xref: /openbmc/qemu/target/i386/ops_sse.h (revision b14df228)
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 
23 #if SHIFT == 0
24 #define Reg MMXReg
25 #define XMM_ONLY(...)
26 #define B(n) MMX_B(n)
27 #define W(n) MMX_W(n)
28 #define L(n) MMX_L(n)
29 #define Q(n) MMX_Q(n)
30 #define SUFFIX _mmx
31 #else
32 #define Reg ZMMReg
33 #define XMM_ONLY(...) __VA_ARGS__
34 #define B(n) ZMM_B(n)
35 #define W(n) ZMM_W(n)
36 #define L(n) ZMM_L(n)
37 #define Q(n) ZMM_Q(n)
38 #define SUFFIX _xmm
39 #endif
40 
41 #define LANE_WIDTH (SHIFT ? 16 : 8)
42 #define PACK_WIDTH (LANE_WIDTH / 2)
43 
44 #if SHIFT == 0
45 #define FPSRL(x, c) ((x) >> shift)
46 #define FPSRAW(x, c) ((int16_t)(x) >> shift)
47 #define FPSRAL(x, c) ((int32_t)(x) >> shift)
48 #define FPSLL(x, c) ((x) << shift)
49 #endif
50 
51 void glue(helper_psrlw, SUFFIX)(CPUX86State *env, Reg *d, Reg *c)
52 {
53     Reg *s = d;
54     int shift;
55     if (c->Q(0) > 15) {
56         for (int i = 0; i < 1 << SHIFT; i++) {
57             d->Q(i) = 0;
58         }
59     } else {
60         shift = c->B(0);
61         for (int i = 0; i < 4 << SHIFT; i++) {
62             d->W(i) = FPSRL(s->W(i), shift);
63         }
64     }
65 }
66 
67 void glue(helper_psllw, SUFFIX)(CPUX86State *env, Reg *d, Reg *c)
68 {
69     Reg *s = d;
70     int shift;
71     if (c->Q(0) > 15) {
72         for (int i = 0; i < 1 << SHIFT; i++) {
73             d->Q(i) = 0;
74         }
75     } else {
76         shift = c->B(0);
77         for (int i = 0; i < 4 << SHIFT; i++) {
78             d->W(i) = FPSLL(s->W(i), shift);
79         }
80     }
81 }
82 
83 void glue(helper_psraw, SUFFIX)(CPUX86State *env, Reg *d, Reg *c)
84 {
85     Reg *s = d;
86     int shift;
87     if (c->Q(0) > 15) {
88         shift = 15;
89     } else {
90         shift = c->B(0);
91     }
92     for (int i = 0; i < 4 << SHIFT; i++) {
93         d->W(i) = FPSRAW(s->W(i), shift);
94     }
95 }
96 
97 void glue(helper_psrld, SUFFIX)(CPUX86State *env, Reg *d, Reg *c)
98 {
99     Reg *s = d;
100     int shift;
101     if (c->Q(0) > 31) {
102         for (int i = 0; i < 1 << SHIFT; i++) {
103             d->Q(i) = 0;
104         }
105     } else {
106         shift = c->B(0);
107         for (int i = 0; i < 2 << SHIFT; i++) {
108             d->L(i) = FPSRL(s->L(i), shift);
109         }
110     }
111 }
112 
113 void glue(helper_pslld, SUFFIX)(CPUX86State *env, Reg *d, Reg *c)
114 {
115     Reg *s = d;
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 *c)
130 {
131     Reg *s = d;
132     int shift;
133     if (c->Q(0) > 31) {
134         shift = 31;
135     } else {
136         shift = c->B(0);
137     }
138     for (int i = 0; i < 2 << SHIFT; i++) {
139         d->L(i) = FPSRAL(s->L(i), shift);
140     }
141 }
142 
143 void glue(helper_psrlq, SUFFIX)(CPUX86State *env, Reg *d, Reg *c)
144 {
145     Reg *s = d;
146     int shift;
147     if (c->Q(0) > 63) {
148         for (int i = 0; i < 1 << SHIFT; i++) {
149             d->Q(i) = 0;
150         }
151     } else {
152         shift = c->B(0);
153         for (int i = 0; i < 1 << SHIFT; i++) {
154             d->Q(i) = FPSRL(s->Q(i), shift);
155         }
156     }
157 }
158 
159 void glue(helper_psllq, SUFFIX)(CPUX86State *env, Reg *d, Reg *c)
160 {
161     Reg *s = d;
162     int shift;
163     if (c->Q(0) > 63) {
164         for (int i = 0; i < 1 << SHIFT; i++) {
165             d->Q(i) = 0;
166         }
167     } else {
168         shift = c->B(0);
169         for (int i = 0; i < 1 << SHIFT; i++) {
170             d->Q(i) = FPSLL(s->Q(i), shift);
171         }
172     }
173 }
174 
175 #if SHIFT >= 1
176 void glue(helper_psrldq, SUFFIX)(CPUX86State *env, Reg *d, Reg *c)
177 {
178     Reg *s = d;
179     int shift, i, j;
180 
181     shift = c->L(0);
182     if (shift > 16) {
183         shift = 16;
184     }
185     for (j = 0; j < 8 << SHIFT; j += LANE_WIDTH) {
186         for (i = 0; i < 16 - shift; i++) {
187             d->B(j + i) = s->B(j + i + shift);
188         }
189         for (i = 16 - shift; i < 16; i++) {
190             d->B(j + i) = 0;
191         }
192     }
193 }
194 
195 void glue(helper_pslldq, SUFFIX)(CPUX86State *env, Reg *d, Reg *c)
196 {
197     Reg *s = d;
198     int shift, i, j;
199 
200     shift = c->L(0);
201     if (shift > 16) {
202         shift = 16;
203     }
204     for (j = 0; j < 8 << SHIFT; j += LANE_WIDTH) {
205         for (i = 15; i >= shift; i--) {
206             d->B(j + i) = s->B(j + i - shift);
207         }
208         for (i = 0; i < shift; i++) {
209             d->B(j + i) = 0;
210         }
211     }
212 }
213 #endif
214 
215 #define SSE_HELPER_1(name, elem, num, F)                        \
216     void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)   \
217     {                                                           \
218         int n = num;                                            \
219         for (int i = 0; i < n; i++) {                           \
220             d->elem(i) = F(s->elem(i));                         \
221         }                                                       \
222     }
223 
224 #define SSE_HELPER_2(name, elem, num, F)                        \
225     void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)   \
226     {                                                           \
227         Reg *v = d;                                             \
228         int n = num;                                            \
229         for (int i = 0; i < n; i++) {                           \
230             d->elem(i) = F(v->elem(i), s->elem(i));             \
231         }                                                       \
232     }
233 
234 #define SSE_HELPER_B(name, F)                                   \
235     SSE_HELPER_2(name, B, 8 << SHIFT, F)
236 
237 #define SSE_HELPER_W(name, F)                                   \
238     SSE_HELPER_2(name, W, 4 << SHIFT, F)
239 
240 #define SSE_HELPER_L(name, F)                                   \
241     SSE_HELPER_2(name, L, 2 << SHIFT, F)
242 
243 #define SSE_HELPER_Q(name, F)                                   \
244     SSE_HELPER_2(name, Q, 1 << SHIFT, F)
245 
246 #if SHIFT == 0
247 static inline int satub(int x)
248 {
249     if (x < 0) {
250         return 0;
251     } else if (x > 255) {
252         return 255;
253     } else {
254         return x;
255     }
256 }
257 
258 static inline int satuw(int x)
259 {
260     if (x < 0) {
261         return 0;
262     } else if (x > 65535) {
263         return 65535;
264     } else {
265         return x;
266     }
267 }
268 
269 static inline int satsb(int x)
270 {
271     if (x < -128) {
272         return -128;
273     } else if (x > 127) {
274         return 127;
275     } else {
276         return x;
277     }
278 }
279 
280 static inline int satsw(int x)
281 {
282     if (x < -32768) {
283         return -32768;
284     } else if (x > 32767) {
285         return 32767;
286     } else {
287         return x;
288     }
289 }
290 
291 #define FADD(a, b) ((a) + (b))
292 #define FADDUB(a, b) satub((a) + (b))
293 #define FADDUW(a, b) satuw((a) + (b))
294 #define FADDSB(a, b) satsb((int8_t)(a) + (int8_t)(b))
