = (AESState *)&d->ZMM_X(i);
2145 AESState *st = (AESState *)&v->ZMM_X(i);
2146 AESState *rk = (AESState *)&s->ZMM_X(i);
2147
2148 aesdec_ISB_ISR_IMC_AK(ad, st, rk, false);
2149 }
2150 }
2151
glue(helper_aesdeclast,SUFFIX)2152 void glue(helper_aesdeclast, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2153 {
2154 for (int i = 0; i < SHIFT; i++) {
2155 AESState *ad = (AESState *)&d->ZMM_X(i);
2156 AESState *st = (AESState *)&v->ZMM_X(i);
2157 AESState *rk = (AESState *)&s->ZMM_X(i);
2158
2159 aesdec_ISB_ISR_AK(ad, st, rk, false);
2160 }
2161 }
2162
glue(helper_aesenc,SUFFIX)2163 void glue(helper_aesenc, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2164 {
2165 for (int i = 0; i < SHIFT; i++) {
2166 AESState *ad = (AESState *)&d->ZMM_X(i);
2167 AESState *st = (AESState *)&v->ZMM_X(i);
2168 AESState *rk = (AESState *)&s->ZMM_X(i);
2169
2170 aesenc_SB_SR_MC_AK(ad, st, rk, false);
2171 }
2172 }
2173
glue(helper_aesenclast,SUFFIX)2174 void glue(helper_aesenclast, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2175 {
2176 for (int i = 0; i < SHIFT; i++) {
2177 AESState *ad = (AESState *)&d->ZMM_X(i);
2178 AESState *st = (AESState *)&v->ZMM_X(i);
2179 AESState *rk = (AESState *)&s->ZMM_X(i);
2180
2181 aesenc_SB_SR_AK(ad, st, rk, false);
2182 }
2183 }
2184
2185 #if SHIFT == 1
glue(helper_aesimc,SUFFIX)2186 void glue(helper_aesimc, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
2187 {
2188 AESState *ad = (AESState *)&d->ZMM_X(0);
2189 AESState *st = (AESState *)&s->ZMM_X(0);
2190
2191 aesdec_IMC(ad, st, false);
2192 }
2193
glue(helper_aeskeygenassist,SUFFIX)2194 void glue(helper_aeskeygenassist, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2195 uint32_t ctrl)
2196 {
2197 int i;
2198 Reg tmp = *s;
2199
2200 for (i = 0 ; i < 4 ; i++) {
2201 d->B(i) = AES_sbox[tmp.B(i + 4)];
2202 d->B(i + 8) = AES_sbox[tmp.B(i + 12)];
2203 }
2204 d->L(1) = (d->L(0) << 24 | d->L(0) >> 8) ^ ctrl;
2205 d->L(3) = (d->L(2) << 24 | d->L(2) >> 8) ^ ctrl;
2206 }
2207 #endif
2208 #endif
2209
2210 #if SHIFT >= 1
glue(helper_vpermilpd,SUFFIX)2211 void glue(helper_vpermilpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2212 {
2213 uint64_t r0, r1;
2214 int i;
2215
2216 for (i = 0; i < 1 << SHIFT; i += 2) {
2217 r0 = v->Q(i + ((s->Q(i) >> 1) & 1));
2218 r1 = v->Q(i + ((s->Q(i+1) >> 1) & 1));
2219 d->Q(i) = r0;
2220 d->Q(i+1) = r1;
2221 }
2222 }
2223
glue(helper_vpermilps,SUFFIX)2224 void glue(helper_vpermilps, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2225 {
