xref: /openbmc/qemu/target/i386/ops_sse.h (revision e052944a)
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 
24 #if SHIFT == 0
25 #define Reg MMXReg
26 #define XMM_ONLY(...)
27 #define B(n) MMX_B(n)
28 #define W(n) MMX_W(n)
29 #define L(n) MMX_L(n)
30 #define Q(n) MMX_Q(n)
31 #define SUFFIX _mmx
32 #else
33 #define Reg ZMMReg
34 #define XMM_ONLY(...) __VA_ARGS__
35 #define B(n) ZMM_B(n)
36 #define W(n) ZMM_W(n)
37 #define L(n) ZMM_L(n)
38 #define Q(n) ZMM_Q(n)
39 #if SHIFT == 1
40 #define SUFFIX _xmm
41 #else
42 #define SUFFIX _ymm
43 #endif
44 #endif
45 
46 #define LANE_WIDTH (SHIFT ? 16 : 8)
47 #define PACK_WIDTH (LANE_WIDTH / 2)
48 
49 #if SHIFT == 0
50 #define FPSRL(x, c) ((x) >> shift)
51 #define FPSRAW(x, c) ((int16_t)(x) >> shift)
52 #define FPSRAL(x, c) ((int32_t)(x) >> shift)
53 #define FPSLL(x, c) ((x) << shift)
54 #endif
55 
56 void glue(helper_psrlw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
57 {
58     int shift;
59     if (c->Q(0) > 15) {
60         for (int i = 0; i < 1 << SHIFT; i++) {
61             d->Q(i) = 0;
62         }
63     } else {
64         shift = c->B(0);
65         for (int i = 0; i < 4 << SHIFT; i++) {
66             d->W(i) = FPSRL(s->W(i), shift);
67         }
68     }
69 }
70 
71 void glue(helper_psllw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
72 {
73     int shift;
74     if (c->Q(0) > 15) {
75         for (int i = 0; i < 1 << SHIFT; i++) {
76             d->Q(i) = 0;
77         }
78     } else {
79         shift = c->B(0);
80         for (int i = 0; i < 4 << SHIFT; i++) {
81             d->W(i) = FPSLL(s->W(i), shift);
82         }
83     }
84 }
85 
86 void glue(helper_psraw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
87 {
88     int shift;
89     if (c->Q(0) > 15) {
90         shift = 15;
91     } else {
92         shift = c->B(0);
93     }
94     for (int i = 0; i < 4 << SHIFT; i++) {
95         d->W(i) = FPSRAW(s->W(i), shift);
96     }
97 }
98 
99 void glue(helper_psrld, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
100 {
101     int shift;
102     if (c->Q(0) > 31) {
103         for (int i = 0; i < 1 << SHIFT; i++) {
104             d->Q(i) = 0;
105         }
106     } else {
107         shift = c->B(0);
108         for (int i = 0; i < 2 << SHIFT; i++) {
109             d->L(i) = FPSRL(s->L(i), shift);
110         }
111     }
112 }
113 
114 void glue(helper_pslld, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
115 {
116     int shift;
117     if (c->Q(0) > 31) {
118         for (int i = 0; i < 1 << SHIFT; i++) {
119             d->Q(i) = 0;
120         }
121     } else {
122         shift = c->B(0);
123         for (int i = 0; i < 2 << SHIFT; i++) {
124             d->L(i) = FPSLL(s->L(i), shift);
125         }
126     }
127 }
128 
129 void glue(helper_psrad, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
130 {
131     int shift;
132     if (c->Q(0) > 31) {
133         shift = 31;
134     } else {
135         shift = c->B(0);
136     }
137     for (int i = 0; i < 2 << SHIFT; i++) {
138         d->L(i) = FPSRAL(s->L(i), shift);
139     }
140 }
141 
142 void glue(helper_psrlq, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
143 {
144     int shift;
145     if (c->Q(0) > 63) {
146         for (int i = 0; i < 1 << SHIFT; i++) {
147             d->Q(i) = 0;
148         }
149     } else {
150         shift = c->B(0);
151         for (int i = 0; i < 1 << SHIFT; i++) {
152             d->Q(i) = FPSRL(s->Q(i), shift);
153         }
154     }
155 }
156 
157 void glue(helper_psllq, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
158 {
159     int shift;
160     if (c->Q(0) > 63) {
161         for (int i = 0; i < 1 << SHIFT; i++) {
162             d->Q(i) = 0;
163         }
164     } else {
165         shift = c->B(0);
166         for (int i = 0; i < 1 << SHIFT; i++) {
167             d->Q(i) = FPSLL(s->Q(i), shift);
168         }
169     }
170 }
171 
172 #if SHIFT >= 1
173 void glue(helper_psrldq, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
174 {
175     int shift, i, j;
176 
177     shift = c->L(0);
178     if (shift > 16) {
179         shift = 16;
180     }
181     for (j = 0; j < 8 << SHIFT; j += LANE_WIDTH) {
182         for (i = 0; i < 16 - shift; i++) {
183             d->B(j + i) = s->B(j + i + shift);
184         }
185         for (i = 16 - shift; i < 16; i++) {
186             d->B(j + i) = 0;
187         }
188     }
189 }
190 
191 void glue(helper_pslldq, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
192 {
193     int shift, i, j;
194 
195     shift = c->L(0);
196     if (shift > 16) {
197         shift = 16;
198     }
199     for (j = 0; j < 8 << SHIFT; j += LANE_WIDTH) {
200         for (i = 15; i >= shift; i--) {
201             d->B(j + i) = s->B(j + i - shift);
202         }
203         for (i = 0; i < shift; i++) {
204             d->B(j + i) = 0;
205         }
206     }
207 }
208 #endif
209 
210 #define SSE_HELPER_1(name, elem, num, F)                        \
211     void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)   \
212     {                                                           \
213         int n = num;                                            \
214         for (int i = 0; i < n; i++) {                           \
215             d->elem(i) = F(s->elem(i));                         \
216         }                                                       \
217     }
218 
219 #define SSE_HELPER_2(name, elem, num, F)                        \
220     void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)   \
221     {                                                           \
222         int n = num;                                            \
223         for (int i = 0; i < n; i++) {                           \
224             d->elem(i) = F(v->elem(i), s->elem(i));             \
225         }                                                       \
226     }
227 
228 #define SSE_HELPER_B(name, F)                                   \
229     SSE_HELPER_2(name, B, 8 << SHIFT, F)
230 
231 #define SSE_HELPER_W(name, F)                                   \
232     SSE_HELPER_2(name, W, 4 << SHIFT, F)
233 
234 #define SSE_HELPER_L(name, F)                                   \
235     SSE_HELPER_2(name, L, 2 << SHIFT, F)
236 
237 #define SSE_HELPER_Q(name, F)                                   \
238     SSE_HELPER_2(name, Q, 1 << SHIFT, F)
239 
240 #if SHIFT == 0
241 static inline int satub(int x)
242 {
243     if (x < 0) {
244         return 0;
245     } else if (x > 255) {
246         return 255;
247     } else {
248         return x;
249     }
250 }
251 
252 static inline int satuw(int x)
253 {
254     if (x < 0) {
255         return 0;
256     } else if (x > 65535) {
257         return 65535;
258     } else {
259         return x;
260     }
261 }
262 
263 static inline int satsb(int x)
264 {
265     if (x < -128) {
266         return -128;
267     } else if (x > 127) {
268         return 127;
269     } else {
270         return x;
271     }
272 }
273 
274 static inline int satsw(int x)
275 {
276     if (x < -32768) {
277         return -32768;
278     } else if (x > 32767) {
279         return 32767;
280     } else {
281         return x;
282     }
283 }
284 
285 #define FADD(a, b) ((a) + (b))
286 #define FADDUB(a, b) satub((a) + (b))
287 #define FADDUW(a, b) satuw((a) + (b))
288 #define FADDSB(a, b) satsb((int8_t)(a) + (int8_t)(b))
289 #define FADDSW(a, b) satsw((int16_t)(a) + (int16_t)(b))
290 
291 #define FSUB(a, b) ((a) - (b))
292 #define FSUBUB(a, b) satub((a) - (b))
293 #define FSUBUW(a, b) satuw((a) - (b))
294 #define FSUBSB(a, b) satsb((int8_t)(a) - (int8_t)(b))
295 #define FSUBSW(a, b) satsw((int16_t)(a) - (int16_t)(b))
296 #define FMINUB(a, b) ((a) < (b)) ? (a) : (b)
297 #define FMINSW(a, b) ((int16_t)(a) < (int16_t)(b)) ? (a) : (b)
298 #define FMAXUB(a, b) ((a) > (b)) ? (a) : (b)
299 #define FMAXSW(a, b) ((int16_t)(a) > (int16_t)(b)) ? (a) : (b)
300 
301 #define FMULHRW(a, b) (((int16_t)(a) * (int16_t)(b) + 0x8000) >> 16)
302 #define FMULHUW(a, b) ((a) * (b) >> 16)
303 #define FMULHW(a, b) ((int16_t)(a) * (int16_t)(b) >> 16)
304 
305 #define FAVG(a, b) (((a) + (b) + 1) >> 1)
306 #endif
307 
308 SSE_HELPER_W(helper_pmulhuw, FMULHUW)
309 SSE_HELPER_W(helper_pmulhw, FMULHW)
310 
311 #if SHIFT == 0
312 void glue(helper_pmulhrw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
313 {
314     d->W(0) = FMULHRW(d->W(0), s->W(0));
315     d->W(1) = FMULHRW(d->W(1), s->W(1));
316     d->W(2) = FMULHRW(d->W(2), s->W(2));
317     d->W(3) = FMULHRW(d->W(3), s->W(3));
318 }
319 #endif
320 
321 SSE_HELPER_B(helper_pavgb, FAVG)
322 SSE_HELPER_W(helper_pavgw, FAVG)
323 
324 void glue(helper_pmuludq, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
325 {
326     int i;
327 
328     for (i = 0; i < (1 << SHIFT); i++) {
329         d->Q(i) = (uint64_t)s->L(i * 2) * (uint64_t)v->L(i * 2);
330     }
331 }
332 
333 void glue(helper_pmaddwd, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
334 {
335     int i;
336 
337     for (i = 0; i < (2 << SHIFT); i++) {
338         d->L(i) = (int16_t)s->W(2 * i) * (int16_t)v->W(2 * i) +
339             (int16_t)s->W(2 * i + 1) * (int16_t)v->W(2 * i + 1);
340     }
341 }
342 
343 #if SHIFT == 0
344 static inline int abs1(int a)
345 {
346     if (a < 0) {
347         return -a;
348     } else {
349         return a;
350     }
351 }
352 #endif
353 void glue(helper_psadbw, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
354 {
355     int i;
356 
357     for (i = 0; i < (1 << SHIFT); i++) {
358         unsigned int val = 0;
359         val += abs1(v->B(8 * i + 0) - s->B(8 * i + 0));
360         val += abs1(v->B(8 * i + 1) - s->B(8 * i + 1));
361         val += abs1(v->B(8 * i + 2) - s->B(8 * i + 2));
362         val += abs1(v->B(8 * i + 3) - s->B(8 * i + 3));
363         val += abs1(v->B(8 * i + 4) - s->B(8 * i + 4));
364         val += abs1(v->B(8 * i + 5) - s->B(8 * i + 5));
365         val += abs1(v->B(8 * i + 6) - s->B(8 * i + 6));
366         val += abs1(v->B(8 * i + 7) - s->B(8 * i + 7));
367         d->Q(i) = val;
368     }
369 }
370 
371 #if SHIFT < 2
372 void glue(helper_maskmov, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
373                                   target_ulong a0)
374 {
375     int i;
376 
377     for (i = 0; i < (8 << SHIFT); i++) {
378         if (s->B(i) & 0x80) {
379             cpu_stb_data_ra(env, a0 + i, d->B(i), GETPC());
380         }
381     }
382 }
383 #endif
384 
385 #define SHUFFLE4(F, a, b, offset) do {      \
386     r0 = a->F((order & 3) + offset);        \
387     r1 = a->F(((order >> 2) & 3) + offset); \
388     r2 = b->F(((order >> 4) & 3) + offset); \
389     r3 = b->F(((order >> 6) & 3) + offset); \
390     d->F(offset) = r0;                      \