295 #define FADDSW(a, b) satsw((int16_t)(a) + (int16_t)(b))
296 
297 #define FSUB(a, b) ((a) - (b))
298 #define FSUBUB(a, b) satub((a) - (b))
299 #define FSUBUW(a, b) satuw((a) - (b))
300 #define FSUBSB(a, b) satsb((int8_t)(a) - (int8_t)(b))
301 #define FSUBSW(a, b) satsw((int16_t)(a) - (int16_t)(b))
302 #define FMINUB(a, b) ((a) < (b)) ? (a) : (b)
303 #define FMINSW(a, b) ((int16_t)(a) < (int16_t)(b)) ? (a) : (b)
304 #define FMAXUB(a, b) ((a) > (b)) ? (a) : (b)
305 #define FMAXSW(a, b) ((int16_t)(a) > (int16_t)(b)) ? (a) : (b)
306 
307 #define FAND(a, b) ((a) & (b))
308 #define FANDN(a, b) ((~(a)) & (b))
309 #define FOR(a, b) ((a) | (b))
310 #define FXOR(a, b) ((a) ^ (b))
311 
312 #define FCMPGTB(a, b) ((int8_t)(a) > (int8_t)(b) ? -1 : 0)
313 #define FCMPGTW(a, b) ((int16_t)(a) > (int16_t)(b) ? -1 : 0)
314 #define FCMPGTL(a, b) ((int32_t)(a) > (int32_t)(b) ? -1 : 0)
315 #define FCMPEQ(a, b) ((a) == (b) ? -1 : 0)
316 
317 #define FMULLW(a, b) ((a) * (b))
318 #define FMULHRW(a, b) (((int16_t)(a) * (int16_t)(b) + 0x8000) >> 16)
319 #define FMULHUW(a, b) ((a) * (b) >> 16)
320 #define FMULHW(a, b) ((int16_t)(a) * (int16_t)(b) >> 16)
321 
322 #define FAVG(a, b) (((a) + (b) + 1) >> 1)
323 #endif
324 
325 SSE_HELPER_B(helper_paddb, FADD)
326 SSE_HELPER_W(helper_paddw, FADD)
327 SSE_HELPER_L(helper_paddl, FADD)
328 SSE_HELPER_Q(helper_paddq, FADD)
329 
330 SSE_HELPER_B(helper_psubb, FSUB)
331 SSE_HELPER_W(helper_psubw, FSUB)
332 SSE_HELPER_L(helper_psubl, FSUB)
333 SSE_HELPER_Q(helper_psubq, FSUB)
334 
335 SSE_HELPER_B(helper_paddusb, FADDUB)
336 SSE_HELPER_B(helper_paddsb, FADDSB)
337 SSE_HELPER_B(helper_psubusb, FSUBUB)
338 SSE_HELPER_B(helper_psubsb, FSUBSB)
339 
340 SSE_HELPER_W(helper_paddusw, FADDUW)
341 SSE_HELPER_W(helper_paddsw, FADDSW)
342 SSE_HELPER_W(helper_psubusw, FSUBUW)
343 SSE_HELPER_W(helper_psubsw, FSUBSW)
344 
345 SSE_HELPER_B(helper_pminub, FMINUB)
346 SSE_HELPER_B(helper_pmaxub, FMAXUB)
347 
348 SSE_HELPER_W(helper_pminsw, FMINSW)
349 SSE_HELPER_W(helper_pmaxsw, FMAXSW)
350 
351 SSE_HELPER_Q(helper_pand, FAND)
352 SSE_HELPER_Q(helper_pandn, FANDN)
353 SSE_HELPER_Q(helper_por, FOR)
354 SSE_HELPER_Q(helper_pxor, FXOR)
355 
356 SSE_HELPER_B(helper_pcmpgtb, FCMPGTB)
357 SSE_HELPER_W(helper_pcmpgtw, FCMPGTW)
358 SSE_HELPER_L(helper_pcmpgtl, FCMPGTL)
359 
360 SSE_HELPER_B(helper_pcmpeqb, FCMPEQ)
361 SSE_HELPER_W(helper_pcmpeqw, FCMPEQ)
362 SSE_HELPER_L(helper_pcmpeql, FCMPEQ)
363 
364 SSE_HELPER_W(helper_pmullw, FMULLW)
365 #if SHIFT == 0
366 SSE_HELPER_W(helper_pmulhrw, FMULHRW)
367 #endif
368 SSE_HELPER_W(helper_pmulhuw, FMULHUW)
369 SSE_HELPER_W(helper_pmulhw, FMULHW)
370 
371 SSE_HELPER_B(helper_pavgb, FAVG)
372 SSE_HELPER_W(helper_pavgw, FAVG)
373 
374 void glue(helper_pmuludq, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
375 {
376     Reg *v = d;
377     int i;
378 
379     for (i = 0; i < (1 << SHIFT); i++) {
380         d->Q(i) = (uint64_t)s->L(i * 2) * (uint64_t)v->L(i * 2);
381     }
382 }
383 
384 void glue(helper_pmaddwd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
385 {
386     Reg *v = d;
387     int i;
388 
389     for (i = 0; i < (2 << SHIFT); i++) {
390         d->L(i) = (int16_t)s->W(2 * i) * (int16_t)v->W(2 * i) +
391             (int16_t)s->W(2 * i + 1) * (int16_t)v->W(2 * i + 1);
392     }
393 }
394 
395 #if SHIFT == 0
396 static inline int abs1(int a)
397 {
398     if (a < 0) {
399         return -a;
400     } else {
401         return a;
402     }
403 }
404 #endif
405 
406 void glue(helper_psadbw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
407 {
408     Reg *v = d;
409     int i;
410 
411     for (i = 0; i < (1 << SHIFT); i++) {
412         unsigned int val = 0;
413         val += abs1(v->B(8 * i + 0) - s->B(8 * i + 0));
414         val += abs1(v->B(8 * i + 1) - s->B(8 * i + 1));
415         val += abs1(v->B(8 * i + 2) - s->B(8 * i + 2));
416         val += abs1(v->B(8 * i + 3) - s->B(8 * i + 3));
417         val += abs1(v->B(8 * i + 4) - s->B(8 * i + 4));
418         val += abs1(v->B(8 * i + 5) - s->B(8 * i + 5));
419         val += abs1(v->B(8 * i + 6) - s->B(8 * i + 6));
420         val += abs1(v->B(8 * i + 7) - s->B(8 * i + 7));
421         d->Q(i) = val;
422     }
423 }
424 
425 #if SHIFT < 2
426 void glue(helper_maskmov, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
427                                   target_ulong a0)
428 {
429     int i;
430 
431     for (i = 0; i < (8 << SHIFT); i++) {
432         if (s->B(i) & 0x80) {
433             cpu_stb_data_ra(env, a0 + i, d->B(i), GETPC());
434         }
435     }
436 }
437 #endif
438 
439 void glue(helper_movl_mm_T0, SUFFIX)(Reg *d, uint32_t val)
440 {
441     int i;
442 
443     d->L(0) = val;
444     d->L(1) = 0;
445     for (i = 1; i < (1 << SHIFT); i++) {
446         d->Q(i) = 0;
447     }
448 }
449 
450 #ifdef TARGET_X86_64
451 void glue(helper_movq_mm_T0, SUFFIX)(Reg *d, uint64_t val)
452 {
453     int i;
454 
455     d->Q(0) = val;
456     for (i = 1; i < (1 << SHIFT); i++) {
457         d->Q(i) = 0;
458     }
459 }
460 #endif
461 
462 #define SHUFFLE4(F, a, b, offset) do {      \
463     r0 = a->F((order & 3) + offset);        \
464     r1 = a->F(((order >> 2) & 3) + offset); \
465     r2 = b->F(((order >> 4) & 3) + offset); \
466     r3 = b->F(((order >> 6) & 3) + offset); \
467     d->F(offset) = r0;                      \
468     d->F(offset + 1) = r1;                  \
469     d->F(offset + 2) = r2;                  \
470     d->F(offset + 3) = r3;                  \
471     } while (0)
472 
473 #if SHIFT == 0
474 void glue(helper_pshufw, SUFFIX)(Reg *d, Reg *s, int order)
475 {
476     uint16_t r0, r1, r2, r3;
477 
478     SHUFFLE4(W, s, s, 0);
479 }
480 #else
481 void glue(helper_shufps, SUFFIX)(Reg *d, Reg *s, int order)
482 {
483     Reg *v = d;
484     uint32_t r0, r1, r2, r3;
485     int i;
486 
487     for (i = 0; i < 2 << SHIFT; i += 4) {
488         SHUFFLE4(L, v, s, i);
489     }
490 }
491 
492 void glue(helper_shufpd, SUFFIX)(Reg *d, Reg *s, int order)
493 {
494     Reg *v = d;
495     uint64_t r0, r1;
496     int i;
497 
498     for (i = 0; i < 1 << SHIFT; i += 2) {
499         r0 = v->Q(((order & 1) & 1) + i);
500         r1 = s->Q(((order >> 1) & 1) + i);
501         d->Q(i) = r0;
502         d->Q(i + 1) = r1;
503         order >>= 2;
504     }
505 }
506 
507 void glue(helper_pshufd, SUFFIX)(Reg *d, Reg *s, int order)
508 {
509     uint32_t r0, r1, r2, r3;
510     int i;
511 
512     for (i = 0; i < 2 << SHIFT; i += 4) {
513         SHUFFLE4(L, s, s, i);
514     }
515 }
516 
517 void glue(helper_pshuflw, SUFFIX)(Reg *d, Reg *s, int order)
518 {
519     uint16_t r0, r1, r2, r3;
520     int i, j;
521 
522     for (i = 0, j = 1; j < 1 << SHIFT; i += 8, j += 2) {
523         SHUFFLE4(W, s, s, i);
524         d->Q(j) = s->Q(j);
525     }
526 }
527 
528 void glue(helper_pshufhw, SUFFIX)(Reg *d, Reg *s, int order)
529 {
530     uint16_t r0, r1, r2, r3;
531     int i, j;
532 
533     for (i = 4, j = 0; j < 1 << SHIFT; i += 8, j += 2) {
534         d->Q(j) = s->Q(j);
535         SHUFFLE4(W, s, s, i);
536     }
537 }
538 #endif
539 
540 #if SHIFT >= 1
541 /* FPU ops */
542 /* XXX: not accurate */
543 