2226 uint32_t r0, r1, r2, r3;
2227 int i;
2228
2229 for (i = 0; i < 2 << SHIFT; i += 4) {
2230 r0 = v->L(i + (s->L(i) & 3));
2231 r1 = v->L(i + (s->L(i+1) & 3));
2232 r2 = v->L(i + (s->L(i+2) & 3));
2233 r3 = v->L(i + (s->L(i+3) & 3));
2234 d->L(i) = r0;
2235 d->L(i+1) = r1;
2236 d->L(i+2) = r2;
2237 d->L(i+3) = r3;
2238 }
2239 }
2240
glue(helper_vpermilpd_imm,SUFFIX)2241 void glue(helper_vpermilpd_imm, SUFFIX)(Reg *d, Reg *s, uint32_t order)
2242 {
2243 uint64_t r0, r1;
2244 int i;
2245
2246 for (i = 0; i < 1 << SHIFT; i += 2) {
2247 r0 = s->Q(i + ((order >> 0) & 1));
2248 r1 = s->Q(i + ((order >> 1) & 1));
2249 d->Q(i) = r0;
2250 d->Q(i+1) = r1;
2251
2252 order >>= 2;
2253 }
2254 }
2255
glue(helper_vpermilps_imm,SUFFIX)2256 void glue(helper_vpermilps_imm, SUFFIX)(Reg *d, Reg *s, uint32_t order)
2257 {
2258 uint32_t r0, r1, r2, r3;
2259 int i;
2260
2261 for (i = 0; i < 2 << SHIFT; i += 4) {
2262 r0 = s->L(i + ((order >> 0) & 3));
2263 r1 = s->L(i + ((order >> 2) & 3));
2264 r2 = s->L(i + ((order >> 4) & 3));
2265 r3 = s->L(i + ((order >> 6) & 3));
2266 d->L(i) = r0;
2267 d->L(i+1) = r1;
2268 d->L(i+2) = r2;
2269 d->L(i+3) = r3;
2270 }
2271 }
2272
2273 #if SHIFT == 1
2274 #define FPSRLVD(x, c) (c < 32 ? ((x) >> c) : 0)
2275 #define FPSRLVQ(x, c) (c < 64 ? ((x) >> c) : 0)
2276 #define FPSRAVD(x, c) ((int32_t)(x) >> (c < 32 ? c : 31))
2277 #define FPSRAVQ(x, c) ((int64_t)(x) >> (c < 64 ? c : 63))
2278 #define FPSLLVD(x, c) (c < 32 ? ((x) << c) : 0)
2279 #define FPSLLVQ(x, c) (c < 64 ? ((x) << c) : 0)
2280 #endif
2281
SSE_HELPER_L(helper_vpsrlvd,FPSRLVD)2282 SSE_HELPER_L(helper_vpsrlvd, FPSRLVD)
2283 SSE_HELPER_L(helper_vpsravd, FPSRAVD)
2284 SSE_HELPER_L(helper_vpsllvd, FPSLLVD)
2285
2286 SSE_HELPER_Q(helper_vpsrlvq, FPSRLVQ)
2287 SSE_HELPER_Q(helper_vpsravq, FPSRAVQ)
2288 SSE_HELPER_Q(helper_vpsllvq, FPSLLVQ)
2289
2290 void glue(helper_vtestps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
2291 {
2292 uint32_t zf = 0, cf = 0;
2293 int i;
2294
2295 for (i = 0; i < 2 << SHIFT; i++) {
2296 zf |= (s->L(i) & d->L(i));
2297 cf |= (s->L(i) & ~d->L(i));
2298 }
2299 CC_SRC = ((zf >> 31) ? 0 : CC_Z) | ((cf >> 31) ? 0 : CC_C);
2300 }
2301
glue(helper_vtestpd,SUFFIX)2302 void glue(helper_vtestpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
2303 {
2304 uint64_t zf = 0, cf = 0;
2305 int i;
2306
2307 for (i = 0; i < 1 << SHIFT; i++) {
2308 zf |= (s->Q(i) & d->Q(i));
2309 cf |= (s->Q(i) & ~d->Q(i));
2310 }
2311 CC_SRC = ((zf >> 63) ? 0 : CC_Z) | ((cf >> 63) ? 0 : CC_C);