391     d->F(offset + 1) = r1;                  \
392     d->F(offset + 2) = r2;                  \
393     d->F(offset + 3) = r3;                  \
394     } while (0)
395 
396 #if SHIFT == 0
397 void glue(helper_pshufw, SUFFIX)(Reg *d, Reg *s, int order)
398 {
399     uint16_t r0, r1, r2, r3;
400 
401     SHUFFLE4(W, s, s, 0);
402 }
403 #else
404 void glue(helper_shufps, SUFFIX)(Reg *d, Reg *v, Reg *s, int order)
405 {
406     uint32_t r0, r1, r2, r3;
407     int i;
408 
409     for (i = 0; i < 2 << SHIFT; i += 4) {
410         SHUFFLE4(L, v, s, i);
411     }
412 }
413 
414 void glue(helper_shufpd, SUFFIX)(Reg *d, Reg *v, Reg *s, int order)
415 {
416     uint64_t r0, r1;
417     int i;
418 
419     for (i = 0; i < 1 << SHIFT; i += 2) {
420         r0 = v->Q(((order & 1) & 1) + i);
421         r1 = s->Q(((order >> 1) & 1) + i);
422         d->Q(i) = r0;
423         d->Q(i + 1) = r1;
424         order >>= 2;
425     }
426 }
427 
428 void glue(helper_pshufd, SUFFIX)(Reg *d, Reg *s, int order)
429 {
430     uint32_t r0, r1, r2, r3;
431     int i;
432 
433     for (i = 0; i < 2 << SHIFT; i += 4) {
434         SHUFFLE4(L, s, s, i);
435     }
436 }
437 
438 void glue(helper_pshuflw, SUFFIX)(Reg *d, Reg *s, int order)
439 {
440     uint16_t r0, r1, r2, r3;
441     int i, j;
442 
443     for (i = 0, j = 1; j < 1 << SHIFT; i += 8, j += 2) {
444         SHUFFLE4(W, s, s, i);
445         d->Q(j) = s->Q(j);
446     }
447 }
448 
449 void glue(helper_pshufhw, SUFFIX)(Reg *d, Reg *s, int order)
450 {
451     uint16_t r0, r1, r2, r3;
452     int i, j;
453 
454     for (i = 4, j = 0; j < 1 << SHIFT; i += 8, j += 2) {
455         d->Q(j) = s->Q(j);
456         SHUFFLE4(W, s, s, i);
457     }
458 }
459 #endif
460 
461 #if SHIFT >= 1
462 /* FPU ops */
463 /* XXX: not accurate */
464 
465 #define SSE_HELPER_P(name, F)                                           \
466     void glue(helper_ ## name ## ps, SUFFIX)(CPUX86State *env,          \
467             Reg *d, Reg *v, Reg *s)                                     \
468     {                                                                   \
469         int i;                                                          \
470         for (i = 0; i < 2 << SHIFT; i++) {                              \
471             d->ZMM_S(i) = F(32, v->ZMM_S(i), s->ZMM_S(i));              \
472         }                                                               \
473     }                                                                   \
474                                                                         \
475     void glue(helper_ ## name ## pd, SUFFIX)(CPUX86State *env,          \
476             Reg *d, Reg *v, Reg *s)                                     \
477     {                                                                   \
478         int i;                                                          \
479         for (i = 0; i < 1 << SHIFT; i++) {                              \
480             d->ZMM_D(i) = F(64, v->ZMM_D(i), s->ZMM_D(i));              \
481         }                                                               \
482     }
483 
484 #if SHIFT == 1
485 
486 #define SSE_HELPER_S(name, F)                                           \
487     SSE_HELPER_P(name, F)                                               \
488                                                                         \
489     void helper_ ## name ## ss(CPUX86State *env, Reg *d, Reg *v, Reg *s)\
490     {                                                                   \
491         int i;                                                          \
492         d->ZMM_S(0) = F(32, v->ZMM_S(0), s->ZMM_S(0));                  \
493         for (i = 1; i < 2 << SHIFT; i++) {                              \
494             d->ZMM_L(i) = v->ZMM_L(i);                                  \
495         }                                                               \
496     }                                                                   \
497                                                                         \
498     void helper_ ## name ## sd(CPUX86State *env, Reg *d, Reg *v, Reg *s)\
499     {                                                                   \
500         int i;                                                          \
501         d->ZMM_D(0) = F(64, v->ZMM_D(0), s->ZMM_D(0));                  \
502         for (i = 1; i < 1 << SHIFT; i++) {                              \
503             d->ZMM_Q(i) = v->ZMM_Q(i);                                  \
504         }                                                               \
505     }
506 
507 #else
508 
509 #define SSE_HELPER_S(name, F) SSE_HELPER_P(name, F)
510 
511 #endif
512 
513 #define FPU_ADD(size, a, b) float ## size ## _add(a, b, &env->sse_status)
514 #define FPU_SUB(size, a, b) float ## size ## _sub(a, b, &env->sse_status)
515 #define FPU_MUL(size, a, b) float ## size ## _mul(a, b, &env->sse_status)
516 #define FPU_DIV(size, a, b) float ## size ## _div(a, b, &env->sse_status)
517 
518 /* Note that the choice of comparison op here is important to get the
519  * special cases right: for min and max Intel specifies that (-0,0),
520  * (NaN, anything) and (anything, NaN) return the second argument.
521  */
522 #define FPU_MIN(size, a, b)                                     \
523     (float ## size ## _lt(a, b, &env->sse_status) ? (a) : (b))
524 #define FPU_MAX(size, a, b)                                     \
525     (float ## size ## _lt(b, a, &env->sse_status) ? (a) : (b))
526 
527 SSE_HELPER_S(add, FPU_ADD)
528 SSE_HELPER_S(sub, FPU_SUB)
529 SSE_HELPER_S(mul, FPU_MUL)
530 SSE_HELPER_S(div, FPU_DIV)
531 SSE_HELPER_S(min, FPU_MIN)
532 SSE_HELPER_S(max, FPU_MAX)
533 
534 void glue(helper_sqrtps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
535 {
536     int i;
537     for (i = 0; i < 2 << SHIFT; i++) {
538         d->ZMM_S(i) = float32_sqrt(s->ZMM_S(i), &env->sse_status);
539     }
540 }
541 
542 void glue(helper_sqrtpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
543 {
544     int i;
545     for (i = 0; i < 1 << SHIFT; i++) {
546         d->ZMM_D(i) = float64_sqrt(s->ZMM_D(i), &env->sse_status);
547     }
548 }
549 
550 #if SHIFT == 1
551 void helper_sqrtss(CPUX86State *env, Reg *d, Reg *v, Reg *s)
552 {
553     int i;
554     d->ZMM_S(0) = float32_sqrt(s->ZMM_S(0), &env->sse_status);
555     for (i = 1; i < 2 << SHIFT; i++) {
556         d->ZMM_L(i) = v->ZMM_L(i);
557     }
558 }
559 
560 void helper_sqrtsd(CPUX86State *env, Reg *d, Reg *v, Reg *s)
561 {
562     int i;
563     d->ZMM_D(0) = float64_sqrt(s->ZMM_D(0), &env->sse_status);
564     for (i = 1; i < 1 << SHIFT; i++) {
565         d->ZMM_Q(i) = v->ZMM_Q(i);
566     }
567 }
568 #endif
569 
570 /* float to float conversions */
571 void glue(helper_cvtps2pd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
572 {
573     int i;
574     for (i = 1 << SHIFT; --i >= 0; ) {
575         d->ZMM_D(i) = float32_to_float64(s->ZMM_S(i), &env->sse_status);
576     }
577 }
578 
579 void glue(helper_cvtpd2ps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
580 {
581     int i;
582     for (i = 0; i < 1 << SHIFT; i++) {
583          d->ZMM_S(i) = float64_to_float32(s->ZMM_D(i), &env->sse_status);
584     }
585     for (i >>= 1; i < 1 << SHIFT; i++) {
586          d->Q(i) = 0;
587     }
588 }
589 
590 #if SHIFT >= 1
591 void glue(helper_cvtph2ps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
592 {
593     int i;
594 
595     for (i = 2 << SHIFT; --i >= 0; ) {
596          d->ZMM_S(i) = float16_to_float32(s->ZMM_H(i), true, &env->sse_status);
597     }
598 }
599 
600 void glue(helper_cvtps2ph, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, int mode)
601 {
602     int i;
603     FloatRoundMode prev_rounding_mode = env->sse_status.float_rounding_mode;
604     if (!(mode & (1 << 2))) {
605         set_x86_rounding_mode(mode & 3, &env->sse_status);
606     }
607 
608     for (i = 0; i < 2 << SHIFT; i++) {
609         d->ZMM_H(i) = float32_to_float16(s->ZMM_S(i), true, &env->sse_status);
610     }
611     for (i >>= 2; i < 1 << SHIFT; i++) {
612         d->Q(i) = 0;
613     }
614 
615     env->sse_status.float_rounding_mode = prev_rounding_mode;
616 }
617 #endif
618 
619 #if SHIFT == 1
620 void helper_cvtss2sd(CPUX86State *env, Reg *d, Reg *v, Reg *s)
621 {
622     int i;
623     d->ZMM_D(0) = float32_to_float64(s->ZMM_S(0), &env->sse_status);
624     for (i = 1; i < 1 << SHIFT; i++) {
625         d->ZMM_Q(i) = v->ZMM_Q(i);
626     }
627 }
628 
629 void helper_cvtsd2ss(CPUX86State *env, Reg *d, Reg *v, Reg *s)
630 {
631     int i;
632     d->ZMM_S(0) = float64_to_float32(s->ZMM_D(0), &env->sse_status);
633     for (i = 1; i < 2 << SHIFT; i++) {
634         d->ZMM_L(i) = v->ZMM_L(i);
635     }
636 }
637 #endif
638 
639 /* integer to float */
640 void glue(helper_cvtdq2ps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
641 {
642     int i;
643     for (i = 0; i < 2 << SHIFT; i++) {
644         d->ZMM_S(i) = int32_to_float32(s->ZMM_L(i), &env->sse_status);
645     }
646 }
647 
648 void glue(helper_cvtdq2pd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
649 {
650     int i;
651     for (i = 1 << SHIFT; --i >= 0; ) {
652         int32_t l = s->ZMM_L(i);
653         d->ZMM_D(i) = int32_to_float64(l, &env->sse_status);
654     }
655 }
656 
657 #if SHIFT == 1
658 void helper_cvtpi2ps(CPUX86State *env, ZMMReg *d, MMXReg *s)
659 {
660     d->ZMM_S(0) = int32_to_float32(s->MMX_L(0), &env->sse_status);
661     d->ZMM_S(1) = int32_to_float32(s->MMX_L(1), &env->sse_status);
662 }
663 
664 void helper_cvtpi2pd(CPUX86State *env, ZMMReg *d, MMXReg *s)
665 {
666     d->ZMM_D(0) = int32_to_float64(s->MMX_L(0), &env->sse_status);
667     d->ZMM_D(1) = int32_to_float64(s->MMX_L(1), &env->sse_status);
668 }
669 
670 void helper_cvtsi2ss(CPUX86State *env, ZMMReg *d, uint32_t val)
671 {
672     d->ZMM_S(0) = int32_to_float32(val, &env->sse_status);
673 }
674 
675 void helper_cvtsi2sd(CPUX86State *env, ZMMReg *d, uint32_t val)
676 {
677     d->ZMM_D(0) = int32_to_float64(val, &env->sse_status);
678 }
679 
680 #ifdef TARGET_X86_64
681 void helper_cvtsq2ss(CPUX86State *env, ZMMReg *d, uint64_t val)
682 {
683     d->ZMM_S(0) = int64_to_float32(val, &env->sse_status);
684 }
685 
686 void helper_cvtsq2sd(CPUX86State *env, ZMMReg *d, uint64_t val)
687 {
688     d->ZMM_D(0) = int64_to_float64(val, &env->sse_status);
689 }
690 #endif
691 
692 #endif
693 
694 /* float to integer */
695 
696 #if SHIFT == 1
697 /*
698  * x86 mandates that we return the indefinite integer value for the result
699  * of any float-to-integer conversion that raises the 'invalid' exception.
700  * Wrap the softfloat functions to get this behaviour.