544 #define SSE_HELPER_P(name, F)                                           \
545     void glue(helper_ ## name ## ps, SUFFIX)(CPUX86State *env,          \
546             Reg *d, Reg *s)                                             \
547     {                                                                   \
548         Reg *v = d;                                                     \
549         int i;                                                          \
550         for (i = 0; i < 2 << SHIFT; i++) {                              \
551             d->ZMM_S(i) = F(32, v->ZMM_S(i), s->ZMM_S(i));              \
552         }                                                               \
553     }                                                                   \
554                                                                         \
555     void glue(helper_ ## name ## pd, SUFFIX)(CPUX86State *env,          \
556             Reg *d, Reg *s)                                     \
557     {                                                                   \
558         Reg *v = d;                                                     \
559         int i;                                                          \
560         for (i = 0; i < 1 << SHIFT; i++) {                              \
561             d->ZMM_D(i) = F(64, v->ZMM_D(i), s->ZMM_D(i));              \
562         }                                                               \
563     }
564 
565 #if SHIFT == 1
566 
567 #define SSE_HELPER_S(name, F)                                           \
568     SSE_HELPER_P(name, F)                                               \
569                                                                         \
570     void helper_ ## name ## ss(CPUX86State *env, Reg *d, Reg *s)\
571     {                                                                   \
572         Reg *v = d;                                                     \
573         d->ZMM_S(0) = F(32, v->ZMM_S(0), s->ZMM_S(0));                  \
574     }                                                                   \
575                                                                         \
576     void helper_ ## name ## sd(CPUX86State *env, Reg *d, Reg *s)\
577     {                                                                   \
578         Reg *v = d;                                                     \
579         d->ZMM_D(0) = F(64, v->ZMM_D(0), s->ZMM_D(0));                  \
580     }
581 
582 #else
583 
584 #define SSE_HELPER_S(name, F) SSE_HELPER_P(name, F)
585 
586 #endif
587 
588 #define FPU_ADD(size, a, b) float ## size ## _add(a, b, &env->sse_status)
589 #define FPU_SUB(size, a, b) float ## size ## _sub(a, b, &env->sse_status)
590 #define FPU_MUL(size, a, b) float ## size ## _mul(a, b, &env->sse_status)
591 #define FPU_DIV(size, a, b) float ## size ## _div(a, b, &env->sse_status)
592 
593 /* Note that the choice of comparison op here is important to get the
594  * special cases right: for min and max Intel specifies that (-0,0),
595  * (NaN, anything) and (anything, NaN) return the second argument.
596  */
597 #define FPU_MIN(size, a, b)                                     \
598     (float ## size ## _lt(a, b, &env->sse_status) ? (a) : (b))
599 #define FPU_MAX(size, a, b)                                     \
600     (float ## size ## _lt(b, a, &env->sse_status) ? (a) : (b))
601 
602 SSE_HELPER_S(add, FPU_ADD)
603 SSE_HELPER_S(sub, FPU_SUB)
604 SSE_HELPER_S(mul, FPU_MUL)
605 SSE_HELPER_S(div, FPU_DIV)
606 SSE_HELPER_S(min, FPU_MIN)
607 SSE_HELPER_S(max, FPU_MAX)
608 
609 void glue(helper_sqrtps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
610 {
611     int i;
612     for (i = 0; i < 2 << SHIFT; i++) {
613         d->ZMM_S(i) = float32_sqrt(s->ZMM_S(i), &env->sse_status);
614     }
615 }
616 
617 void glue(helper_sqrtpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
618 {
619     int i;
620     for (i = 0; i < 1 << SHIFT; i++) {
621         d->ZMM_D(i) = float64_sqrt(s->ZMM_D(i), &env->sse_status);
622     }
623 }
624 
625 #if SHIFT == 1
626 void helper_sqrtss(CPUX86State *env, Reg *d, Reg *s)
627 {
628     d->ZMM_S(0) = float32_sqrt(s->ZMM_S(0), &env->sse_status);
629 }
630 
631 void helper_sqrtsd(CPUX86State *env, Reg *d, Reg *s)
632 {
633     d->ZMM_D(0) = float64_sqrt(s->ZMM_D(0), &env->sse_status);
634 }
635 #endif
636 
637 /* float to float conversions */
638 void glue(helper_cvtps2pd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
639 {
640     int i;
641     for (i = 1 << SHIFT; --i >= 0; ) {
642         d->ZMM_D(i) = float32_to_float64(s->ZMM_S(i), &env->sse_status);
643     }
644 }
645 
646 void glue(helper_cvtpd2ps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
647 {
648     int i;
649     for (i = 0; i < 1 << SHIFT; i++) {
650          d->ZMM_S(i) = float64_to_float32(s->ZMM_D(i), &env->sse_status);
651     }
652     for (i >>= 1; i < 1 << SHIFT; i++) {
653          d->Q(i) = 0;
654     }
655 }
656 
657 #if SHIFT == 1
658 void helper_cvtss2sd(CPUX86State *env, Reg *d, Reg *s)
659 {
660     d->ZMM_D(0) = float32_to_float64(s->ZMM_S(0), &env->sse_status);
661 }
662 
663 void helper_cvtsd2ss(CPUX86State *env, Reg *d, Reg *s)
664 {
665     d->ZMM_S(0) = float64_to_float32(s->ZMM_D(0), &env->sse_status);
666 }
667 #endif
668 
669 /* integer to float */
670 void glue(helper_cvtdq2ps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
671 {
672     int i;
673     for (i = 0; i < 2 << SHIFT; i++) {
674         d->ZMM_S(i) = int32_to_float32(s->ZMM_L(i), &env->sse_status);
675     }
676 }
677 
678 void glue(helper_cvtdq2pd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
679 {
680     int i;
681     for (i = 1 << SHIFT; --i >= 0; ) {
682         int32_t l = s->ZMM_L(i);
683         d->ZMM_D(i) = int32_to_float64(l, &env->sse_status);
684     }
685 }
686 
687 #if SHIFT == 1
688 void helper_cvtpi2ps(CPUX86State *env, ZMMReg *d, MMXReg *s)
689 {
690     d->ZMM_S(0) = int32_to_float32(s->MMX_L(0), &env->sse_status);
691     d->ZMM_S(1) = int32_to_float32(s->MMX_L(1), &env->sse_status);
692 }
693 
694 void helper_cvtpi2pd(CPUX86State *env, ZMMReg *d, MMXReg *s)
695 {
696     d->ZMM_D(0) = int32_to_float64(s->MMX_L(0), &env->sse_status);
697     d->ZMM_D(1) = int32_to_float64(s->MMX_L(1), &env->sse_status);
698 }
699 
700 void helper_cvtsi2ss(CPUX86State *env, ZMMReg *d, uint32_t val)
701 {
702     d->ZMM_S(0) = int32_to_float32(val, &env->sse_status);
703 }
704 
705 void helper_cvtsi2sd(CPUX86State *env, ZMMReg *d, uint32_t val)
706 {
707     d->ZMM_D(0) = int32_to_float64(val, &env->sse_status);
708 }
709 
710 #ifdef TARGET_X86_64
711 void helper_cvtsq2ss(CPUX86State *env, ZMMReg *d, uint64_t val)
712 {
713     d->ZMM_S(0) = int64_to_float32(val, &env->sse_status);
714 }
715 
716 void helper_cvtsq2sd(CPUX86State *env, ZMMReg *d, uint64_t val)
717 {
718     d->ZMM_D(0) = int64_to_float64(val, &env->sse_status);
719 }
720 #endif
721 
722 #endif
723 
724 /* float to integer */
725 
726 #if SHIFT == 1
727 /*
728  * x86 mandates that we return the indefinite integer value for the result
729  * of any float-to-integer conversion that raises the 'invalid' exception.