2312 }
2313
glue(helper_vpmaskmovd_st,SUFFIX)2314 void glue(helper_vpmaskmovd_st, SUFFIX)(CPUX86State *env,
2315 Reg *v, Reg *s, target_ulong a0)
2316 {
2317 int i;
2318
2319 for (i = 0; i < (2 << SHIFT); i++) {
2320 if (v->L(i) >> 31) {
2321 cpu_stl_data_ra(env, a0 + i * 4, s->L(i), GETPC());
2322 }
2323 }
2324 }
2325
glue(helper_vpmaskmovq_st,SUFFIX)2326 void glue(helper_vpmaskmovq_st, SUFFIX)(CPUX86State *env,
2327 Reg *v, Reg *s, target_ulong a0)
2328 {
2329 int i;
2330
2331 for (i = 0; i < (1 << SHIFT); i++) {
2332 if (v->Q(i) >> 63) {
2333 cpu_stq_data_ra(env, a0 + i * 8, s->Q(i), GETPC());
2334 }
2335 }
2336 }
2337
glue(helper_vpmaskmovd,SUFFIX)2338 void glue(helper_vpmaskmovd, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2339 {
2340 int i;
2341
2342 for (i = 0; i < (2 << SHIFT); i++) {
2343 d->L(i) = (v->L(i) >> 31) ? s->L(i) : 0;
2344 }
2345 }
2346
glue(helper_vpmaskmovq,SUFFIX)2347 void glue(helper_vpmaskmovq, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2348 {
2349 int i;
2350
2351 for (i = 0; i < (1 << SHIFT); i++) {
2352 d->Q(i) = (v->Q(i) >> 63) ? s->Q(i) : 0;
2353 }
2354 }
2355
glue(helper_vpgatherdd,SUFFIX)2356 void glue(helper_vpgatherdd, SUFFIX)(CPUX86State *env,
2357 Reg *d, Reg *v, Reg *s, target_ulong a0, unsigned scale)
2358 {
2359 int i;
2360 for (i = 0; i < (2 << SHIFT); i++) {
2361 if (v->L(i) >> 31) {
2362 target_ulong addr = a0
2363 + ((target_ulong)(int32_t)s->L(i) << scale);
2364 d->L(i) = cpu_ldl_data_ra(env, addr, GETPC());
2365 }
2366 v->L(i) = 0;
2367 }
2368 }
2369
glue(helper_vpgatherdq,SUFFIX)2370 void glue(helper_vpgatherdq, SUFFIX)(CPUX86State *env,
2371 Reg *d, Reg *v, Reg *s, target_ulong a0, unsigned scale)
2372 {
2373 int i;
2374 for (i = 0; i < (1 << SHIFT); i++) {
2375 if (v->Q(i) >> 63) {
2376 target_ulong addr = a0
2377 + ((target_ulong)(int32_t)s->L(i) << scale);
2378 d->Q(i) = cpu_ldq_data_ra(env, addr, GETPC());
2379 }
2380 v->Q(i) = 0;
2381 }
2382 }
2383
glue(helper_vpgatherqd,SUFFIX)2384 void glue(helper_vpgatherqd, SUFFIX)(CPUX86State *env,
2385 Reg *d, Reg *v, Reg *s, target_ulong a0, unsigned scale)
2386 {
2387 int i;
2388 for (i = 0; i < (1 << SHIFT); i++) {
2389 if (v->L(i) >> 31) {
2390 target_ulong addr = a0
2391 + ((target_ulong)(int64_t)s->Q(i) << scale);
2392 d->L(i) = cpu_ldl_data_ra(env, addr, GETPC());
2393 }
2394 v->L(i) = 0;
2395 }
2396 for (i /= 2; i < 1 << SHIFT; i++) {
2397 d->Q(i) = 0;
2398 v->Q(i) = 0;
2399 }
2400 }
2401
glue(helper_vpgatherqq,SUFFIX)2402 void glue(helper_vpgatherqq, SUFFIX)(CPUX86State *env,