701  */
702 #define WRAP_FLOATCONV(RETTYPE, FN, FLOATTYPE, INDEFVALUE)              \
703     static inline RETTYPE x86_##FN(FLOATTYPE a, float_status *s)        \
704     {                                                                   \
705         int oldflags, newflags;                                         \
706         RETTYPE r;                                                      \
707                                                                         \
708         oldflags = get_float_exception_flags(s);                        \
709         set_float_exception_flags(0, s);                                \
710         r = FN(a, s);                                                   \
711         newflags = get_float_exception_flags(s);                        \
712         if (newflags & float_flag_invalid) {                            \
713             r = INDEFVALUE;                                             \
714         }                                                               \
715         set_float_exception_flags(newflags | oldflags, s);              \
716         return r;                                                       \
717     }
718 
719 WRAP_FLOATCONV(int32_t, float32_to_int32, float32, INT32_MIN)
720 WRAP_FLOATCONV(int32_t, float32_to_int32_round_to_zero, float32, INT32_MIN)
721 WRAP_FLOATCONV(int32_t, float64_to_int32, float64, INT32_MIN)
722 WRAP_FLOATCONV(int32_t, float64_to_int32_round_to_zero, float64, INT32_MIN)
723 WRAP_FLOATCONV(int64_t, float32_to_int64, float32, INT64_MIN)
724 WRAP_FLOATCONV(int64_t, float32_to_int64_round_to_zero, float32, INT64_MIN)
725 WRAP_FLOATCONV(int64_t, float64_to_int64, float64, INT64_MIN)
726 WRAP_FLOATCONV(int64_t, float64_to_int64_round_to_zero, float64, INT64_MIN)
727 #endif
728 
729 void glue(helper_cvtps2dq, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
730 {
731     int i;
732     for (i = 0; i < 2 << SHIFT; i++) {
733         d->ZMM_L(i) = x86_float32_to_int32(s->ZMM_S(i), &env->sse_status);
734     }
735 }
736 
737 void glue(helper_cvtpd2dq, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
738 {
739     int i;
740     for (i = 0; i < 1 << SHIFT; i++) {
741         d->ZMM_L(i) = x86_float64_to_int32(s->ZMM_D(i), &env->sse_status);
742     }
743     for (i >>= 1; i < 1 << SHIFT; i++) {
744          d->Q(i) = 0;
745     }
746 }
747 
748 #if SHIFT == 1
749 void helper_cvtps2pi(CPUX86State *env, MMXReg *d, ZMMReg *s)
750 {
751     d->MMX_L(0) = x86_float32_to_int32(s->ZMM_S(0), &env->sse_status);
752     d->MMX_L(1) = x86_float32_to_int32(s->ZMM_S(1), &env->sse_status);
753 }
754 
755 void helper_cvtpd2pi(CPUX86State *env, MMXReg *d, ZMMReg *s)
756 {
757     d->MMX_L(0) = x86_float64_to_int32(s->ZMM_D(0), &env->sse_status);
758     d->MMX_L(1) = x86_float64_to_int32(s->ZMM_D(1), &env->sse_status);
759 }
760 
761 int32_t helper_cvtss2si(CPUX86State *env, ZMMReg *s)
762 {
763     return x86_float32_to_int32(s->ZMM_S(0), &env->sse_status);
764 }
765 
766 int32_t helper_cvtsd2si(CPUX86State *env, ZMMReg *s)
767 {
768     return x86_float64_to_int32(s->ZMM_D(0), &env->sse_status);
769 }
770 
771 #ifdef TARGET_X86_64
772 int64_t helper_cvtss2sq(CPUX86State *env, ZMMReg *s)
773 {
774     return x86_float32_to_int64(s->ZMM_S(0), &env->sse_status);
775 }
776 
777 int64_t helper_cvtsd2sq(CPUX86State *env, ZMMReg *s)
778 {
779     return x86_float64_to_int64(s->ZMM_D(0), &env->sse_status);
780 }
781 #endif
782 #endif
783 
784 /* float to integer truncated */
785 void glue(helper_cvttps2dq, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
786 {
787     int i;
788     for (i = 0; i < 2 << SHIFT; i++) {
789         d->ZMM_L(i) = x86_float32_to_int32_round_to_zero(s->ZMM_S(i),
790                                                          &env->sse_status);
791     }
792 }
793 
794 void glue(helper_cvttpd2dq, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
795 {
796     int i;
797     for (i = 0; i < 1 << SHIFT; i++) {
798         d->ZMM_L(i) = x86_float64_to_int32_round_to_zero(s->ZMM_D(i),
799                                                          &env->sse_status);
800     }
801     for (i >>= 1; i < 1 << SHIFT; i++) {
802          d->Q(i) = 0;
803     }
804 }
805 
806 #if SHIFT == 1
807 void helper_cvttps2pi(CPUX86State *env, MMXReg *d, ZMMReg *s)
808 {
809     d->MMX_L(0) = x86_float32_to_int32_round_to_zero(s->ZMM_S(0), &env->sse_status);
810     d->MMX_L(1) = x86_float32_to_int32_round_to_zero(s->ZMM_S(1), &env->sse_status);
811 }
812 
813 void helper_cvttpd2pi(CPUX86State *env, MMXReg *d, ZMMReg *s)
814 {
815     d->MMX_L(0) = x86_float64_to_int32_round_to_zero(s->ZMM_D(0), &env->sse_status);
816     d->MMX_L(1) = x86_float64_to_int32_round_to_zero(s->ZMM_D(1), &env->sse_status);
817 }
818 
819 int32_t helper_cvttss2si(CPUX86State *env, ZMMReg *s)
820 {
821     return x86_float32_to_int32_round_to_zero(s->ZMM_S(0), &env->sse_status);
822 }
823 
824 int32_t helper_cvttsd2si(CPUX86State *env, ZMMReg *s)
825 {
826     return x86_float64_to_int32_round_to_zero(s->ZMM_D(0), &env->sse_status);
827 }
828 
829 #ifdef TARGET_X86_64
830 int64_t helper_cvttss2sq(CPUX86State *env, ZMMReg *s)
831 {
832     return x86_float32_to_int64_round_to_zero(s->ZMM_S(0), &env->sse_status);
833 }
834 
835 int64_t helper_cvttsd2sq(CPUX86State *env, ZMMReg *s)
836 {
837     return x86_float64_to_int64_round_to_zero(s->ZMM_D(0), &env->sse_status);
838 }
839 #endif
840 #endif
841 
842 void glue(helper_rsqrtps, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
843 {
844     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
845     int i;
846     for (i = 0; i < 2 << SHIFT; i++) {
847         d->ZMM_S(i) = float32_div(float32_one,
848                                   float32_sqrt(s->ZMM_S(i), &env->sse_status),
849                                   &env->sse_status);
850     }
851     set_float_exception_flags(old_flags, &env->sse_status);
852 }
853 
854 #if SHIFT == 1
855 void helper_rsqrtss(CPUX86State *env, ZMMReg *d, ZMMReg *v, ZMMReg *s)
856 {
857     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
858     int i;
859     d->ZMM_S(0) = float32_div(float32_one,
860                               float32_sqrt(s->ZMM_S(0), &env->sse_status),
861                               &env->sse_status);
862     set_float_exception_flags(old_flags, &env->sse_status);
863     for (i = 1; i < 2 << SHIFT; i++) {
864         d->ZMM_L(i) = v->ZMM_L(i);
865     }
866 }
867 #endif
868 
869 void glue(helper_rcpps, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
870 {
871     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
872     int i;
873     for (i = 0; i < 2 << SHIFT; i++) {
874         d->ZMM_S(i) = float32_div(float32_one, s->ZMM_S(i), &env->sse_status);
875     }
876     set_float_exception_flags(old_flags, &env->sse_status);
877 }
878 
879 #if SHIFT == 1
880 void helper_rcpss(CPUX86State *env, ZMMReg *d, ZMMReg *v, ZMMReg *s)
881 {
882     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
883     int i;
884     d->ZMM_S(0) = float32_div(float32_one, s->ZMM_S(0), &env->sse_status);
885     for (i = 1; i < 2 << SHIFT; i++) {
886         d->ZMM_L(i) = v->ZMM_L(i);
887     }
888     set_float_exception_flags(old_flags, &env->sse_status);
889 }
890 #endif
891 
892 #if SHIFT == 1
893 static inline uint64_t helper_extrq(uint64_t src, int shift, int len)
894 {
895     uint64_t mask;
896 
897     if (len == 0) {
898         mask = ~0LL;
899     } else {
900         mask = (1ULL << len) - 1;
901     }
902     return (src >> shift) & mask;
903 }
904 
905 void helper_extrq_r(CPUX86State *env, ZMMReg *d, ZMMReg *s)
906 {
907     d->ZMM_Q(0) = helper_extrq(d->ZMM_Q(0), s->ZMM_B(1) & 63, s->ZMM_B(0) & 63);
908 }
909 
910 void helper_extrq_i(CPUX86State *env, ZMMReg *d, int index, int length)
911 {
912     d->ZMM_Q(0) = helper_extrq(d->ZMM_Q(0), index, length);
913 }
914 
915 static inline uint64_t helper_insertq(uint64_t dest, uint64_t src, int shift, int len)
916 {
917     uint64_t mask;
918 
919     if (len == 0) {
920         mask = ~0ULL;
921     } else {
922         mask = (1ULL << len) - 1;
923     }
924     return (dest & ~(mask << shift)) | ((src & mask) << shift);
925 }
926 
927 void helper_insertq_r(CPUX86State *env, ZMMReg *d, ZMMReg *s)
928 {
929     d->ZMM_Q(0) = helper_insertq(d->ZMM_Q(0), s->ZMM_Q(0), s->ZMM_B(9) & 63, s->ZMM_B(8) & 63);
930 }
931 
932 void helper_insertq_i(CPUX86State *env, ZMMReg *d, ZMMReg *s, int index, int length)
933 {
934     d->ZMM_Q(0) = helper_insertq(d->ZMM_Q(0), s->ZMM_Q(0), index, length);
935 }
936 #endif
937 
938 #define SSE_HELPER_HPS(name, F)  \
939 void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s) \
940 {                                                                 \
941     float32 r[2 << SHIFT];                                        \
942     int i, j, k;                                                  \
943     for (k = 0; k < 2 << SHIFT; k += LANE_WIDTH / 4) {            \
944         for (i = j = 0; j < 4; i++, j += 2) {                     \
945             r[i + k] = F(v->ZMM_S(j + k), v->ZMM_S(j + k + 1), &env->sse_status); \
946         }                                                         \
947         for (j = 0; j < 4; i++, j += 2) {                         \
948             r[i + k] = F(s->ZMM_S(j + k), s->ZMM_S(j + k + 1), &env->sse_status); \
949         }                                                         \
950     }                                                             \
951     for (i = 0; i < 2 << SHIFT; i++) {                            \
952         d->ZMM_S(i) = r[i];                                       \
953     }                                                             \
954 }
955 
956 SSE_HELPER_HPS(haddps, float32_add)
957 SSE_HELPER_HPS(hsubps, float32_sub)
958 
959 #define SSE_HELPER_HPD(name, F)  \
960 void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s) \
961 {                                                                 \
962     float64 r[1 << SHIFT];                                        \
963     int i, j, k;                                                  \
964     for (k = 0; k < 1 << SHIFT; k += LANE_WIDTH / 8) {            \
965         for (i = j = 0; j < 2; i++, j += 2) {                     \
966             r[i + k] = F(v->ZMM_D(j + k), v->ZMM_D(j + k + 1), &env->sse_status); \
967         }                                                         \
968         for (j = 0; j < 2; i++, j += 2) {                         \
969             r[i + k] = F(s->ZMM_D(j + k), s->ZMM_D(j + k + 1), &env->sse_status); \
970         }                                                         \
971     }                                                             \
972     for (i = 0; i < 1 << SHIFT; i++) {                            \
973         d->ZMM_D(i) = r[i];                                       \
974     }                                                             \
975 }
976 
977 SSE_HELPER_HPD(haddpd, float64_add)
978 SSE_HELPER_HPD(hsubpd, float64_sub)
979 
980 void glue(helper_addsubps, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
981 {
982     int i;
983     for (i = 0; i < 2 << SHIFT; i += 2) {
984         d->ZMM_S(i) = float32_sub(v->ZMM_S(i), s->ZMM_S(i), &env->sse_status);
985         d->ZMM_S(i+1) = float32_add(v->ZMM_S(i+1), s->ZMM_S(i+1), &env->sse_status);
986     }
987 }
988 
989 void glue(helper_addsubpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
990 {
991     int i;
992     for (i = 0; i < 1 << SHIFT; i += 2) {
993         d->ZMM_D(i) = float64_sub(v->ZMM_D(i), s->ZMM_D(i), &env->sse_status);