730  * Wrap the softfloat functions to get this behaviour.
731  */
732 #define WRAP_FLOATCONV(RETTYPE, FN, FLOATTYPE, INDEFVALUE)              \
733     static inline RETTYPE x86_##FN(FLOATTYPE a, float_status *s)        \
734     {                                                                   \
735         int oldflags, newflags;                                         \
736         RETTYPE r;                                                      \
737                                                                         \
738         oldflags = get_float_exception_flags(s);                        \
739         set_float_exception_flags(0, s);                                \
740         r = FN(a, s);                                                   \
741         newflags = get_float_exception_flags(s);                        \
742         if (newflags & float_flag_invalid) {                            \
743             r = INDEFVALUE;                                             \
744         }                                                               \
745         set_float_exception_flags(newflags | oldflags, s);              \
746         return r;                                                       \
747     }
748 
749 WRAP_FLOATCONV(int32_t, float32_to_int32, float32, INT32_MIN)
750 WRAP_FLOATCONV(int32_t, float32_to_int32_round_to_zero, float32, INT32_MIN)
751 WRAP_FLOATCONV(int32_t, float64_to_int32, float64, INT32_MIN)
752 WRAP_FLOATCONV(int32_t, float64_to_int32_round_to_zero, float64, INT32_MIN)
753 WRAP_FLOATCONV(int64_t, float32_to_int64, float32, INT64_MIN)
754 WRAP_FLOATCONV(int64_t, float32_to_int64_round_to_zero, float32, INT64_MIN)
755 WRAP_FLOATCONV(int64_t, float64_to_int64, float64, INT64_MIN)
756 WRAP_FLOATCONV(int64_t, float64_to_int64_round_to_zero, float64, INT64_MIN)
757 #endif
758 
759 void glue(helper_cvtps2dq, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
760 {
761     int i;
762     for (i = 0; i < 2 << SHIFT; i++) {
763         d->ZMM_L(i) = x86_float32_to_int32(s->ZMM_S(i), &env->sse_status);
764     }
765 }
766 
767 void glue(helper_cvtpd2dq, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
768 {
769     int i;
770     for (i = 0; i < 1 << SHIFT; i++) {
771         d->ZMM_L(i) = x86_float64_to_int32(s->ZMM_D(i), &env->sse_status);
772     }
773     for (i >>= 1; i < 1 << SHIFT; i++) {
774          d->Q(i) = 0;
775     }
776 }
777 
778 #if SHIFT == 1
779 void helper_cvtps2pi(CPUX86State *env, MMXReg *d, ZMMReg *s)
780 {
781     d->MMX_L(0) = x86_float32_to_int32(s->ZMM_S(0), &env->sse_status);
782     d->MMX_L(1) = x86_float32_to_int32(s->ZMM_S(1), &env->sse_status);
783 }
784 
785 void helper_cvtpd2pi(CPUX86State *env, MMXReg *d, ZMMReg *s)
786 {
787     d->MMX_L(0) = x86_float64_to_int32(s->ZMM_D(0), &env->sse_status);
788     d->MMX_L(1) = x86_float64_to_int32(s->ZMM_D(1), &env->sse_status);
789 }
790 
791 int32_t helper_cvtss2si(CPUX86State *env, ZMMReg *s)
792 {
793     return x86_float32_to_int32(s->ZMM_S(0), &env->sse_status);
794 }
795 
796 int32_t helper_cvtsd2si(CPUX86State *env, ZMMReg *s)
797 {
798     return x86_float64_to_int32(s->ZMM_D(0), &env->sse_status);
799 }
800 
801 #ifdef TARGET_X86_64
802 int64_t helper_cvtss2sq(CPUX86State *env, ZMMReg *s)
803 {
804     return x86_float32_to_int64(s->ZMM_S(0), &env->sse_status);
805 }
806 
807 int64_t helper_cvtsd2sq(CPUX86State *env, ZMMReg *s)
808 {
809     return x86_float64_to_int64(s->ZMM_D(0), &env->sse_status);
810 }
811 #endif
812 #endif
813 
814 /* float to integer truncated */
815 void glue(helper_cvttps2dq, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
816 {
817     int i;
818     for (i = 0; i < 2 << SHIFT; i++) {
819         d->ZMM_L(i) = x86_float32_to_int32_round_to_zero(s->ZMM_S(i),
820                                                          &env->sse_status);
821     }
822 }
823 
824 void glue(helper_cvttpd2dq, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
825 {
826     int i;
827     for (i = 0; i < 1 << SHIFT; i++) {
828         d->ZMM_L(i) = x86_float64_to_int32_round_to_zero(s->ZMM_D(i),
829                                                          &env->sse_status);
830     }
831     for (i >>= 1; i < 1 << SHIFT; i++) {
832          d->Q(i) = 0;
833     }
834 }
835 
836 #if SHIFT == 1
837 void helper_cvttps2pi(CPUX86State *env, MMXReg *d, ZMMReg *s)
838 {
839     d->MMX_L(0) = x86_float32_to_int32_round_to_zero(s->ZMM_S(0), &env->sse_status);
840     d->MMX_L(1) = x86_float32_to_int32_round_to_zero(s->ZMM_S(1), &env->sse_status);
841 }
842 
843 void helper_cvttpd2pi(CPUX86State *env, MMXReg *d, ZMMReg *s)
844 {
845     d->MMX_L(0) = x86_float64_to_int32_round_to_zero(s->ZMM_D(0), &env->sse_status);
846     d->MMX_L(1) = x86_float64_to_int32_round_to_zero(s->ZMM_D(1), &env->sse_status);
847 }
848 
849 int32_t helper_cvttss2si(CPUX86State *env, ZMMReg *s)
850 {
851     return x86_float32_to_int32_round_to_zero(s->ZMM_S(0), &env->sse_status);
852 }
853 
854 int32_t helper_cvttsd2si(CPUX86State *env, ZMMReg *s)
855 {
856     return x86_float64_to_int32_round_to_zero(s->ZMM_D(0), &env->sse_status);
857 }
858 
859 #ifdef TARGET_X86_64
860 int64_t helper_cvttss2sq(CPUX86State *env, ZMMReg *s)
861 {
862     return x86_float32_to_int64_round_to_zero(s->ZMM_S(0), &env->sse_status);
863 }
864 
865 int64_t helper_cvttsd2sq(CPUX86State *env, ZMMReg *s)
866 {
867     return x86_float64_to_int64_round_to_zero(s->ZMM_D(0), &env->sse_status);
868 }
869 #endif
870 #endif
871 
872 void glue(helper_rsqrtps, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
873 {
874     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
875     int i;
876     for (i = 0; i < 2 << SHIFT; i++) {
877         d->ZMM_S(i) = float32_div(float32_one,
878                                   float32_sqrt(s->ZMM_S(i), &env->sse_status),
879                                   &env->sse_status);
880     }
881     set_float_exception_flags(old_flags, &env->sse_status);
882 }
883 
884 #if SHIFT == 1
885 void helper_rsqrtss(CPUX86State *env, ZMMReg *d, ZMMReg *s)
886 {
887     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
888     d->ZMM_S(0) = float32_div(float32_one,
889                               float32_sqrt(s->ZMM_S(0), &env->sse_status),
890                               &env->sse_status);
891     set_float_exception_flags(old_flags, &env->sse_status);
892 }
893 #endif
894 
895 void glue(helper_rcpps, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
896 {
897     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
898     int i;
899     for (i = 0; i < 2 << SHIFT; i++) {
900         d->ZMM_S(i) = float32_div(float32_one, s->ZMM_S(i), &env->sse_status);
901     }
902     set_float_exception_flags(old_flags, &env->sse_status);
903 }
904 
905 #if SHIFT == 1
906 void helper_rcpss(CPUX86State *env, ZMMReg *d, ZMMReg *s)
907 {
908     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
909     d->ZMM_S(0) = float32_div(float32_one, s->ZMM_S(0), &env->sse_status);
910     set_float_exception_flags(old_flags, &env->sse_status);
911 }
912 #endif
913 
914 #if SHIFT == 1
915 static inline uint64_t helper_extrq(uint64_t src, int shift, int len)
916 {
917     uint64_t mask;
918 
919     if (len == 0) {
920         mask = ~0LL;
921     } else {
922         mask = (1ULL << len) - 1;
923     }
924     return (src >> shift) & mask;
925 }
926 
927 void helper_extrq_r(CPUX86State *env, ZMMReg *d, ZMMReg *s)
928 {
929     d->ZMM_Q(0) = helper_extrq(d->ZMM_Q(0), s->ZMM_B(1), s->ZMM_B(0));
930 }
931 
932 void helper_extrq_i(CPUX86State *env, ZMMReg *d, int index, int length)
933 {
934     d->ZMM_Q(0) = helper_extrq(d->ZMM_Q(0), index, length);
935 }
936 
937 static inline uint64_t helper_insertq(uint64_t src, int shift, int len)
938 {
939     uint64_t mask;
940 
941     if (len == 0) {
942         mask = ~0ULL;
943     } else {
944         mask = (1ULL << len) - 1;
945     }
946     return (src & ~(mask << shift)) | ((src & mask) << shift);
947 }
948 
949 void helper_insertq_r(CPUX86State *env, ZMMReg *d, ZMMReg *s)
950 {
951     d->ZMM_Q(0) = helper_insertq(s->ZMM_Q(0), s->ZMM_B(9), s->ZMM_B(8));