2403 Reg *d, Reg *v, Reg *s, target_ulong a0, unsigned scale)
2404 {
2405 int i;
2406 for (i = 0; i < (1 << SHIFT); i++) {
2407 if (v->Q(i) >> 63) {
2408 target_ulong addr = a0
2409 + ((target_ulong)(int64_t)s->Q(i) << scale);
2410 d->Q(i) = cpu_ldq_data_ra(env, addr, GETPC());
2411 }
2412 v->Q(i) = 0;
2413 }
2414 }
2415 #endif
2416
2417 #if SHIFT >= 2
helper_vpermdq_ymm(Reg * d,Reg * v,Reg * s,uint32_t order)2418 void helper_vpermdq_ymm(Reg *d, Reg *v, Reg *s, uint32_t order)
2419 {
2420 uint64_t r0, r1, r2, r3;
2421
2422 switch (order & 3) {
2423 case 0:
2424 r0 = v->Q(0);
2425 r1 = v->Q(1);
2426 break;
2427 case 1:
2428 r0 = v->Q(2);
2429 r1 = v->Q(3);
2430 break;
2431 case 2:
2432 r0 = s->Q(0);
2433 r1 = s->Q(1);
2434 break;
2435 case 3:
2436 r0 = s->Q(2);
2437 r1 = s->Q(3);
2438 break;
2439 default: /* default case added to help the compiler to avoid warnings */
2440 g_assert_not_reached();
2441 }
2442 switch ((order >> 4) & 3) {
2443 case 0:
2444 r2 = v->Q(0);
2445 r3 = v->Q(1);
2446 break;
2447 case 1:
2448 r2 = v->Q(2);
2449 r3 = v->Q(3);
2450 break;
2451 case 2:
2452 r2 = s->Q(0);
2453 r3 = s->Q(1);
2454 break;
2455 case 3:
2456 r2 = s->Q(2);
2457 r3 = s->Q(3);
2458 break;
2459 default: /* default case added to help the compiler to avoid warnings */
2460 g_assert_not_reached();
2461 }
2462 d->Q(0) = r0;
2463 d->Q(1) = r1;
2464 d->Q(2) = r2;
2465 d->Q(3) = r3;
2466 if (order & 0x8) {
2467 d->Q(0) = 0;
2468 d->Q(1) = 0;
2469 }
2470 if (order & 0x80) {
2471 d->Q(2) = 0;
2472 d->Q(3) = 0;
2473 }
2474 }
2475
helper_vpermq_ymm(Reg * d,Reg * s,uint32_t order)2476 void helper_vpermq_ymm(Reg *d, Reg *s, uint32_t order)
2477 {
2478 uint64_t r0, r1, r2, r3;
2479 r0 = s->Q(order & 3);
2480 r1 = s->Q((order >> 2) & 3);
2481 r2 = s->Q((order >> 4) & 3);
2482 r3 = s->Q((order >> 6) & 3);
2483 d->Q(0) = r0;
2484 d->Q(1) = r1;
2485 d->Q(2) = r2;
2486 d->Q(3) = r3;
2487 }
2488
helper_vpermd_ymm(Reg * d,Reg * v,Reg * s)2489 void helper_vpermd_ymm(Reg *d, Reg *v, Reg *s)
2490 {
2491 uint32_t r[8];
2492 int i;
2493
2494 for (i = 0; i < 8; i++) {
2495 r[i] = s->L(v->L(i) & 7);
2496 }
2497 for (i = 0; i < 8; i++) {
2498 d->L(i) = r[i];
2499 }
2500 }
2501 #endif
2502
2503 /* FMA3 op helpers */
2504 #if SHIFT == 1
2505 #define SSE_HELPER_FMAS(name, elem, F) \
2506 void name(CPUX86State *env, Reg *d, Reg *a, Reg *b, Reg *c, int flags) \
2507 { \
2508 d->elem(0) = F(a->elem(0), b->elem(0), c->elem(0), flags, &env->sse_status); \
2509 }