994         d->ZMM_D(i+1) = float64_add(v->ZMM_D(i+1), s->ZMM_D(i+1), &env->sse_status);
995     }
996 }
997 
998 #define SSE_HELPER_CMP_P(name, F, C)                                    \
999     void glue(helper_ ## name ## ps, SUFFIX)(CPUX86State *env,          \
1000                                              Reg *d, Reg *v, Reg *s)    \
1001     {                                                                   \
1002         int i;                                                          \
1003         for (i = 0; i < 2 << SHIFT; i++) {                              \
1004             d->ZMM_L(i) = C(F(32, v->ZMM_S(i), s->ZMM_S(i))) ? -1 : 0;  \
1005         }                                                               \
1006     }                                                                   \
1007                                                                         \
1008     void glue(helper_ ## name ## pd, SUFFIX)(CPUX86State *env,          \
1009                                              Reg *d, Reg *v, Reg *s)    \
1010     {                                                                   \
1011         int i;                                                          \
1012         for (i = 0; i < 1 << SHIFT; i++) {                              \
1013             d->ZMM_Q(i) = C(F(64, v->ZMM_D(i), s->ZMM_D(i))) ? -1 : 0;  \
1014         }                                                               \
1015     }
1016 
1017 #if SHIFT == 1
1018 #define SSE_HELPER_CMP(name, F, C)                                          \
1019     SSE_HELPER_CMP_P(name, F, C)                                            \
1020     void helper_ ## name ## ss(CPUX86State *env, Reg *d, Reg *v, Reg *s)    \
1021     {                                                                       \
1022         int i;                                                              \
1023         d->ZMM_L(0) = C(F(32, v->ZMM_S(0), s->ZMM_S(0))) ? -1 : 0;          \
1024         for (i = 1; i < 2 << SHIFT; i++) {                                  \
1025             d->ZMM_L(i) = v->ZMM_L(i);                                      \
1026         }                                                                   \
1027     }                                                                       \
1028                                                                             \
1029     void helper_ ## name ## sd(CPUX86State *env, Reg *d, Reg *v, Reg *s)    \
1030     {                                                                       \
1031         int i;                                                              \
1032         d->ZMM_Q(0) = C(F(64, v->ZMM_D(0), s->ZMM_D(0))) ? -1 : 0;          \
1033         for (i = 1; i < 1 << SHIFT; i++) {                                  \
1034             d->ZMM_Q(i) = v->ZMM_Q(i);                                      \
1035         }                                                                   \
1036     }
1037 
1038 static inline bool FPU_EQU(FloatRelation x)
1039 {
1040     return (x == float_relation_equal || x == float_relation_unordered);
1041 }
1042 static inline bool FPU_GE(FloatRelation x)
1043 {
1044     return (x == float_relation_equal || x == float_relation_greater);
1045 }
1046 #define FPU_EQ(x) (x == float_relation_equal)
1047 #define FPU_LT(x) (x == float_relation_less)
1048 #define FPU_LE(x) (x <= float_relation_equal)
1049 #define FPU_GT(x) (x == float_relation_greater)
1050 #define FPU_UNORD(x) (x == float_relation_unordered)
1051 /* We must make sure we evaluate the argument in case it is a signalling NAN */
1052 #define FPU_FALSE(x) (x == float_relation_equal && 0)
1053 
1054 #define FPU_CMPQ(size, a, b) \
1055     float ## size ## _compare_quiet(a, b, &env->sse_status)
1056 #define FPU_CMPS(size, a, b) \
1057     float ## size ## _compare(a, b, &env->sse_status)
1058 
1059 #else
1060 #define SSE_HELPER_CMP(name, F, C) SSE_HELPER_CMP_P(name, F, C)
1061 #endif
1062 
1063 SSE_HELPER_CMP(cmpeq, FPU_CMPQ, FPU_EQ)
1064 SSE_HELPER_CMP(cmplt, FPU_CMPS, FPU_LT)
1065 SSE_HELPER_CMP(cmple, FPU_CMPS, FPU_LE)
1066 SSE_HELPER_CMP(cmpunord, FPU_CMPQ,  FPU_UNORD)
1067 SSE_HELPER_CMP(cmpneq, FPU_CMPQ, !FPU_EQ)
1068 SSE_HELPER_CMP(cmpnlt, FPU_CMPS, !FPU_LT)
1069 SSE_HELPER_CMP(cmpnle, FPU_CMPS, !FPU_LE)
1070 SSE_HELPER_CMP(cmpord, FPU_CMPQ, !FPU_UNORD)
1071 
1072 SSE_HELPER_CMP(cmpequ, FPU_CMPQ, FPU_EQU)
1073 SSE_HELPER_CMP(cmpnge, FPU_CMPS, !FPU_GE)
1074 SSE_HELPER_CMP(cmpngt, FPU_CMPS, !FPU_GT)
1075 SSE_HELPER_CMP(cmpfalse, FPU_CMPQ,  FPU_FALSE)
1076 SSE_HELPER_CMP(cmpnequ, FPU_CMPQ, !FPU_EQU)
1077 SSE_HELPER_CMP(cmpge, FPU_CMPS, FPU_GE)
1078 SSE_HELPER_CMP(cmpgt, FPU_CMPS, FPU_GT)
1079 SSE_HELPER_CMP(cmptrue, FPU_CMPQ,  !FPU_FALSE)
1080 
1081 SSE_HELPER_CMP(cmpeqs, FPU_CMPS, FPU_EQ)
1082 SSE_HELPER_CMP(cmpltq, FPU_CMPQ, FPU_LT)
1083 SSE_HELPER_CMP(cmpleq, FPU_CMPQ, FPU_LE)
1084 SSE_HELPER_CMP(cmpunords, FPU_CMPS,  FPU_UNORD)
1085 SSE_HELPER_CMP(cmpneqq, FPU_CMPS, !FPU_EQ)
1086 SSE_HELPER_CMP(cmpnltq, FPU_CMPQ, !FPU_LT)
1087 SSE_HELPER_CMP(cmpnleq, FPU_CMPQ, !FPU_LE)
1088 SSE_HELPER_CMP(cmpords, FPU_CMPS, !FPU_UNORD)
1089 
1090 SSE_HELPER_CMP(cmpequs, FPU_CMPS, FPU_EQU)
1091 SSE_HELPER_CMP(cmpngeq, FPU_CMPQ, !FPU_GE)
1092 SSE_HELPER_CMP(cmpngtq, FPU_CMPQ, !FPU_GT)
1093 SSE_HELPER_CMP(cmpfalses, FPU_CMPS,  FPU_FALSE)
1094 SSE_HELPER_CMP(cmpnequs, FPU_CMPS, !FPU_EQU)
1095 SSE_HELPER_CMP(cmpgeq, FPU_CMPQ, FPU_GE)
1096 SSE_HELPER_CMP(cmpgtq, FPU_CMPQ, FPU_GT)
1097 SSE_HELPER_CMP(cmptrues, FPU_CMPS,  !FPU_FALSE)
1098 
1099 #undef SSE_HELPER_CMP
1100 
1101 #if SHIFT == 1
1102 static const int comis_eflags[4] = {CC_C, CC_Z, 0, CC_Z | CC_P | CC_C};
1103 
1104 void helper_ucomiss(CPUX86State *env, Reg *d, Reg *s)
1105 {
1106     FloatRelation ret;
1107     float32 s0, s1;
1108 
1109     s0 = d->ZMM_S(0);
1110     s1 = s->ZMM_S(0);
1111     ret = float32_compare_quiet(s0, s1, &env->sse_status);
1112     CC_SRC = comis_eflags[ret + 1];
1113 }
1114 
1115 void helper_comiss(CPUX86State *env, Reg *d, Reg *s)
1116 {
1117     FloatRelation ret;
1118     float32 s0, s1;
1119 
1120     s0 = d->ZMM_S(0);
1121     s1 = s->ZMM_S(0);
1122     ret = float32_compare(s0, s1, &env->sse_status);
1123     CC_SRC = comis_eflags[ret + 1];
1124 }
1125 
1126 void helper_ucomisd(CPUX86State *env, Reg *d, Reg *s)
1127 {
1128     FloatRelation ret;
1129     float64 d0, d1;
1130 
1131     d0 = d->ZMM_D(0);
1132     d1 = s->ZMM_D(0);
1133     ret = float64_compare_quiet(d0, d1, &env->sse_status);
1134     CC_SRC = comis_eflags[ret + 1];
1135 }
1136 
1137 void helper_comisd(CPUX86State *env, Reg *d, Reg *s)
1138 {
1139     FloatRelation ret;
1140     float64 d0, d1;
1141 
1142     d0 = d->ZMM_D(0);
1143     d1 = s->ZMM_D(0);
1144     ret = float64_compare(d0, d1, &env->sse_status);
1145     CC_SRC = comis_eflags[ret + 1];
1146 }
1147 #endif
1148 
1149 uint32_t glue(helper_movmskps, SUFFIX)(CPUX86State *env, Reg *s)
1150 {
1151     uint32_t mask;
1152     int i;
1153 
1154     mask = 0;
1155     for (i = 0; i < 2 << SHIFT; i++) {
1156         mask |= (s->ZMM_L(i) >> (31 - i)) & (1 << i);
1157     }
1158     return mask;
1159 }
1160 
1161 uint32_t glue(helper_movmskpd, SUFFIX)(CPUX86State *env, Reg *s)
1162 {
1163     uint32_t mask;
1164     int i;
1165 
1166     mask = 0;
1167     for (i = 0; i < 1 << SHIFT; i++) {
1168         mask |= (s->ZMM_Q(i) >> (63 - i)) & (1 << i);
1169     }
1170     return mask;
1171 }
1172 
1173 #endif
1174 
1175 #define PACK_HELPER_B(name, F) \
1176 void glue(helper_pack ## name, SUFFIX)(CPUX86State *env,      \
1177         Reg *d, Reg *v, Reg *s)                               \
1178 {                                                             \
1179     uint8_t r[PACK_WIDTH * 2];                                \
1180     int j, k;                                                 \
1181     for (j = 0; j < 4 << SHIFT; j += PACK_WIDTH) {            \
1182         for (k = 0; k < PACK_WIDTH; k++) {                    \
1183             r[k] = F((int16_t)v->W(j + k));                   \
1184         }                                                     \
1185         for (k = 0; k < PACK_WIDTH; k++) {                    \
1186             r[PACK_WIDTH + k] = F((int16_t)s->W(j + k));      \
1187         }                                                     \
1188         for (k = 0; k < PACK_WIDTH * 2; k++) {                \
1189             d->B(2 * j + k) = r[k];                           \
1190         }                                                     \
1191     }                                                         \
1192 }
1193 
1194 PACK_HELPER_B(sswb, satsb)
1195 PACK_HELPER_B(uswb, satub)
1196 
1197 void glue(helper_packssdw, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
1198 {
1199     uint16_t r[PACK_WIDTH];
1200     int j, k;
1201 
1202     for (j = 0; j < 2 << SHIFT; j += PACK_WIDTH / 2) {
1203         for (k = 0; k < PACK_WIDTH / 2; k++) {
1204             r[k] = satsw(v->L(j + k));
1205         }
1206         for (k = 0; k < PACK_WIDTH / 2; k++) {
1207             r[PACK_WIDTH / 2 + k] = satsw(s->L(j + k));
1208         }
1209         for (k = 0; k < PACK_WIDTH; k++) {
1210             d->W(2 * j + k) = r[k];
1211         }
1212     }
1213 }
1214 
1215 #define UNPCK_OP(base_name, base)                                       \
1216                                                                         \
1217     void glue(helper_punpck ## base_name ## bw, SUFFIX)(CPUX86State *env,\
1218                                                 Reg *d, Reg *v, Reg *s) \
1219     {                                                                   \
1220         uint8_t r[PACK_WIDTH * 2];                                      \
1221         int j, i;                                                       \
1222                                                                         \
1223         for (j = 0; j < 8 << SHIFT; ) {                                 \
1224             int k = j + base * PACK_WIDTH;                              \
1225             for (i = 0; i < PACK_WIDTH; i++) {                          \
1226                 r[2 * i] = v->B(k + i);                                 \
1227                 r[2 * i + 1] = s->B(k + i);                             \
1228             }                                                           \
1229             for (i = 0; i < PACK_WIDTH * 2; i++, j++) {                 \
1230                 d->B(j) = r[i];                                         \
1231             }                                                           \
1232         }                                                               \
1233     }                                                                   \
1234                                                                         \
1235     void glue(helper_punpck ## base_name ## wd, SUFFIX)(CPUX86State *env,\
1236                                                 Reg *d, Reg *v, Reg *s) \
1237     {                                                                   \
1238         uint16_t r[PACK_WIDTH];                                         \
1239         int j, i;                                                       \
1240                                                                         \
1241         for (j = 0; j < 4 << SHIFT; ) {                                 \
1242             int k = j + base * PACK_WIDTH / 2;                          \
1243             for (i = 0; i < PACK_WIDTH / 2; i++) {                      \
1244                 r[2 * i] = v->W(k + i);                                 \