952 }
953 
954 void helper_insertq_i(CPUX86State *env, ZMMReg *d, int index, int length)
955 {
956     d->ZMM_Q(0) = helper_insertq(d->ZMM_Q(0), index, length);
957 }
958 #endif
959 
960 #define SSE_HELPER_HPS(name, F)  \
961 void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s) \
962 {                                                                 \
963     Reg *v = d;                                                   \
964     float32 r[2 << SHIFT];                                        \
965     int i, j, k;                                                  \
966     for (k = 0; k < 2 << SHIFT; k += LANE_WIDTH / 4) {            \
967         for (i = j = 0; j < 4; i++, j += 2) {                     \
968             r[i + k] = F(v->ZMM_S(j + k), v->ZMM_S(j + k + 1), &env->sse_status); \
969         }                                                         \
970         for (j = 0; j < 4; i++, j += 2) {                         \
971             r[i + k] = F(s->ZMM_S(j + k), s->ZMM_S(j + k + 1), &env->sse_status); \
972         }                                                         \
973     }                                                             \
974     for (i = 0; i < 2 << SHIFT; i++) {                            \
975         d->ZMM_S(i) = r[i];                                       \
976     }                                                             \
977 }
978 
979 SSE_HELPER_HPS(haddps, float32_add)
980 SSE_HELPER_HPS(hsubps, float32_sub)
981 
982 #define SSE_HELPER_HPD(name, F)  \
983 void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s) \
984 {                                                                 \
985     Reg *v = d;                                                   \
986     float64 r[1 << SHIFT];                                        \
987     int i, j, k;                                                  \
988     for (k = 0; k < 1 << SHIFT; k += LANE_WIDTH / 8) {            \
989         for (i = j = 0; j < 2; i++, j += 2) {                     \
990             r[i + k] = F(v->ZMM_D(j + k), v->ZMM_D(j + k + 1), &env->sse_status); \
991         }                                                         \
992         for (j = 0; j < 2; i++, j += 2) {                         \
993             r[i + k] = F(s->ZMM_D(j + k), s->ZMM_D(j + k + 1), &env->sse_status); \
994         }                                                         \
995     }                                                             \
996     for (i = 0; i < 1 << SHIFT; i++) {                            \
997         d->ZMM_D(i) = r[i];                                       \
998     }                                                             \
999 }
1000 
1001 SSE_HELPER_HPD(haddpd, float64_add)
1002 SSE_HELPER_HPD(hsubpd, float64_sub)
1003 
1004 void glue(helper_addsubps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
1005 {
1006     Reg *v = d;
1007     int i;
1008     for (i = 0; i < 2 << SHIFT; i += 2) {
1009         d->ZMM_S(i) = float32_sub(v->ZMM_S(i), s->ZMM_S(i), &env->sse_status);
1010         d->ZMM_S(i+1) = float32_add(v->ZMM_S(i+1), s->ZMM_S(i+1), &env->sse_status);
1011     }
1012 }
1013 
1014 void glue(helper_addsubpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
1015 {
1016     Reg *v = d;
1017     int i;
1018     for (i = 0; i < 1 << SHIFT; i += 2) {
1019         d->ZMM_D(i) = float64_sub(v->ZMM_D(i), s->ZMM_D(i), &env->sse_status);
1020         d->ZMM_D(i+1) = float64_add(v->ZMM_D(i+1), s->ZMM_D(i+1), &env->sse_status);
1021     }
1022 }
1023 
1024 #define SSE_HELPER_CMP_P(name, F, C)                                    \
1025     void glue(helper_ ## name ## ps, SUFFIX)(CPUX86State *env,          \
1026                                              Reg *d, Reg *s)    \
1027     {                                                                   \
1028         Reg *v = d;                                                     \
1029         int i;                                                          \
1030         for (i = 0; i < 2 << SHIFT; i++) {                              \
1031             d->ZMM_L(i) = C(F(32, v->ZMM_S(i), s->ZMM_S(i))) ? -1 : 0;  \
1032         }                                                               \
1033     }                                                                   \
1034                                                                         \
1035     void glue(helper_ ## name ## pd, SUFFIX)(CPUX86State *env,          \
1036                                              Reg *d, Reg *s)    \
1037     {                                                                   \
1038         Reg *v = d;                                                     \
1039         int i;                                                          \
1040         for (i = 0; i < 1 << SHIFT; i++) {                              \
1041             d->ZMM_Q(i) = C(F(64, v->ZMM_D(i), s->ZMM_D(i))) ? -1 : 0;  \
1042         }                                                               \
1043     }
1044 
1045 #if SHIFT == 1
1046 #define SSE_HELPER_CMP(name, F, C)                                          \
1047     SSE_HELPER_CMP_P(name, F, C)                                            \
1048     void helper_ ## name ## ss(CPUX86State *env, Reg *d, Reg *s)    \
1049     {                                                                       \
1050         Reg *v = d;                                                         \
1051         d->ZMM_L(0) = C(F(32, v->ZMM_S(0), s->ZMM_S(0))) ? -1 : 0;          \
1052     }                                                                       \
1053                                                                             \
1054     void helper_ ## name ## sd(CPUX86State *env, Reg *d, Reg *s)    \
1055     {                                                                       \
1056         Reg *v = d;                                                         \
1057         d->ZMM_Q(0) = C(F(64, v->ZMM_D(0), s->ZMM_D(0))) ? -1 : 0;          \
1058     }
1059 
1060 #define FPU_EQ(x) (x == float_relation_equal)
1061 #define FPU_LT(x) (x == float_relation_less)
1062 #define FPU_LE(x) (x <= float_relation_equal)
1063 #define FPU_UNORD(x) (x == float_relation_unordered)
1064 
1065 #define FPU_CMPQ(size, a, b) \
1066     float ## size ## _compare_quiet(a, b, &env->sse_status)
1067 #define FPU_CMPS(size, a, b) \
1068     float ## size ## _compare(a, b, &env->sse_status)
1069 
1070 #else
1071 #define SSE_HELPER_CMP(name, F, C) SSE_HELPER_CMP_P(name, F, C)
1072 #endif
1073 
1074 SSE_HELPER_CMP(cmpeq, FPU_CMPQ, FPU_EQ)
1075 SSE_HELPER_CMP(cmplt, FPU_CMPS, FPU_LT)
1076 SSE_HELPER_CMP(cmple, FPU_CMPS, FPU_LE)
1077 SSE_HELPER_CMP(cmpunord, FPU_CMPQ,  FPU_UNORD)
1078 SSE_HELPER_CMP(cmpneq, FPU_CMPQ, !FPU_EQ)
1079 SSE_HELPER_CMP(cmpnlt, FPU_CMPS, !FPU_LT)
1080 SSE_HELPER_CMP(cmpnle, FPU_CMPS, !FPU_LE)
1081 SSE_HELPER_CMP(cmpord, FPU_CMPQ, !FPU_UNORD)
1082 
1083 #undef SSE_HELPER_CMP
1084 
1085 #if SHIFT == 1
1086 static const int comis_eflags[4] = {CC_C, CC_Z, 0, CC_Z | CC_P | CC_C};
1087 
1088 void helper_ucomiss(CPUX86State *env, Reg *d, Reg *s)
1089 {
1090     FloatRelation ret;
1091     float32 s0, s1;
1092 
1093     s0 = d->ZMM_S(0);
1094     s1 = s->ZMM_S(0);
1095     ret = float32_compare_quiet(s0, s1, &env->sse_status);
1096     CC_SRC = comis_eflags[ret + 1];
1097 }
1098 
1099 void helper_comiss(CPUX86State *env, Reg *d, Reg *s)
1100 {
1101     FloatRelation ret;
1102     float32 s0, s1;
1103 
1104     s0 = d->ZMM_S(0);
1105     s1 = s->ZMM_S(0);
1106     ret = float32_compare(s0, s1, &env->sse_status);
1107     CC_SRC = comis_eflags[ret + 1];
1108 }
1109 
1110 void helper_ucomisd(CPUX86State *env, Reg *d, Reg *s)
1111 {
1112     FloatRelation ret;
1113     float64 d0, d1;
1114 
1115     d0 = d->ZMM_D(0);
1116     d1 = s->ZMM_D(0);
1117     ret = float64_compare_quiet(d0, d1, &env->sse_status);
1118     CC_SRC = comis_eflags[ret + 1];
1119 }
1120 
1121 void helper_comisd(CPUX86State *env, Reg *d, Reg *s)
1122 {
1123     FloatRelation ret;