2510 #define SSE_HELPER_FMAP(name, elem, num, F) \
2511 void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *a, Reg *b, Reg *c, \
2512 int flags, int flip) \
2513 { \
2514 int i; \
2515 for (i = 0; i < num; i++) { \
2516 d->elem(i) = F(a->elem(i), b->elem(i), c->elem(i), flags, &env->sse_status); \
2517 flags ^= flip; \
2518 } \
2519 }
2520
SSE_HELPER_FMAS(helper_fma4ss,ZMM_S,float32_muladd)2521 SSE_HELPER_FMAS(helper_fma4ss, ZMM_S, float32_muladd)
2522 SSE_HELPER_FMAS(helper_fma4sd, ZMM_D, float64_muladd)
2523 #endif
2524
2525 #if SHIFT >= 1
2526 SSE_HELPER_FMAP(helper_fma4ps, ZMM_S, 2 << SHIFT, float32_muladd)
2527 SSE_HELPER_FMAP(helper_fma4pd, ZMM_D, 1 << SHIFT, float64_muladd)
2528 #endif
2529
2530 #if SHIFT == 1
2531 #define SSE_HELPER_SHA1RNDS4(name, F, K) \
2532 void name(Reg *d, Reg *a, Reg *b) \
2533 { \
2534 uint32_t A, B, C, D, E, t, i; \
2535 \
2536 A = a->L(3); \
2537 B = a->L(2); \
2538 C = a->L(1); \
2539 D = a->L(0); \
2540 E = 0; \
2541 \
2542 for (i = 0; i <= 3; i++) { \
2543 t = F(B, C, D) + rol32(A, 5) + b->L(3 - i) + E + K; \
2544 E = D; \
2545 D = C; \
2546 C = rol32(B, 30); \
2547 B = A; \
2548 A = t; \
2549 } \
2550 \
2551 d->L(3) = A; \
2552 d->L(2) = B; \
2553 d->L(1) = C; \
2554 d->L(0) = D; \
2555 }
2556
2557 #define SHA1_F0(b, c, d) (((b) & (c)) ^ (~(b) & (d)))
2558 #define SHA1_F1(b, c, d) ((b) ^ (c) ^ (d))
2559 #define SHA1_F2(b, c, d) (((b) & (c)) ^ ((b) & (d)) ^ ((c) & (d)))
2560
2561 SSE_HELPER_SHA1RNDS4(helper_sha1rnds4_f0, SHA1_F0, 0x5A827999)
2562 SSE_HELPER_SHA1RNDS4(helper_sha1rnds4_f1, SHA1_F1, 0x6ED9EBA1)
2563 SSE_HELPER_SHA1RNDS4(helper_sha1rnds4_f2, SHA1_F2, 0x8F1BBCDC)
2564 SSE_HELPER_SHA1RNDS4(helper_sha1rnds4_f3, SHA1_F1, 0xCA62C1D6)
2565
2566 void helper_sha1nexte(Reg *d, Reg *a, Reg *b)
2567 {
2568 d->L(3) = b->L(3) + rol32(a->L(3), 30);
2569 d->L(2) = b->L(2);
2570 d->L(1) = b->L(1);
2571 d->L(0) = b->L(0);
2572 }
2573
helper_sha1msg1(Reg * d,Reg * a,Reg * b)2574 void helper_sha1msg1(Reg *d, Reg *a, Reg *b)
2575 {
2576 /* These could be overwritten by the first two assignments, save them. */
2577 uint32_t b3 = b->L(3);
2578 uint32_t b2 = b->L(2);
2579
2580 d->L(3) = a->L(3) ^ a->L(1);
2581 d->L(2) = a->L(2) ^ a->L(0);
2582 d->L(1) = a->L(1) ^ b3;
2583 d->L(0) = a->L(0) ^ b2;
2584 }
2585
helper_sha1msg2(Reg * d,Reg * a,Reg * b)2586 void helper_sha1msg2(Reg *d, Reg *a, Reg *b)
2587 {
2588 d->L(3) = rol32(a->L(3) ^ b->L(2), 1);
2589 d->L(2) = rol32(a->L(2) ^ b->L(1), 1);
2590 d->L(1) = rol32(a->L(1) ^ b->L(0), 1);