1245                 r[2 * i + 1] = s->W(k + i);                             \
1246             }                                                           \
1247             for (i = 0; i < PACK_WIDTH; i++, j++) {                     \
1248                 d->W(j) = r[i];                                         \
1249             }                                                           \
1250         }                                                               \
1251     }                                                                   \
1252                                                                         \
1253     void glue(helper_punpck ## base_name ## dq, SUFFIX)(CPUX86State *env,\
1254                                                 Reg *d, Reg *v, Reg *s) \
1255     {                                                                   \
1256         uint32_t r[PACK_WIDTH / 2];                                     \
1257         int j, i;                                                       \
1258                                                                         \
1259         for (j = 0; j < 2 << SHIFT; ) {                                 \
1260             int k = j + base * PACK_WIDTH / 4;                          \
1261             for (i = 0; i < PACK_WIDTH / 4; i++) {                      \
1262                 r[2 * i] = v->L(k + i);                                 \
1263                 r[2 * i + 1] = s->L(k + i);                             \
1264             }                                                           \
1265             for (i = 0; i < PACK_WIDTH / 2; i++, j++) {                 \
1266                 d->L(j) = r[i];                                         \
1267             }                                                           \
1268         }                                                               \
1269     }                                                                   \
1270                                                                         \
1271     XMM_ONLY(                                                           \
1272              void glue(helper_punpck ## base_name ## qdq, SUFFIX)(      \
1273                         CPUX86State *env, Reg *d, Reg *v, Reg *s)       \
1274              {                                                          \
1275                  uint64_t r[2];                                         \
1276                  int i;                                                 \
1277                                                                         \
1278                  for (i = 0; i < 1 << SHIFT; i += 2) {                  \
1279                      r[0] = v->Q(base + i);                             \
1280                      r[1] = s->Q(base + i);                             \
1281                      d->Q(i) = r[0];                                    \
1282                      d->Q(i + 1) = r[1];                                \
1283                  }                                                      \
1284              }                                                          \
1285                                                                         )
1286 
1287 UNPCK_OP(l, 0)
1288 UNPCK_OP(h, 1)
1289 
1290 #undef PACK_WIDTH
1291 #undef PACK_HELPER_B
1292 #undef UNPCK_OP
1293 
1294 
1295 /* 3DNow! float ops */
1296 #if SHIFT == 0
1297 void helper_pi2fd(CPUX86State *env, MMXReg *d, MMXReg *s)
1298 {
1299     d->MMX_S(0) = int32_to_float32(s->MMX_L(0), &env->mmx_status);
1300     d->MMX_S(1) = int32_to_float32(s->MMX_L(1), &env->mmx_status);
1301 }
1302 
1303 void helper_pi2fw(CPUX86State *env, MMXReg *d, MMXReg *s)
1304 {
1305     d->MMX_S(0) = int32_to_float32((int16_t)s->MMX_W(0), &env->mmx_status);
1306     d->MMX_S(1) = int32_to_float32((int16_t)s->MMX_W(2), &env->mmx_status);
1307 }
1308 
1309 void helper_pf2id(CPUX86State *env, MMXReg *d, MMXReg *s)
1310 {
1311     d->MMX_L(0) = float32_to_int32_round_to_zero(s->MMX_S(0), &env->mmx_status);
1312     d->MMX_L(1) = float32_to_int32_round_to_zero(s->MMX_S(1), &env->mmx_status);
1313 }
1314 
1315 void helper_pf2iw(CPUX86State *env, MMXReg *d, MMXReg *s)
1316 {
1317     d->MMX_L(0) = satsw(float32_to_int32_round_to_zero(s->MMX_S(0),
1318                                                        &env->mmx_status));
1319     d->MMX_L(1) = satsw(float32_to_int32_round_to_zero(s->MMX_S(1),
1320                                                        &env->mmx_status));
1321 }
1322 
1323 void helper_pfacc(CPUX86State *env, MMXReg *d, MMXReg *s)
1324 {
1325     float32 r;
1326 
1327     r = float32_add(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1328     d->MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1329     d->MMX_S(0) = r;
1330 }
1331 
1332 void helper_pfadd(CPUX86State *env, MMXReg *d, MMXReg *s)
1333 {
1334     d->MMX_S(0) = float32_add(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1335     d->MMX_S(1) = float32_add(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1336 }
1337 
1338 void helper_pfcmpeq(CPUX86State *env, MMXReg *d, MMXReg *s)
1339 {
1340     d->MMX_L(0) = float32_eq_quiet(d->MMX_S(0), s->MMX_S(0),
1341                                    &env->mmx_status) ? -1 : 0;
1342     d->MMX_L(1) = float32_eq_quiet(d->MMX_S(1), s->MMX_S(1),
1343                                    &env->mmx_status) ? -1 : 0;
1344 }
1345 
1346 void helper_pfcmpge(CPUX86State *env, MMXReg *d, MMXReg *s)
1347 {
1348     d->MMX_L(0) = float32_le(s->MMX_S(0), d->MMX_S(0),
1349                              &env->mmx_status) ? -1 : 0;
1350     d->MMX_L(1) = float32_le(s->MMX_S(1), d->MMX_S(1),
1351                              &env->mmx_status) ? -1 : 0;
1352 }
1353 
1354 void helper_pfcmpgt(CPUX86State *env, MMXReg *d, MMXReg *s)
1355 {
1356     d->MMX_L(0) = float32_lt(s->MMX_S(0), d->MMX_S(0),
1357                              &env->mmx_status) ? -1 : 0;
1358     d->MMX_L(1) = float32_lt(s->MMX_S(1), d->MMX_S(1),
1359                              &env->mmx_status) ? -1 : 0;
1360 }
1361 
1362 void helper_pfmax(CPUX86State *env, MMXReg *d, MMXReg *s)
1363 {
1364     if (float32_lt(d->MMX_S(0), s->MMX_S(0), &env->mmx_status)) {
1365         d->MMX_S(0) = s->MMX_S(0);
1366     }
1367     if (float32_lt(d->MMX_S(1), s->MMX_S(1), &env->mmx_status)) {
1368         d->MMX_S(1) = s->MMX_S(1);
1369     }
1370 }
1371 
1372 void helper_pfmin(CPUX86State *env, MMXReg *d, MMXReg *s)
1373 {
1374     if (float32_lt(s->MMX_S(0), d->MMX_S(0), &env->mmx_status)) {
1375         d->MMX_S(0) = s->MMX_S(0);
1376     }
1377     if (float32_lt(s->MMX_S(1), d->MMX_S(1), &env->mmx_status)) {
1378         d->MMX_S(1) = s->MMX_S(1);
1379     }
1380 }
1381 
1382 void helper_pfmul(CPUX86State *env, MMXReg *d, MMXReg *s)
1383 {
1384     d->MMX_S(0) = float32_mul(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1385     d->MMX_S(1) = float32_mul(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1386 }
1387 
1388 void helper_pfnacc(CPUX86State *env, MMXReg *d, MMXReg *s)
1389 {
1390     float32 r;
1391 
1392     r = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1393     d->MMX_S(1) = float32_sub(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1394     d->MMX_S(0) = r;
1395 }
1396 
1397 void helper_pfpnacc(CPUX86State *env, MMXReg *d, MMXReg *s)
1398 {
1399     float32 r;
1400 
1401     r = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1402     d->MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1403     d->MMX_S(0) = r;
1404 }
1405 
1406 void helper_pfrcp(CPUX86State *env, MMXReg *d, MMXReg *s)
1407 {
1408     d->MMX_S(0) = float32_div(float32_one, s->MMX_S(0), &env->mmx_status);
1409     d->MMX_S(1) = d->MMX_S(0);
1410 }
1411 
1412 void helper_pfrsqrt(CPUX86State *env, MMXReg *d, MMXReg *s)
1413 {
1414     d->MMX_L(1) = s->MMX_L(0) & 0x7fffffff;
1415     d->MMX_S(1) = float32_div(float32_one,
1416                               float32_sqrt(d->MMX_S(1), &env->mmx_status),
1417                               &env->mmx_status);
1418     d->MMX_L(1) |= s->MMX_L(0) & 0x80000000;
1419     d->MMX_L(0) = d->MMX_L(1);
1420 }
1421 
1422 void helper_pfsub(CPUX86State *env, MMXReg *d, MMXReg *s)
1423 {
1424     d->MMX_S(0) = float32_sub(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1425     d->MMX_S(1) = float32_sub(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1426 }
1427 
1428 void helper_pfsubr(CPUX86State *env, MMXReg *d, MMXReg *s)
1429 {
1430     d->MMX_S(0) = float32_sub(s->MMX_S(0), d->MMX_S(0), &env->mmx_status);
1431     d->MMX_S(1) = float32_sub(s->MMX_S(1), d->MMX_S(1), &env->mmx_status);
1432 }
1433 
1434 void helper_pswapd(CPUX86State *env, MMXReg *d, MMXReg *s)
1435 {
1436     uint32_t r;
1437 
1438     r = s->MMX_L(0);
1439     d->MMX_L(0) = s->MMX_L(1);
1440     d->MMX_L(1) = r;
1441 }
1442 #endif
1443 
1444 /* SSSE3 op helpers */
1445 void glue(helper_pshufb, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
1446 {
1447     int i;
1448 #if SHIFT == 0
1449     uint8_t r[8];
1450 
1451     for (i = 0; i < 8; i++) {
1452         r[i] = (s->B(i) & 0x80) ? 0 : (v->B(s->B(i) & 7));
1453     }
1454     for (i = 0; i < 8; i++) {
1455         d->B(i) = r[i];
1456     }
1457 #else
1458     uint8_t r[8 << SHIFT];
1459 
1460     for (i = 0; i < 8 << SHIFT; i++) {
1461         int j = i & ~0xf;
1462         r[i] = (s->B(i) & 0x80) ? 0 : v->B(j | (s->B(i) & 0xf));
1463     }
1464     for (i = 0; i < 8 << SHIFT; i++) {
1465         d->B(i) = r[i];
1466     }
1467 #endif
1468 }
1469 
1470 #define SSE_HELPER_HW(name, F)  \
1471 void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s) \
1472 {                                                          \
1473     uint16_t r[4 << SHIFT];                                \
1474     int i, j, k;                                           \
1475     for (k = 0; k < 4 << SHIFT; k += LANE_WIDTH / 2) {     \
1476         for (i = j = 0; j < LANE_WIDTH / 2; i++, j += 2) { \
1477             r[i + k] = F(v->W(j + k), v->W(j + k + 1));    \
1478         }                                                  \
1479         for (j = 0; j < LANE_WIDTH / 2; i++, j += 2) {     \
1480             r[i + k] = F(s->W(j + k), s->W(j + k + 1));    \
1481         }                                                  \
1482     }                                                      \
1483     for (i = 0; i < 4 << SHIFT; i++) {                     \
1484         d->W(i) = r[i];                                    \
1485     }                                                      \
1486 }
1487 
1488 #define SSE_HELPER_HL(name, F)  \
1489 void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s) \
1490 {                                                          \
1491     uint32_t r[2 << SHIFT];                                \
1492     int i, j, k;                                           \
1493     for (k = 0; k < 2 << SHIFT; k += LANE_WIDTH / 4) {     \
1494         for (i = j = 0; j < LANE_WIDTH / 4; i++, j += 2) { \
1495             r[i + k] = F(v->L(j + k), v->L(j + k + 1));    \
1496         }                                                  \
1497         for (j = 0; j < LANE_WIDTH / 4; i++, j += 2) {     \
1498             r[i + k] = F(s->L(j + k), s->L(j + k + 1));    \
1499         }                                                  \
1500     }                                                      \
1501     for (i = 0; i < 2 << SHIFT; i++) {                     \
1502         d->L(i) = r[i];                                    \
1503     }                                                      \
1504 }
1505 
1506 SSE_HELPER_HW(phaddw, FADD)