1124     float64 d0, d1;
1125 
1126     d0 = d->ZMM_D(0);
1127     d1 = s->ZMM_D(0);
1128     ret = float64_compare(d0, d1, &env->sse_status);
1129     CC_SRC = comis_eflags[ret + 1];
1130 }
1131 #endif
1132 
1133 uint32_t glue(helper_movmskps, SUFFIX)(CPUX86State *env, Reg *s)
1134 {
1135     uint32_t mask;
1136     int i;
1137 
1138     mask = 0;
1139     for (i = 0; i < 2 << SHIFT; i++) {
1140         mask |= (s->ZMM_L(i) >> (31 - i)) & (1 << i);
1141     }
1142     return mask;
1143 }
1144 
1145 uint32_t glue(helper_movmskpd, SUFFIX)(CPUX86State *env, Reg *s)
1146 {
1147     uint32_t mask;
1148     int i;
1149 
1150     mask = 0;
1151     for (i = 0; i < 1 << SHIFT; i++) {
1152         mask |= (s->ZMM_Q(i) >> (63 - i)) & (1 << i);
1153     }
1154     return mask;
1155 }
1156 
1157 #endif
1158 
1159 uint32_t glue(helper_pmovmskb, SUFFIX)(CPUX86State *env, Reg *s)
1160 {
1161     uint32_t val;
1162     int i;
1163 
1164     val = 0;
1165     for (i = 0; i < (1 << SHIFT); i++) {
1166         uint8_t byte = 0;
1167         byte |= (s->B(8 * i + 0) >> 7);
1168         byte |= (s->B(8 * i + 1) >> 6) & 0x02;
1169         byte |= (s->B(8 * i + 2) >> 5) & 0x04;
1170         byte |= (s->B(8 * i + 3) >> 4) & 0x08;
1171         byte |= (s->B(8 * i + 4) >> 3) & 0x10;
1172         byte |= (s->B(8 * i + 5) >> 2) & 0x20;
1173         byte |= (s->B(8 * i + 6) >> 1) & 0x40;
1174         byte |= (s->B(8 * i + 7)) & 0x80;
1175         val |= byte << (8 * i);
1176     }
1177     return val;
1178 }
1179 
1180 #define PACK_HELPER_B(name, F) \
1181 void glue(helper_pack ## name, SUFFIX)(CPUX86State *env,      \
1182         Reg *d, Reg *s)                                       \
1183 {                                                             \
1184     Reg *v = d;                                               \
1185     uint8_t r[PACK_WIDTH * 2];                                \
1186     int j, k;                                                 \
1187     for (j = 0; j < 4 << SHIFT; j += PACK_WIDTH) {            \
1188         for (k = 0; k < PACK_WIDTH; k++) {                    \
1189             r[k] = F((int16_t)v->W(j + k));                   \
1190         }                                                     \
1191         for (k = 0; k < PACK_WIDTH; k++) {                    \
1192             r[PACK_WIDTH + k] = F((int16_t)s->W(j + k));      \
1193         }                                                     \
1194         for (k = 0; k < PACK_WIDTH * 2; k++) {                \
1195             d->B(2 * j + k) = r[k];                           \
1196         }                                                     \
1197     }                                                         \
1198 }
1199 
1200 PACK_HELPER_B(sswb, satsb)
1201 PACK_HELPER_B(uswb, satub)
1202 
1203 void glue(helper_packssdw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
1204 {
1205     Reg *v = d;
1206     uint16_t r[PACK_WIDTH];
1207     int j, k;
1208 
1209     for (j = 0; j < 2 << SHIFT; j += PACK_WIDTH / 2) {
1210         for (k = 0; k < PACK_WIDTH / 2; k++) {
1211             r[k] = satsw(v->L(j + k));
1212         }
1213         for (k = 0; k < PACK_WIDTH / 2; k++) {
1214             r[PACK_WIDTH / 2 + k] = satsw(s->L(j + k));
1215         }
1216         for (k = 0; k < PACK_WIDTH; k++) {
1217             d->W(2 * j + k) = r[k];
1218         }
1219     }
1220 }
1221 
1222 #define UNPCK_OP(base_name, base)                                       \
1223                                                                         \
1224     void glue(helper_punpck ## base_name ## bw, SUFFIX)(CPUX86State *env,\
1225                                                 Reg *d, Reg *s) \
1226     {                                                                   \
1227         Reg *v = d;                                                     \
1228         uint8_t r[PACK_WIDTH * 2];                                      \
1229         int j, i;                                                       \
1230                                                                         \
1231         for (j = 0; j < 8 << SHIFT; ) {                                 \
1232             int k = j + base * PACK_WIDTH;                              \
1233             for (i = 0; i < PACK_WIDTH; i++) {                          \
1234                 r[2 * i] = v->B(k + i);                                 \
1235                 r[2 * i + 1] = s->B(k + i);                             \
1236             }                                                           \
1237             for (i = 0; i < PACK_WIDTH * 2; i++, j++) {                 \
1238                 d->B(j) = r[i];                                         \
1239             }                                                           \
1240         }                                                               \
1241     }                                                                   \
1242                                                                         \
1243     void glue(helper_punpck ## base_name ## wd, SUFFIX)(CPUX86State *env,\
1244                                                 Reg *d, Reg *s) \
1245     {                                                                   \
1246         Reg *v = d;                                                     \
1247         uint16_t r[PACK_WIDTH];                                         \
1248         int j, i;                                                       \
1249                                                                         \
1250         for (j = 0; j < 4 << SHIFT; ) {                                 \
1251             int k = j + base * PACK_WIDTH / 2;                          \
1252             for (i = 0; i < PACK_WIDTH / 2; i++) {                      \
1253                 r[2 * i] = v->W(k + i);                                 \
1254                 r[2 * i + 1] = s->W(k + i);                             \
1255             }                                                           \
1256             for (i = 0; i < PACK_WIDTH; i++, j++) {                     \
1257                 d->W(j) = r[i];                                         \
1258             }                                                           \
1259         }                                                               \
1260     }                                                                   \
1261                                                                         \
1262     void glue(helper_punpck ## base_name ## dq, SUFFIX)(CPUX86State *env,\
1263                                                 Reg *d, Reg *s) \
1264     {                                                                   \
1265         Reg *v = d;                                                     \
1266         uint32_t r[PACK_WIDTH / 2];                                     \
1267         int j, i;                                                       \
1268                                                                         \
1269         for (j = 0; j < 2 << SHIFT; ) {                                 \
1270             int k = j + base * PACK_WIDTH / 4;                          \
1271             for (i = 0; i < PACK_WIDTH / 4; i++) {                      \
1272                 r[2 * i] = v->L(k + i);                                 \
1273                 r[2 * i + 1] = s->L(k + i);                             \
1274             }                                                           \
1275             for (i = 0; i < PACK_WIDTH / 2; i++, j++) {                 \
1276                 d->L(j) = r[i];                                         \
1277             }                                                           \
1278         }                                                               \
1279     }                                                                   \
1280                                                                         \
1281     XMM_ONLY(                                                           \