2591 d->L(0) = rol32(a->L(0) ^ d->L(3), 1);
2592 }
2593
2594 #define SHA256_CH(e, f, g) (((e) & (f)) ^ (~(e) & (g)))
2595 #define SHA256_MAJ(a, b, c) (((a) & (b)) ^ ((a) & (c)) ^ ((b) & (c)))
2596
2597 #define SHA256_RNDS0(w) (ror32((w), 2) ^ ror32((w), 13) ^ ror32((w), 22))
2598 #define SHA256_RNDS1(w) (ror32((w), 6) ^ ror32((w), 11) ^ ror32((w), 25))
2599 #define SHA256_MSGS0(w) (ror32((w), 7) ^ ror32((w), 18) ^ ((w) >> 3))
2600 #define SHA256_MSGS1(w) (ror32((w), 17) ^ ror32((w), 19) ^ ((w) >> 10))
2601
helper_sha256rnds2(Reg * d,Reg * a,Reg * b,uint32_t wk0,uint32_t wk1)2602 void helper_sha256rnds2(Reg *d, Reg *a, Reg *b, uint32_t wk0, uint32_t wk1)
2603 {
2604 uint32_t t, AA, EE;
2605
2606 uint32_t A = b->L(3);
2607 uint32_t B = b->L(2);
2608 uint32_t C = a->L(3);
2609 uint32_t D = a->L(2);
2610 uint32_t E = b->L(1);
2611 uint32_t F = b->L(0);
2612 uint32_t G = a->L(1);
2613 uint32_t H = a->L(0);
2614
2615 /* Even round */
2616 t = SHA256_CH(E, F, G) + SHA256_RNDS1(E) + wk0 + H;
2617 AA = t + SHA256_MAJ(A, B, C) + SHA256_RNDS0(A);
2618 EE = t + D;
2619
2620 /* These will be B and F at the end of the odd round */
2621 d->L(2) = AA;
2622 d->L(0) = EE;
2623
2624 D = C, C = B, B = A, A = AA;
2625 H = G, G = F, F = E, E = EE;
2626
2627 /* Odd round */
2628 t = SHA256_CH(E, F, G) + SHA256_RNDS1(E) + wk1 + H;
2629 AA = t + SHA256_MAJ(A, B, C) + SHA256_RNDS0(A);
2630 EE = t + D;
2631
2632 d->L(3) = AA;
2633 d->L(1) = EE;
2634 }
2635
helper_sha256msg1(Reg * d,Reg * a,Reg * b)2636 void helper_sha256msg1(Reg *d, Reg *a, Reg *b)
2637 {
2638 /* b->L(0) could be overwritten by the first assignment, save it. */
2639 uint32_t b0 = b->L(0);
2640
2641 d->L(0) = a->L(0) + SHA256_MSGS0(a->L(1));
2642 d->L(1) = a->L(1) + SHA256_MSGS0(a->L(2));
2643 d->L(2) = a->L(2) + SHA256_MSGS0(a->L(3));
2644 d->L(3) = a->L(3) + SHA256_MSGS0(b0);
2645 }
2646
helper_sha256msg2(Reg * d,Reg * a,Reg * b)2647 void helper_sha256msg2(Reg *d, Reg *a, Reg *b)
2648 {
2649 /* Earlier assignments cannot overwrite any of the two operands. */
2650 d->L(0) = a->L(0) + SHA256_MSGS1(b->L(2));
2651 d->L(1) = a->L(1) + SHA256_MSGS1(b->L(3));
2652 /* Yes, this reuses the previously computed values. */
2653 d->L(2) = a->L(2) + SHA256_MSGS1(d->L(0));
2654 d->L(3) = a->L(3) + SHA256_MSGS1(d->L(1));
2655 }
2656 #endif
2657
2658 #undef SSE_HELPER_S
2659
2660 #undef LANE_WIDTH
2661 #undef SHIFT
2662 #undef XMM_ONLY
2663 #undef Reg
2664 #undef B
2665 #undef W
2666 #undef L
2667 #undef Q
2668 #undef SUFFIX
2669