1507 SSE_HELPER_HW(phsubw, FSUB)
1508 SSE_HELPER_HW(phaddsw, FADDSW)
1509 SSE_HELPER_HW(phsubsw, FSUBSW)
1510 SSE_HELPER_HL(phaddd, FADD)
1511 SSE_HELPER_HL(phsubd, FSUB)
1512 
1513 #undef SSE_HELPER_HW
1514 #undef SSE_HELPER_HL
1515 
1516 void glue(helper_pmaddubsw, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
1517 {
1518     int i;
1519     for (i = 0; i < 4 << SHIFT; i++) {
1520         d->W(i) = satsw((int8_t)s->B(i * 2) * (uint8_t)v->B(i * 2) +
1521                         (int8_t)s->B(i * 2 + 1) * (uint8_t)v->B(i * 2 + 1));
1522     }
1523 }
1524 
1525 #define FMULHRSW(d, s) (((int16_t) d * (int16_t)s + 0x4000) >> 15)
1526 SSE_HELPER_W(helper_pmulhrsw, FMULHRSW)
1527 
1528 #define FSIGNB(d, s) (s <= INT8_MAX  ? s ? d : 0 : -(int8_t)d)
1529 #define FSIGNW(d, s) (s <= INT16_MAX ? s ? d : 0 : -(int16_t)d)
1530 #define FSIGNL(d, s) (s <= INT32_MAX ? s ? d : 0 : -(int32_t)d)
1531 SSE_HELPER_B(helper_psignb, FSIGNB)
1532 SSE_HELPER_W(helper_psignw, FSIGNW)
1533 SSE_HELPER_L(helper_psignd, FSIGNL)
1534 
1535 void glue(helper_palignr, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
1536                                   uint32_t imm)
1537 {
1538     int i;
1539 
1540     /* XXX could be checked during translation */
1541     if (imm >= (SHIFT ? 32 : 16)) {
1542         for (i = 0; i < (1 << SHIFT); i++) {
1543             d->Q(i) = 0;
1544         }
1545     } else {
1546         int shift = imm * 8;
1547 #define SHR(v, i) (i < 64 && i > -64 ? i > 0 ? v >> (i) : (v << -(i)) : 0)
1548 #if SHIFT == 0
1549         d->Q(0) = SHR(s->Q(0), shift - 0) |
1550             SHR(v->Q(0), shift -  64);
1551 #else
1552         for (i = 0; i < (1 << SHIFT); i += 2) {
1553             uint64_t r0, r1;
1554 
1555             r0 = SHR(s->Q(i), shift - 0) |
1556                  SHR(s->Q(i + 1), shift -  64) |
1557                  SHR(v->Q(i), shift - 128) |
1558                  SHR(v->Q(i + 1), shift - 192);
1559             r1 = SHR(s->Q(i), shift + 64) |
1560                  SHR(s->Q(i + 1), shift -   0) |
1561                  SHR(v->Q(i), shift -  64) |
1562                  SHR(v->Q(i + 1), shift - 128);
1563             d->Q(i) = r0;
1564             d->Q(i + 1) = r1;
1565         }
1566 #endif
1567 #undef SHR
1568     }
1569 }
1570 
1571 #if SHIFT >= 1
1572 
1573 #define SSE_HELPER_V(name, elem, num, F)                                \
1574     void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,   \
1575                             Reg *m)                                     \
1576     {                                                                   \
1577         int i;                                                          \
1578         for (i = 0; i < num; i++) {                                     \
1579             d->elem(i) = F(v->elem(i), s->elem(i), m->elem(i));         \
1580         }                                                               \
1581     }
1582 
1583 #define SSE_HELPER_I(name, elem, num, F)                                \
1584     void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,   \
1585                             uint32_t imm)                               \
1586     {                                                                   \
1587         int i;                                                          \
1588         for (i = 0; i < num; i++) {                                     \
1589             int j = i & 7;                                              \
1590             d->elem(i) = F(v->elem(i), s->elem(i), (imm >> j) & 1);     \
1591         }                                                               \
1592     }
1593 
1594 /* SSE4.1 op helpers */
1595 #define FBLENDVB(v, s, m) ((m & 0x80) ? s : v)
1596 #define FBLENDVPS(v, s, m) ((m & 0x80000000) ? s : v)
1597 #define FBLENDVPD(v, s, m) ((m & 0x8000000000000000LL) ? s : v)
1598 SSE_HELPER_V(helper_pblendvb, B, 8 << SHIFT, FBLENDVB)
1599 SSE_HELPER_V(helper_blendvps, L, 2 << SHIFT, FBLENDVPS)
1600 SSE_HELPER_V(helper_blendvpd, Q, 1 << SHIFT, FBLENDVPD)
1601 
1602 void glue(helper_ptest, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
1603 {
1604     uint64_t zf = 0, cf = 0;
1605     int i;
1606 
1607     for (i = 0; i < 1 << SHIFT; i++) {
1608         zf |= (s->Q(i) &  d->Q(i));
1609         cf |= (s->Q(i) & ~d->Q(i));
1610     }
1611     CC_SRC = (zf ? 0 : CC_Z) | (cf ? 0 : CC_C);
1612 }
1613 
1614 #define FMOVSLDUP(i) s->L((i) & ~1)
1615 #define FMOVSHDUP(i) s->L((i) | 1)
1616 #define FMOVDLDUP(i) s->Q((i) & ~1)
1617 
1618 #define SSE_HELPER_F(name, elem, num, F)                        \
1619     void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)   \
1620     {                                                           \
1621         int n = num;                                            \
1622         for (int i = n; --i >= 0; ) {                           \
1623             d->elem(i) = F(i);                                  \
1624         }                                                       \
1625     }
1626 
1627 #if SHIFT > 0
1628 SSE_HELPER_F(helper_pmovsxbw, W, 4 << SHIFT, (int8_t) s->B)
1629 SSE_HELPER_F(helper_pmovsxbd, L, 2 << SHIFT, (int8_t) s->B)
1630 SSE_HELPER_F(helper_pmovsxbq, Q, 1 << SHIFT, (int8_t) s->B)
1631 SSE_HELPER_F(helper_pmovsxwd, L, 2 << SHIFT, (int16_t) s->W)
1632 SSE_HELPER_F(helper_pmovsxwq, Q, 1 << SHIFT, (int16_t) s->W)
1633 SSE_HELPER_F(helper_pmovsxdq, Q, 1 << SHIFT, (int32_t) s->L)
1634 SSE_HELPER_F(helper_pmovzxbw, W, 4 << SHIFT, s->B)
1635 SSE_HELPER_F(helper_pmovzxbd, L, 2 << SHIFT, s->B)
1636 SSE_HELPER_F(helper_pmovzxbq, Q, 1 << SHIFT, s->B)
1637 SSE_HELPER_F(helper_pmovzxwd, L, 2 << SHIFT, s->W)
1638 SSE_HELPER_F(helper_pmovzxwq, Q, 1 << SHIFT, s->W)
1639 SSE_HELPER_F(helper_pmovzxdq, Q, 1 << SHIFT, s->L)
1640 SSE_HELPER_F(helper_pmovsldup, L, 2 << SHIFT, FMOVSLDUP)
1641 SSE_HELPER_F(helper_pmovshdup, L, 2 << SHIFT, FMOVSHDUP)
1642 SSE_HELPER_F(helper_pmovdldup, Q, 1 << SHIFT, FMOVDLDUP)
1643 #endif
1644 
1645 void glue(helper_pmuldq, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
1646 {
1647     int i;
1648 
1649     for (i = 0; i < 1 << SHIFT; i++) {
1650         d->Q(i) = (int64_t)(int32_t) v->L(2 * i) * (int32_t) s->L(2 * i);
1651     }
1652 }
1653 
1654 void glue(helper_packusdw, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
1655 {
1656     uint16_t r[8];
1657     int i, j, k;
1658 
1659     for (i = 0, j = 0; i <= 2 << SHIFT; i += 8, j += 4) {
1660         r[0] = satuw(v->L(j));
1661         r[1] = satuw(v->L(j + 1));
1662         r[2] = satuw(v->L(j + 2));
1663         r[3] = satuw(v->L(j + 3));
1664         r[4] = satuw(s->L(j));
1665         r[5] = satuw(s->L(j + 1));
1666         r[6] = satuw(s->L(j + 2));
1667         r[7] = satuw(s->L(j + 3));
1668         for (k = 0; k < 8; k++) {
1669             d->W(i + k) = r[k];
1670         }
1671     }
1672 }
1673 
1674 #if SHIFT == 1
1675 void glue(helper_phminposuw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
1676 {
1677     int idx = 0;
1678 
1679     if (s->W(1) < s->W(idx)) {
1680         idx = 1;
1681     }
1682     if (s->W(2) < s->W(idx)) {
1683         idx = 2;
1684     }
1685     if (s->W(3) < s->W(idx)) {
1686         idx = 3;
1687     }
1688     if (s->W(4) < s->W(idx)) {
1689         idx = 4;
1690     }
1691     if (s->W(5) < s->W(idx)) {
1692         idx = 5;
1693     }
1694     if (s->W(6) < s->W(idx)) {
1695         idx = 6;
1696     }
1697     if (s->W(7) < s->W(idx)) {
1698         idx = 7;
1699     }
1700 
1701     d->W(0) = s->W(idx);
1702     d->W(1) = idx;
1703     d->L(1) = 0;
1704     d->Q(1) = 0;
1705 }
1706 #endif
1707 
1708 void glue(helper_roundps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
1709                                   uint32_t mode)
1710 {
1711     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
1712     signed char prev_rounding_mode;
1713     int i;
1714 
1715     prev_rounding_mode = env->sse_status.float_rounding_mode;
1716     if (!(mode & (1 << 2))) {
1717         set_x86_rounding_mode(mode & 3, &env->sse_status);
1718     }
1719 
1720     for (i = 0; i < 2 << SHIFT; i++) {
1721         d->ZMM_S(i) = float32_round_to_int(s->ZMM_S(i), &env->sse_status);
1722     }
1723 
1724     if (mode & (1 << 3) && !(old_flags & float_flag_inexact)) {
1725         set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1726                                   ~float_flag_inexact,
1727                                   &env->sse_status);
1728     }
1729     env->sse_status.float_rounding_mode = prev_rounding_mode;
1730 }
1731 
1732 void glue(helper_roundpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
1733                                   uint32_t mode)
1734 {
1735     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
1736     signed char prev_rounding_mode;
1737     int i;
1738 
1739     prev_rounding_mode = env->sse_status.float_rounding_mode;
1740     if (!(mode & (1 << 2))) {
1741         set_x86_rounding_mode(mode & 3, &env->sse_status);
1742     }
1743 
1744     for (i = 0; i < 1 << SHIFT; i++) {
1745         d->ZMM_D(i) = float64_round_to_int(s->ZMM_D(i), &env->sse_status);
1746     }
1747 
1748     if (mode & (1 << 3) && !(old_flags & float_flag_inexact)) {
1749         set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1750                                   ~float_flag_inexact,
1751                                   &env->sse_status);
1752     }
1753     env->sse_status.float_rounding_mode = prev_rounding_mode;
1754 }
1755 
1756 #if SHIFT == 1
1757 void glue(helper_roundss, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
1758                                   uint32_t mode)
1759 {
1760     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
1761     signed char prev_rounding_mode;
1762     int i;
1763 
1764     prev_rounding_mode = env->sse_status.float_rounding_mode;
1765     if (!(mode & (1 << 2))) {
1766         set_x86_rounding_mode(mode & 3, &env->sse_status);
1767     }
1768 
1769     d->ZMM_S(0) = float32_round_to_int(s->ZMM_S(0), &env->sse_status);
1770     for (i = 1; i < 2 << SHIFT; i++) {
1771         d->ZMM_L(i) = v->ZMM_L(i);
1772     }
1773 
1774     if (mode & (1 << 3) && !(old_flags & float_flag_inexact)) {
1775         set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1776                                   ~float_flag_inexact,
1777                                   &env->sse_status);
1778     }
1779     env->sse_status.float_rounding_mode = prev_rounding_mode;
1780 }
1781 
1782 void glue(helper_roundsd, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
1783                                   uint32_t mode)
1784 {
1785     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
1786     signed char prev_rounding_mode;
1787     int i;
1788 
1789     prev_rounding_mode = env->sse_status.float_rounding_mode;
1790     if (!(mode & (1 << 2))) {
1791         set_x86_rounding_mode(mode & 3, &env->sse_status);
1792     }
1793 
1794     d->ZMM_D(0) = float64_round_to_int(s->ZMM_D(0), &env->sse_status);
1795     for (i = 1; i < 1 << SHIFT; i++) {
1796         d->ZMM_Q(i) = v->ZMM_Q(i);
1797     }
1798 
1799     if (mode & (1 << 3) && !(old_flags & float_flag_inexact)) {
1800         set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1801                                   ~float_flag_inexact,
1802                                   &env->sse_status);
1803     }
1804     env->sse_status.float_rounding_mode = prev_rounding_mode;
1805 }
1806 #endif
1807 
1808 #define FBLENDP(v, s, m) (m ? s : v)
1809 SSE_HELPER_I(helper_blendps, L, 2 << SHIFT, FBLENDP)
1810 SSE_HELPER_I(helper_blendpd, Q, 1 << SHIFT, FBLENDP)