1282              void glue(helper_punpck ## base_name ## qdq, SUFFIX)(      \
1283                         CPUX86State *env, Reg *d, Reg *s)       \
1284              {                                                          \
1285                  Reg *v = d;                                            \
1286                  uint64_t r[2];                                         \
1287                  int i;                                                 \
1288                                                                         \
1289                  for (i = 0; i < 1 << SHIFT; i += 2) {                  \
1290                      r[0] = v->Q(base + i);                             \
1291                      r[1] = s->Q(base + i);                             \
1292                      d->Q(i) = r[0];                                    \
1293                      d->Q(i + 1) = r[1];                                \
1294                  }                                                      \
1295              }                                                          \
1296                                                                         )
1297 
1298 UNPCK_OP(l, 0)
1299 UNPCK_OP(h, 1)
1300 
1301 #undef PACK_WIDTH
1302 #undef PACK_HELPER_B
1303 #undef UNPCK_OP
1304 
1305 
1306 /* 3DNow! float ops */
1307 #if SHIFT == 0
1308 void helper_pi2fd(CPUX86State *env, MMXReg *d, MMXReg *s)
1309 {
1310     d->MMX_S(0) = int32_to_float32(s->MMX_L(0), &env->mmx_status);
1311     d->MMX_S(1) = int32_to_float32(s->MMX_L(1), &env->mmx_status);
1312 }
1313 
1314 void helper_pi2fw(CPUX86State *env, MMXReg *d, MMXReg *s)
1315 {
1316     d->MMX_S(0) = int32_to_float32((int16_t)s->MMX_W(0), &env->mmx_status);
1317     d->MMX_S(1) = int32_to_float32((int16_t)s->MMX_W(2), &env->mmx_status);
1318 }
1319 
1320 void helper_pf2id(CPUX86State *env, MMXReg *d, MMXReg *s)
1321 {
1322     d->MMX_L(0) = float32_to_int32_round_to_zero(s->MMX_S(0), &env->mmx_status);
1323     d->MMX_L(1) = float32_to_int32_round_to_zero(s->MMX_S(1), &env->mmx_status);
1324 }
1325 
1326 void helper_pf2iw(CPUX86State *env, MMXReg *d, MMXReg *s)
1327 {
1328     d->MMX_L(0) = satsw(float32_to_int32_round_to_zero(s->MMX_S(0),
1329                                                        &env->mmx_status));
1330     d->MMX_L(1) = satsw(float32_to_int32_round_to_zero(s->MMX_S(1),
1331                                                        &env->mmx_status));
1332 }
1333 
1334 void helper_pfacc(CPUX86State *env, MMXReg *d, MMXReg *s)
1335 {
1336     float32 r;
1337 
1338     r = float32_add(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1339     d->MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1340     d->MMX_S(0) = r;
1341 }
1342 
1343 void helper_pfadd(CPUX86State *env, MMXReg *d, MMXReg *s)
1344 {
1345     d->MMX_S(0) = float32_add(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1346     d->MMX_S(1) = float32_add(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1347 }
1348 
1349 void helper_pfcmpeq(CPUX86State *env, MMXReg *d, MMXReg *s)
1350 {
1351     d->MMX_L(0) = float32_eq_quiet(d->MMX_S(0), s->MMX_S(0),
1352                                    &env->mmx_status) ? -1 : 0;
1353     d->MMX_L(1) = float32_eq_quiet(d->MMX_S(1), s->MMX_S(1),
1354                                    &env->mmx_status) ? -1 : 0;
1355 }
1356 
1357 void helper_pfcmpge(CPUX86State *env, MMXReg *d, MMXReg *s)
1358 {
1359     d->MMX_L(0) = float32_le(s->MMX_S(0), d->MMX_S(0),
1360                              &env->mmx_status) ? -1 : 0;
1361     d->MMX_L(1) = float32_le(s->MMX_S(1), d->MMX_S(1),
1362                              &env->mmx_status) ? -1 : 0;
1363 }
1364 
1365 void helper_pfcmpgt(CPUX86State *env, MMXReg *d, MMXReg *s)
1366 {
1367     d->MMX_L(0) = float32_lt(s->MMX_S(0), d->MMX_S(0),
1368                              &env->mmx_status) ? -1 : 0;
1369     d->MMX_L(1) = float32_lt(s->MMX_S(1), d->MMX_S(1),
1370                              &env->mmx_status) ? -1 : 0;
1371 }
1372 
1373 void helper_pfmax(CPUX86State *env, MMXReg *d, MMXReg *s)
1374 {
1375     if (float32_lt(d->MMX_S(0), s->MMX_S(0), &env->mmx_status)) {
1376         d->MMX_S(0) = s->MMX_S(0);
1377     }
1378     if (float32_lt(d->MMX_S(1), s->MMX_S(1), &env->mmx_status)) {
1379         d->MMX_S(1) = s->MMX_S(1);
1380     }
1381 }
1382 
1383 void helper_pfmin(CPUX86State *env, MMXReg *d, MMXReg *s)
1384 {
1385     if (float32_lt(s->MMX_S(0), d->MMX_S(0), &env->mmx_status)) {
1386         d->MMX_S(0) = s->MMX_S(0);
1387     }
1388     if (float32_lt(s->MMX_S(1), d->MMX_S(1), &env->mmx_status)) {
1389         d->MMX_S(1) = s->MMX_S(1);
1390     }
1391 }
1392 
1393 void helper_pfmul(CPUX86State *env, MMXReg *d, MMXReg *s)
1394 {
1395     d->MMX_S(0) = float32_mul(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1396     d->MMX_S(1) = float32_mul(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1397 }
1398 
1399 void helper_pfnacc(CPUX86State *env, MMXReg *d, MMXReg *s)
1400 {
1401     float32 r;
1402 
1403     r = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1404     d->MMX_S(1) = float32_sub(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1405     d->MMX_S(0) = r;
1406 }
1407 
1408 void helper_pfpnacc(CPUX86State *env, MMXReg *d, MMXReg *s)
1409 {
1410     float32 r;
1411 
1412     r = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1413     d->MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1414     d->MMX_S(0) = r;
1415 }
1416 
1417 void helper_pfrcp(CPUX86State *env, MMXReg *d, MMXReg *s)
1418 {
1419     d->MMX_S(0) = float32_div(float32_one, s->MMX_S(0), &env->mmx_status);
1420     d->MMX_S(1) = d->MMX_S(0);
1421 }
1422 
1423 void helper_pfrsqrt(CPUX86State *env, MMXReg *d, MMXReg *s)
1424 {
1425     d->MMX_L(1) = s->MMX_L(0) & 0x7fffffff;
1426     d->MMX_S(1) = float32_div(float32_one,
1427                               float32_sqrt(d->MMX_S(1), &env->mmx_status),
1428                               &env->mmx_status);
1429     d->MMX_L(1) |= s->MMX_L(0) & 0x80000000;
1430     d->MMX_L(0) = d->MMX_L(1);
1431 }
1432 
1433 void helper_pfsub(CPUX86State *env, MMXReg *d, MMXReg *s)
1434 {
1435     d->MMX_S(0) = float32_sub(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1436     d->MMX_S(1) = float32_sub(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1437 }
1438 
1439 void helper_pfsubr(CPUX86State *env, MMXReg *d, MMXReg *s)
1440 {
1441     d->MMX_S(0) = float32_sub(s->MMX_S(0), d->MMX_S(0), &env->mmx_status);
1442     d->MMX_S(1) = float32_sub(s->MMX_S(1), d->MMX_S(1), &env->mmx_status);
1443 }
1444 
1445 void helper_pswapd(CPUX86State *env, MMXReg *d, MMXReg *s)
1446 {
1447     uint32_t r;
1448 
1449     r = s->MMX_L(0);
1450     d->MMX_L(0) = s->MMX_L(1);
1451     d->MMX_L(1) = r;
1452 }
1453 #endif
1454 
1455 /* SSSE3 op helpers */
1456 void glue(helper_pshufb, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
1457 {
1458     Reg *v = d;
1459     int i;
1460 #if SHIFT == 0
1461     uint8_t r[8];
1462 
1463     for (i = 0; i < 8; i++) {
1464         r[i] = (s->B(i) & 0x80) ? 0 : (v->B(s->B(i) & 7));
1465     }
1466     for (i = 0; i < 8; i++) {
1467         d->B(i) = r[i];
1468     }
1469 #else
1470     uint8_t r[8 << SHIFT];
1471 
1472     for (i = 0; i < 8 << SHIFT; i++) {
1473         int j = i & ~0xf;
1474         r[i] = (s->B(i) & 0x80) ? 0 : v->B(j | (s->B(i) & 0xf));
1475     }
1476     for (i = 0; i < 8 << SHIFT; i++) {
1477         d->B(i) = r[i];
1478     }
1479 #endif
1480 }
1481 
1482 #define SSE_HELPER_HW(name, F)  \
1483 void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s) \
1484 {                                                          \
1485     Reg *v = d;                                            \
1486     uint16_t r[4 << SHIFT];                                \
1487     int i, j, k;                                           \
1488     for (k = 0; k < 4 << SHIFT; k += LANE_WIDTH / 2) {     \
1489         for (i = j = 0; j < LANE_WIDTH / 2; i++, j += 2) { \
1490             r[i + k] = F(v->W(j + k), v->W(j + k + 1));    \
1491         }                                                  \