1811 SSE_HELPER_I(helper_pblendw, W, 4 << SHIFT, FBLENDP)
1812 
1813 void glue(helper_dpps, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
1814                                uint32_t mask)
1815 {
1816     float32 prod1, prod2, temp2, temp3, temp4;
1817     int i;
1818 
1819     for (i = 0; i < 2 << SHIFT; i += 4) {
1820         /*
1821          * We must evaluate (A+B)+(C+D), not ((A+B)+C)+D
1822          * to correctly round the intermediate results
1823          */
1824         if (mask & (1 << 4)) {
1825             prod1 = float32_mul(v->ZMM_S(i), s->ZMM_S(i), &env->sse_status);
1826         } else {
1827             prod1 = float32_zero;
1828         }
1829         if (mask & (1 << 5)) {
1830             prod2 = float32_mul(v->ZMM_S(i+1), s->ZMM_S(i+1), &env->sse_status);
1831         } else {
1832             prod2 = float32_zero;
1833         }
1834         temp2 = float32_add(prod1, prod2, &env->sse_status);
1835         if (mask & (1 << 6)) {
1836             prod1 = float32_mul(v->ZMM_S(i+2), s->ZMM_S(i+2), &env->sse_status);
1837         } else {
1838             prod1 = float32_zero;
1839         }
1840         if (mask & (1 << 7)) {
1841             prod2 = float32_mul(v->ZMM_S(i+3), s->ZMM_S(i+3), &env->sse_status);
1842         } else {
1843             prod2 = float32_zero;
1844         }
1845         temp3 = float32_add(prod1, prod2, &env->sse_status);
1846         temp4 = float32_add(temp2, temp3, &env->sse_status);
1847 
1848         d->ZMM_S(i) = (mask & (1 << 0)) ? temp4 : float32_zero;
1849         d->ZMM_S(i+1) = (mask & (1 << 1)) ? temp4 : float32_zero;
1850         d->ZMM_S(i+2) = (mask & (1 << 2)) ? temp4 : float32_zero;
1851         d->ZMM_S(i+3) = (mask & (1 << 3)) ? temp4 : float32_zero;
1852     }
1853 }
1854 
1855 #if SHIFT == 1
1856 /* Oddly, there is no ymm version of dppd */
1857 void glue(helper_dppd, SUFFIX)(CPUX86State *env,
1858                                Reg *d, Reg *v, Reg *s, uint32_t mask)
1859 {
1860     float64 prod1, prod2, temp2;
1861 
1862     if (mask & (1 << 4)) {
1863         prod1 = float64_mul(v->ZMM_D(0), s->ZMM_D(0), &env->sse_status);
1864     } else {
1865         prod1 = float64_zero;
1866     }
1867     if (mask & (1 << 5)) {
1868         prod2 = float64_mul(v->ZMM_D(1), s->ZMM_D(1), &env->sse_status);
1869     } else {
1870         prod2 = float64_zero;
1871     }
1872     temp2 = float64_add(prod1, prod2, &env->sse_status);
1873     d->ZMM_D(0) = (mask & (1 << 0)) ? temp2 : float64_zero;
1874     d->ZMM_D(1) = (mask & (1 << 1)) ? temp2 : float64_zero;
1875 }
1876 #endif
1877 
1878 void glue(helper_mpsadbw, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
1879                                   uint32_t offset)
1880 {
1881     int i, j;
1882     uint16_t r[8];
1883 
1884     for (j = 0; j < 4 << SHIFT; ) {
1885         int s0 = (j * 2) + ((offset & 3) << 2);
1886         int d0 = (j * 2) + ((offset & 4) << 0);
1887         for (i = 0; i < LANE_WIDTH / 2; i++, d0++) {
1888             r[i] = 0;
1889             r[i] += abs1(v->B(d0 + 0) - s->B(s0 + 0));
1890             r[i] += abs1(v->B(d0 + 1) - s->B(s0 + 1));
1891             r[i] += abs1(v->B(d0 + 2) - s->B(s0 + 2));
1892             r[i] += abs1(v->B(d0 + 3) - s->B(s0 + 3));
1893         }
1894         for (i = 0; i < LANE_WIDTH / 2; i++, j++) {
1895             d->W(j) = r[i];
1896         }
1897         offset >>= 3;
1898     }
1899 }
1900 
1901 /* SSE4.2 op helpers */
1902 #if SHIFT == 1
1903 static inline int pcmp_elen(CPUX86State *env, int reg, uint32_t ctrl)
1904 {
1905     target_long val, limit;
1906 
1907     /* Presence of REX.W is indicated by a bit higher than 7 set */
1908     if (ctrl >> 8) {
1909         val = (target_long)env->regs[reg];
1910     } else {
1911         val = (int32_t)env->regs[reg];
1912     }
1913     if (ctrl & 1) {
1914         limit = 8;
1915     } else {
1916         limit = 16;
1917     }
1918     if ((val > limit) || (val < -limit)) {
1919         return limit;
1920     }
1921     return abs1(val);
1922 }
1923 
1924 static inline int pcmp_ilen(Reg *r, uint8_t ctrl)
1925 {
1926     int val = 0;
1927 
1928     if (ctrl & 1) {
1929         while (val < 8 && r->W(val)) {
1930             val++;
1931         }
1932     } else {
1933         while (val < 16 && r->B(val)) {
1934             val++;
1935         }
1936     }
1937 
1938     return val;
1939 }
1940 
1941 static inline int pcmp_val(Reg *r, uint8_t ctrl, int i)
1942 {
1943     switch ((ctrl >> 0) & 3) {
1944     case 0:
1945         return r->B(i);
1946     case 1:
1947         return r->W(i);
1948     case 2:
1949         return (int8_t)r->B(i);
1950     case 3:
1951     default:
1952         return (int16_t)r->W(i);
1953     }
1954 }
1955 
1956 static inline unsigned pcmpxstrx(CPUX86State *env, Reg *d, Reg *s,
1957                                  uint8_t ctrl, int valids, int validd)
1958 {
1959     unsigned int res = 0;
1960     int v;
1961     int j, i;
1962     int upper = (ctrl & 1) ? 7 : 15;
1963 
1964     valids--;
1965     validd--;
1966 
1967     CC_SRC = (valids < upper ? CC_Z : 0) | (validd < upper ? CC_S : 0);
1968 
1969     switch ((ctrl >> 2) & 3) {
1970     case 0:
1971         for (j = valids; j >= 0; j--) {
1972             res <<= 1;
1973             v = pcmp_val(s, ctrl, j);
1974             for (i = validd; i >= 0; i--) {
1975                 res |= (v == pcmp_val(d, ctrl, i));
1976             }
1977         }
1978         break;
1979     case 1:
1980         for (j = valids; j >= 0; j--) {
1981             res <<= 1;
1982             v = pcmp_val(s, ctrl, j);
1983             for (i = ((validd - 1) | 1); i >= 0; i -= 2) {
1984                 res |= (pcmp_val(d, ctrl, i - 0) >= v &&
1985                         pcmp_val(d, ctrl, i - 1) <= v);
1986             }
1987         }
1988         break;
1989     case 2:
1990         res = (1 << (upper - MAX(valids, validd))) - 1;
1991         res <<= MAX(valids, validd) - MIN(valids, validd);
1992         for (i = MIN(valids, validd); i >= 0; i--) {
1993             res <<= 1;
1994             v = pcmp_val(s, ctrl, i);
1995             res |= (v == pcmp_val(d, ctrl, i));
1996         }
1997         break;
1998     case 3:
1999         if (validd == -1) {
2000             res = (2 << upper) - 1;
2001             break;
2002         }
2003         for (j = valids == upper ? valids : valids - validd; j >= 0; j--) {
2004             res <<= 1;
2005             v = 1;
2006             for (i = MIN(valids - j, validd); i >= 0; i--) {
2007                 v &= (pcmp_val(s, ctrl, i + j) == pcmp_val(d, ctrl, i));
2008             }
2009             res |= v;
2010         }
2011         break;
2012     }
2013 
2014     switch ((ctrl >> 4) & 3) {
2015     case 1:
2016         res ^= (2 << upper) - 1;
2017         break;
2018     case 3:
2019         res ^= (1 << (valids + 1)) - 1;
2020         break;
2021     }
2022 
2023     if (res) {
2024         CC_SRC |= CC_C;
2025     }
2026     if (res & 1) {
2027         CC_SRC |= CC_O;
2028     }
2029 
2030     return res;
2031 }
2032 
2033 void glue(helper_pcmpestri, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2034                                     uint32_t ctrl)
2035 {
2036     unsigned int res = pcmpxstrx(env, d, s, ctrl,
2037                                  pcmp_elen(env, R_EDX, ctrl),
2038                                  pcmp_elen(env, R_EAX, ctrl));
2039 
2040     if (res) {
2041         env->regs[R_ECX] = (ctrl & (1 << 6)) ? 31 - clz32(res) : ctz32(res);
2042     } else {
2043         env->regs[R_ECX] = 16 >> (ctrl & (1 << 0));
2044     }
2045 }
2046 
2047 void glue(helper_pcmpestrm, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2048                                     uint32_t ctrl)
2049 {
2050     int i;
2051     unsigned int res = pcmpxstrx(env, d, s, ctrl,
2052                                  pcmp_elen(env, R_EDX, ctrl),
2053                                  pcmp_elen(env, R_EAX, ctrl));
2054 
2055     if ((ctrl >> 6) & 1) {
2056         if (ctrl & 1) {
2057             for (i = 0; i < 8; i++, res >>= 1) {
2058                 env->xmm_regs[0].W(i) = (res & 1) ? ~0 : 0;
2059             }
2060         } else {
2061             for (i = 0; i < 16; i++, res >>= 1) {
2062                 env->xmm_regs[0].B(i) = (res & 1) ? ~0 : 0;
2063             }
2064         }
2065     } else {
2066         env->xmm_regs[0].Q(1) = 0;
2067         env->xmm_regs[0].Q(0) = res;
2068     }
2069 }
2070 
2071 void glue(helper_pcmpistri, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2072                                     uint32_t ctrl)
2073 {
2074     unsigned int res = pcmpxstrx(env, d, s, ctrl,
2075                                  pcmp_ilen(s, ctrl),
2076                                  pcmp_ilen(d, ctrl));
2077 
2078     if (res) {
2079         env->regs[R_ECX] = (ctrl & (1 << 6)) ? 31 - clz32(res) : ctz32(res);
2080     } else {
2081         env->regs[R_ECX] = 16 >> (ctrl & (1 << 0));
2082     }
2083 }
2084 
2085 void glue(helper_pcmpistrm, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2086                                     uint32_t ctrl)
2087 {
2088     int i;
2089     unsigned int res = pcmpxstrx(env, d, s, ctrl,
2090                                  pcmp_ilen(s, ctrl),
2091                                  pcmp_ilen(d, ctrl));
2092 
2093     if ((ctrl >> 6) & 1) {
2094         if (ctrl & 1) {
2095             for (i = 0; i < 8; i++, res >>= 1) {
2096                 env->xmm_regs[0].W(i) = (res & 1) ? ~0 : 0;
2097             }
2098         } else {
2099             for (i = 0; i < 16; i++, res >>= 1) {
2100                 env->xmm_regs[0].B(i) = (res & 1) ? ~0 : 0;
2101             }
2102         }
2103     } else {
2104         env->xmm_regs[0].Q(1) = 0;
2105         env->xmm_regs[0].Q(0) = res;
2106     }
2107 }
2108 
2109 #define CRCPOLY        0x1edc6f41
2110 #define CRCPOLY_BITREV 0x82f63b78
2111 target_ulong helper_crc32(uint32_t crc1, target_ulong msg, uint32_t len)
2112 {
2113     target_ulong crc = (msg & ((target_ulong) -1 >>
2114                                (TARGET_LONG_BITS - len))) ^ crc1;
2115 
2116     while (len--) {
2117         crc = (crc >> 1) ^ ((crc & 1) ? CRCPOLY_BITREV : 0);
2118     }
2119 
2120     return crc;
2121 }
2122 
2123 #endif
2124 
2125 #if SHIFT == 1
2126 static void clmulq(uint64_t *dest_l, uint64_t *dest_h,
2127                           uint64_t a, uint64_t b)
2128 {
2129     uint64_t al, ah, resh, resl;
2130 
2131     ah = 0;
2132     al = a;
2133     resh = resl = 0;
2134 
2135     while (b) {
2136         if (b & 1) {
2137             resl ^= al;
2138             resh ^= ah;
2139         }
2140         ah = (ah << 1) | (al >> 63);
2141         al <<= 1;
2142         b >>= 1;
2143     }
2144 
2145     *dest_l = resl;
2146     *dest_h = resh;
2147 }
2148 #endif
2149 
2150 void glue(helper_pclmulqdq, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
2151                                     uint32_t ctrl)
2152 {
2153     uint64_t a, b;
2154     int i;
2155 
2156     for (i = 0; i < 1 << SHIFT; i += 2) {
2157         a = v->Q(((ctrl & 1) != 0) + i);
2158         b = s->Q(((ctrl & 16) != 0) + i);
2159         clmulq(&d->Q(i), &d->Q(i + 1), a, b);
2160     }
2161 }
2162 
2163 void glue(helper_aesdec, 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         aesdec_ISB_ISR_IMC_AK(ad, st, rk, false);
2171     }
2172 }
2173 
2174 void glue(helper_aesdeclast, 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         aesdec_ISB_ISR_AK(ad, st, rk, false);
2182     }
2183 }
2184 
2185 void glue(helper_aesenc, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2186 {
2187     for (int i = 0; i < SHIFT; i++) {
2188         AESState *ad = (AESState *)&d->ZMM_X(i);
2189         AESState *st = (AESState *)&v->ZMM_X(i);
2190         AESState *rk = (AESState *)&s->ZMM_X(i);
2191 
2192         aesenc_SB_SR_MC_AK(ad, st, rk, false);
2193     }
2194 }
2195 
2196 void glue(helper_aesenclast, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2197 {
2198     for (int i = 0; i < SHIFT; i++) {
2199         AESState *ad = (AESState *)&d->ZMM_X(i);
2200         AESState *st = (AESState *)&v->ZMM_X(i);
2201         AESState *rk = (AESState *)&s->ZMM_X(i);
2202 
2203         aesenc_SB_SR_AK(ad, st, rk, false);
2204     }
2205 }
2206 
2207 #if SHIFT == 1
2208 void glue(helper_aesimc, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
2209 {
2210     AESState *ad = (AESState *)&d->ZMM_X(0);
2211     AESState *st = (AESState *)&s->ZMM_X(0);
2212 
2213     aesdec_IMC(ad, st, false);
2214 }
2215 
2216 void glue(helper_aeskeygenassist, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2217                                           uint32_t ctrl)
2218 {
2219     int i;
2220     Reg tmp = *s;
2221 
2222     for (i = 0 ; i < 4 ; i++) {
2223         d->B(i) = AES_sbox[tmp.B(i + 4)];
2224         d->B(i + 8) = AES_sbox[tmp.B(i + 12)];
2225     }
2226     d->L(1) = (d->L(0) << 24 | d->L(0) >> 8) ^ ctrl;
2227     d->L(3) = (d->L(2) << 24 | d->L(2) >> 8) ^ ctrl;
2228 }
2229 #endif
2230 #endif
2231 
2232 #if SHIFT >= 1
2233 void glue(helper_vpermilpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2234 {
2235     uint64_t r0, r1;
2236     int i;
2237 
2238     for (i = 0; i < 1 << SHIFT; i += 2) {
2239         r0 = v->Q(i + ((s->Q(i) >> 1) & 1));
2240         r1 = v->Q(i + ((s->Q(i+1) >> 1) & 1));
2241         d->Q(i) = r0;
2242         d->Q(i+1) = r1;
2243     }
2244 }
2245 
2246 void glue(helper_vpermilps, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2247 {
2248     uint32_t r0, r1, r2, r3;
2249     int i;
2250 
2251     for (i = 0; i < 2 << SHIFT; i += 4) {
2252         r0 = v->L(i + (s->L(i) & 3));
2253         r1 = v->L(i + (s->L(i+1) & 3));
2254         r2 = v->L(i + (s->L(i+2) & 3));
2255         r3 = v->L(i + (s->L(i+3) & 3));
2256         d->L(i) = r0;
2257         d->L(i+1) = r1;
2258         d->L(i+2) = r2;
2259         d->L(i+3) = r3;
2260     }
2261 }
2262 
2263 void glue(helper_vpermilpd_imm, SUFFIX)(Reg *d, Reg *s, uint32_t order)
2264 {
2265     uint64_t r0, r1;
2266     int i;
2267 
2268     for (i = 0; i < 1 << SHIFT; i += 2) {
2269         r0 = s->Q(i + ((order >> 0) & 1));
2270         r1 = s->Q(i + ((order >> 1) & 1));
2271         d->Q(i) = r0;
2272         d->Q(i+1) = r1;
2273 
2274         order >>= 2;
2275     }
2276 }
2277 
2278 void glue(helper_vpermilps_imm, SUFFIX)(Reg *d, Reg *s, uint32_t order)
2279 {
2280     uint32_t r0, r1, r2, r3;
2281     int i;
2282 
2283     for (i = 0; i < 2 << SHIFT; i += 4) {
2284         r0 = s->L(i + ((order >> 0) & 3));
2285         r1 = s->L(i + ((order >> 2) & 3));
2286         r2 = s->L(i + ((order >> 4) & 3));
2287         r3 = s->L(i + ((order >> 6) & 3));
2288         d->L(i) = r0;
2289         d->L(i+1) = r1;
2290         d->L(i+2) = r2;
2291         d->L(i+3) = r3;
2292     }
2293 }
2294 
2295 #if SHIFT == 1
2296 #define FPSRLVD(x, c) (c < 32 ? ((x) >> c) : 0)
2297 #define FPSRLVQ(x, c) (c < 64 ? ((x) >> c) : 0)
2298 #define FPSRAVD(x, c) ((int32_t)(x) >> (c < 32 ? c : 31))
2299 #define FPSRAVQ(x, c) ((int64_t)(x) >> (c < 64 ? c : 63))
2300 #define FPSLLVD(x, c) (c < 32 ? ((x) << c) : 0)
2301 #define FPSLLVQ(x, c) (c < 64 ? ((x) << c) : 0)
2302 #endif
2303 
2304 SSE_HELPER_L(helper_vpsrlvd, FPSRLVD)
2305 SSE_HELPER_L(helper_vpsravd, FPSRAVD)
2306 SSE_HELPER_L(helper_vpsllvd, FPSLLVD)
2307 
2308 SSE_HELPER_Q(helper_vpsrlvq, FPSRLVQ)
2309 SSE_HELPER_Q(helper_vpsravq, FPSRAVQ)
2310 SSE_HELPER_Q(helper_vpsllvq, FPSLLVQ)
2311 
2312 void glue(helper_vtestps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
2313 {
2314     uint32_t zf = 0, cf = 0;
2315     int i;
2316 
2317     for (i = 0; i < 2 << SHIFT; i++) {
2318         zf |= (s->L(i) &  d->L(i));
2319         cf |= (s->L(i) & ~d->L(i));
2320     }
2321     CC_SRC = ((zf >> 31) ? 0 : CC_Z) | ((cf >> 31) ? 0 : CC_C);
2322 }
2323 
2324 void glue(helper_vtestpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
2325 {
2326     uint64_t zf = 0, cf = 0;
2327     int i;
2328 
2329     for (i = 0; i < 1 << SHIFT; i++) {
2330         zf |= (s->Q(i) &  d->Q(i));
2331         cf |= (s->Q(i) & ~d->Q(i));
2332     }
2333     CC_SRC = ((zf >> 63) ? 0 : CC_Z) | ((cf >> 63) ? 0 : CC_C);
2334 }
2335 
2336 void glue(helper_vpmaskmovd_st, SUFFIX)(CPUX86State *env,
2337                                         Reg *v, Reg *s, target_ulong a0)
2338 {
2339     int i;
2340 
2341     for (i = 0; i < (2 << SHIFT); i++) {
2342         if (v->L(i) >> 31) {
2343             cpu_stl_data_ra(env, a0 + i * 4, s->L(i), GETPC());
2344         }
2345     }
2346 }
2347 
2348 void glue(helper_vpmaskmovq_st, SUFFIX)(CPUX86State *env,
2349                                         Reg *v, Reg *s, target_ulong a0)
2350 {
2351     int i;
2352 
2353     for (i = 0; i < (1 << SHIFT); i++) {
2354         if (v->Q(i) >> 63) {
2355             cpu_stq_data_ra(env, a0 + i * 8, s->Q(i), GETPC());
2356         }
2357     }
2358 }
2359 
2360 void glue(helper_vpmaskmovd, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2361 {
2362     int i;
2363 
2364     for (i = 0; i < (2 << SHIFT); i++) {
2365         d->L(i) = (v->L(i) >> 31) ? s->L(i) : 0;
2366     }
2367 }
2368 
2369 void glue(helper_vpmaskmovq, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2370 {
2371     int i;
2372 
2373     for (i = 0; i < (1 << SHIFT); i++) {
2374         d->Q(i) = (v->Q(i) >> 63) ? s->Q(i) : 0;
2375     }
2376 }
2377 
2378 void glue(helper_vpgatherdd, SUFFIX)(CPUX86State *env,
2379         Reg *d, Reg *v, Reg *s, target_ulong a0, unsigned scale)
2380 {
2381     int i;
2382     for (i = 0; i < (2 << SHIFT); i++) {
2383         if (v->L(i) >> 31) {
2384             target_ulong addr = a0
2385                 + ((target_ulong)(int32_t)s->L(i) << scale);
2386             d->L(i) = cpu_ldl_data_ra(env, addr, GETPC());
2387         }
2388         v->L(i) = 0;
2389     }
2390 }
2391 
2392 void glue(helper_vpgatherdq, SUFFIX)(CPUX86State *env,
2393         Reg *d, Reg *v, Reg *s, target_ulong a0, unsigned scale)
2394 {
2395     int i;
2396     for (i = 0; i < (1 << SHIFT); i++) {
2397         if (v->Q(i) >> 63) {
2398             target_ulong addr = a0
2399                 + ((target_ulong)(int32_t)s->L(i) << scale);
2400             d->Q(i) = cpu_ldq_data_ra(env, addr, GETPC());
2401         }
2402         v->Q(i) = 0;
2403     }
2404 }
2405 
2406 void glue(helper_vpgatherqd, SUFFIX)(CPUX86State *env,
2407         Reg *d, Reg *v, Reg *s, target_ulong a0, unsigned scale)
2408 {
2409     int i;
2410     for (i = 0; i < (1 << SHIFT); i++) {
2411         if (v->L(i) >> 31) {
2412             target_ulong addr = a0
2413                 + ((target_ulong)(int64_t)s->Q(i) << scale);
2414             d->L(i) = cpu_ldl_data_ra(env, addr, GETPC());
2415         }
2416         v->L(i) = 0;
2417     }
2418     for (i /= 2; i < 1 << SHIFT; i++) {
2419         d->Q(i) = 0;
2420         v->Q(i) = 0;
2421     }
2422 }
2423 
2424 void glue(helper_vpgatherqq, SUFFIX)(CPUX86State *env,
2425         Reg *d, Reg *v, Reg *s, target_ulong a0, unsigned scale)
2426 {
2427     int i;
2428     for (i = 0; i < (1 << SHIFT); i++) {
2429         if (v->Q(i) >> 63) {
2430             target_ulong addr = a0
2431                 + ((target_ulong)(int64_t)s->Q(i) << scale);
2432             d->Q(i) = cpu_ldq_data_ra(env, addr, GETPC());
2433         }
2434         v->Q(i) = 0;
2435     }
2436 }
2437 #endif
2438 
2439 #if SHIFT >= 2
2440 void helper_vpermdq_ymm(Reg *d, Reg *v, Reg *s, uint32_t order)
2441 {
2442     uint64_t r0, r1, r2, r3;
2443 
2444     switch (order & 3) {
2445     case 0:
2446         r0 = v->Q(0);
2447         r1 = v->Q(1);
2448         break;
2449     case 1:
2450         r0 = v->Q(2);
2451         r1 = v->Q(3);
2452         break;
2453     case 2:
2454         r0 = s->Q(0);
2455         r1 = s->Q(1);
2456         break;
2457     case 3:
2458         r0 = s->Q(2);
2459         r1 = s->Q(3);
2460         break;
2461     default: /* default case added to help the compiler to avoid warnings */
2462         g_assert_not_reached();
2463     }
2464     switch ((order >> 4) & 3) {
2465     case 0:
2466         r2 = v->Q(0);
2467         r3 = v->Q(1);
2468         break;
2469     case 1:
2470         r2 = v->Q(2);
2471         r3 = v->Q(3);
2472         break;
2473     case 2:
2474         r2 = s->Q(0);
2475         r3 = s->Q(1);
2476         break;
2477     case 3:
2478         r2 = s->Q(2);
2479         r3 = s->Q(3);
2480         break;
2481     default: /* default case added to help the compiler to avoid warnings */
2482         g_assert_not_reached();
2483     }
2484     d->Q(0) = r0;
2485     d->Q(1) = r1;
2486     d->Q(2) = r2;
2487     d->Q(3) = r3;
2488     if (order & 0x8) {
2489         d->Q(0) = 0;
2490         d->Q(1) = 0;
2491     }
2492     if (order & 0x80) {
2493         d->Q(2) = 0;
2494         d->Q(3) = 0;
2495     }
2496 }
2497 
2498 void helper_vpermq_ymm(Reg *d, Reg *s, uint32_t order)
2499 {
2500     uint64_t r0, r1, r2, r3;
2501     r0 = s->Q(order & 3);
2502     r1 = s->Q((order >> 2) & 3);
2503     r2 = s->Q((order >> 4) & 3);
2504     r3 = s->Q((order >> 6) & 3);
2505     d->Q(0) = r0;
2506     d->Q(1) = r1;
2507     d->Q(2) = r2;
2508     d->Q(3) = r3;
2509 }
2510 
2511 void helper_vpermd_ymm(Reg *d, Reg *v, Reg *s)
2512 {
2513     uint32_t r[8];
2514     int i;
2515 
2516     for (i = 0; i < 8; i++) {
2517         r[i] = s->L(v->L(i) & 7);
2518     }
2519     for (i = 0; i < 8; i++) {
2520         d->L(i) = r[i];
2521     }
2522 }
2523 #endif
2524 
2525 /* FMA3 op helpers */
2526 #if SHIFT == 1
2527 #define SSE_HELPER_FMAS(name, elem, F)                                         \
2528     void name(CPUX86State *env, Reg *d, Reg *a, Reg *b, Reg *c, int flags)     \
2529     {                                                                          \
2530         d->elem(0) = F(a->elem(0), b->elem(0), c->elem(0), flags, &env->sse_status); \
2531     }
2532 #define SSE_HELPER_FMAP(name, elem, num, F)                                    \
2533     void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *a, Reg *b, Reg *c,  \
2534                             int flags, int flip)                               \
2535     {                                                                          \
2536         int i;                                                                 \
2537         for (i = 0; i < num; i++) {                                            \
2538             d->elem(i) = F(a->elem(i), b->elem(i), c->elem(i), flags, &env->sse_status); \
2539             flags ^= flip;                                                     \
2540         }                                                                      \
2541     }
2542 
2543 SSE_HELPER_FMAS(helper_fma4ss,  ZMM_S, float32_muladd)
2544 SSE_HELPER_FMAS(helper_fma4sd,  ZMM_D, float64_muladd)
2545 #endif
2546 
2547 #if SHIFT >= 1
2548 SSE_HELPER_FMAP(helper_fma4ps,  ZMM_S, 2 << SHIFT, float32_muladd)
2549 SSE_HELPER_FMAP(helper_fma4pd,  ZMM_D, 1 << SHIFT, float64_muladd)
2550 #endif
2551 
2552 #undef SSE_HELPER_S
2553 
2554 #undef LANE_WIDTH
2555 #undef SHIFT
2556 #undef XMM_ONLY
2557 #undef Reg
2558 #undef B
2559 #undef W
2560 #undef L
2561 #undef Q
2562 #undef SUFFIX
2563