1492         for (j = 0; j < LANE_WIDTH / 2; i++, j += 2) {     \
1493             r[i + k] = F(s->W(j + k), s->W(j + k + 1));    \
1494         }                                                  \
1495     }                                                      \
1496     for (i = 0; i < 4 << SHIFT; i++) {                     \
1497         d->W(i) = r[i];                                    \
1498     }                                                      \
1499 }
1500 
1501 #define SSE_HELPER_HL(name, F)  \
1502 void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s) \
1503 {                                                          \
1504     Reg *v = d;                                            \
1505     uint32_t r[2 << SHIFT];                                \
1506     int i, j, k;                                           \
1507     for (k = 0; k < 2 << SHIFT; k += LANE_WIDTH / 4) {     \
1508         for (i = j = 0; j < LANE_WIDTH / 4; i++, j += 2) { \
1509             r[i + k] = F(v->L(j + k), v->L(j + k + 1));    \
1510         }                                                  \
1511         for (j = 0; j < LANE_WIDTH / 4; i++, j += 2) {     \
1512             r[i + k] = F(s->L(j + k), s->L(j + k + 1));    \
1513         }                                                  \
1514     }                                                      \
1515     for (i = 0; i < 2 << SHIFT; i++) {                     \
1516         d->L(i) = r[i];                                    \
1517     }                                                      \
1518 }
1519 
1520 SSE_HELPER_HW(phaddw, FADD)
1521 SSE_HELPER_HW(phsubw, FSUB)
1522 SSE_HELPER_HW(phaddsw, FADDSW)
1523 SSE_HELPER_HW(phsubsw, FSUBSW)
1524 SSE_HELPER_HL(phaddd, FADD)
1525 SSE_HELPER_HL(phsubd, FSUB)
1526 
1527 #undef SSE_HELPER_HW
1528 #undef SSE_HELPER_HL
1529 
1530 void glue(helper_pmaddubsw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
1531 {
1532     Reg *v = d;
1533     int i;
1534     for (i = 0; i < 4 << SHIFT; i++) {
1535         d->W(i) = satsw((int8_t)s->B(i * 2) * (uint8_t)v->B(i * 2) +
1536                         (int8_t)s->B(i * 2 + 1) * (uint8_t)v->B(i * 2 + 1));
1537     }
1538 }
1539 
1540 #define FABSB(x) (x > INT8_MAX  ? -(int8_t)x : x)
1541 #define FABSW(x) (x > INT16_MAX ? -(int16_t)x : x)
1542 #define FABSL(x) (x > INT32_MAX ? -(int32_t)x : x)
1543 SSE_HELPER_1(helper_pabsb, B, 8 << SHIFT, FABSB)
1544 SSE_HELPER_1(helper_pabsw, W, 4 << SHIFT, FABSW)
1545 SSE_HELPER_1(helper_pabsd, L, 2 << SHIFT, FABSL)
1546 
1547 #define FMULHRSW(d, s) (((int16_t) d * (int16_t)s + 0x4000) >> 15)
1548 SSE_HELPER_W(helper_pmulhrsw, FMULHRSW)
1549 
1550 #define FSIGNB(d, s) (s <= INT8_MAX  ? s ? d : 0 : -(int8_t)d)
1551 #define FSIGNW(d, s) (s <= INT16_MAX ? s ? d : 0 : -(int16_t)d)
1552 #define FSIGNL(d, s) (s <= INT32_MAX ? s ? d : 0 : -(int32_t)d)
1553 SSE_HELPER_B(helper_psignb, FSIGNB)
1554 SSE_HELPER_W(helper_psignw, FSIGNW)
1555 SSE_HELPER_L(helper_psignd, FSIGNL)
1556 
1557 void glue(helper_palignr, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
1558                                   int32_t shift)
1559 {
1560     Reg *v = d;
1561     int i;
1562 
1563     /* XXX could be checked during translation */
1564     if (shift >= (SHIFT ? 32 : 16)) {
1565         for (i = 0; i < (1 << SHIFT); i++) {
1566             d->Q(i) = 0;
1567         }
1568     } else {
1569         shift <<= 3;
1570 #define SHR(v, i) (i < 64 && i > -64 ? i > 0 ? v >> (i) : (v << -(i)) : 0)
1571 #if SHIFT == 0
1572         d->Q(0) = SHR(s->Q(0), shift - 0) |
1573             SHR(v->Q(0), shift -  64);
1574 #else
1575         for (i = 0; i < (1 << SHIFT); i += 2) {
1576             uint64_t r0, r1;
1577 
1578             r0 = SHR(s->Q(i), shift - 0) |
1579                  SHR(s->Q(i + 1), shift -  64) |
1580                  SHR(v->Q(i), shift - 128) |
1581                  SHR(v->Q(i + 1), shift - 192);
1582             r1 = SHR(s->Q(i), shift + 64) |
1583                  SHR(s->Q(i + 1), shift -   0) |
1584                  SHR(v->Q(i), shift -  64) |
1585                  SHR(v->Q(i + 1), shift - 128);
1586             d->Q(i) = r0;
1587             d->Q(i + 1) = r1;
1588         }
1589 #endif
1590 #undef SHR
1591     }
1592 }
1593 
1594 #if SHIFT >= 1
1595 
1596 #define SSE_HELPER_V(name, elem, num, F)                                \
1597     void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)   \
1598     {                                                                   \
1599         Reg *v = d;                                                     \
1600         Reg *m = &env->xmm_regs[0];                                     \
1601         int i;                                                          \
1602         for (i = 0; i < num; i++) {                                     \
1603             d->elem(i) = F(v->elem(i), s->elem(i), m->elem(i));         \
1604         }                                                               \
1605     }
1606 
1607 #define SSE_HELPER_I(name, elem, num, F)                                \
1608     void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,   \
1609                             uint32_t imm)                               \
1610     {                                                                   \
1611         Reg *v = d;                                                     \
1612         int i;                                                          \
1613         for (i = 0; i < num; i++) {                                     \
1614             int j = i & 7;                                              \
1615             d->elem(i) = F(v->elem(i), s->elem(i), (imm >> j) & 1);     \
1616         }                                                               \
1617     }
1618 
1619 /* SSE4.1 op helpers */
1620 #define FBLENDVB(v, s, m) ((m & 0x80) ? s : v)
1621 #define FBLENDVPS(v, s, m) ((m & 0x80000000) ? s : v)
1622 #define FBLENDVPD(v, s, m) ((m & 0x8000000000000000LL) ? s : v)
1623 SSE_HELPER_V(helper_pblendvb, B, 8 << SHIFT, FBLENDVB)
1624 SSE_HELPER_V(helper_blendvps, L, 2 << SHIFT, FBLENDVPS)
1625 SSE_HELPER_V(helper_blendvpd, Q, 1 << SHIFT, FBLENDVPD)
1626 
1627 void glue(helper_ptest, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
1628 {
1629     uint64_t zf = 0, cf = 0;
1630     int i;
1631 
1632     for (i = 0; i < 1 << SHIFT; i++) {
1633         zf |= (s->Q(i) &  d->Q(i));
1634         cf |= (s->Q(i) & ~d->Q(i));
1635     }
1636     CC_SRC = (zf ? 0 : CC_Z) | (cf ? 0 : CC_C);
1637 }
1638 
1639 #define SSE_HELPER_F(name, elem, num, F)                        \
1640     void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)   \
1641     {                                                           \
1642         int n = num;                                            \
1643         for (int i = n; --i >= 0; ) {                           \
1644             d->elem(i) = F(i);                                  \
1645         }                                                       \
1646     }
1647 
1648 #if SHIFT > 0
1649 SSE_HELPER_F(helper_pmovsxbw, W, 4 << SHIFT, (int8_t) s->B)
1650 SSE_HELPER_F(helper_pmovsxbd, L, 2 << SHIFT, (int8_t) s->B)
1651 SSE_HELPER_F(helper_pmovsxbq, Q, 1 << SHIFT, (int8_t) s->B)
1652 SSE_HELPER_F(helper_pmovsxwd, L, 2 << SHIFT, (int16_t) s->W)
1653 SSE_HELPER_F(helper_pmovsxwq, Q, 1 << SHIFT, (int16_t) s->W)
1654 SSE_HELPER_F(helper_pmovsxdq, Q, 1 << SHIFT, (int32_t) s->L)
1655 SSE_HELPER_F(helper_pmovzxbw, W, 4 << SHIFT, s->B)
1656 SSE_HELPER_F(helper_pmovzxbd, L, 2 << SHIFT, s->B)
1657 SSE_HELPER_F(helper_pmovzxbq, Q, 1 << SHIFT, s->B)
1658 SSE_HELPER_F(helper_pmovzxwd, L, 2 << SHIFT, s->W)
1659 SSE_HELPER_F(helper_pmovzxwq, Q, 1 << SHIFT, s->W)
1660 SSE_HELPER_F(helper_pmovzxdq, Q, 1 << SHIFT, s->L)
1661 #endif
1662 
1663 void glue(helper_pmuldq, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
1664 {
1665     Reg *v = d;
1666     int i;
1667 
1668     for (i = 0; i < 1 << SHIFT; i++) {
1669         d->Q(i) = (int64_t)(int32_t) v->L(2 * i) * (int32_t) s->L(2 * i);
1670     }
1671 }
1672 
1673 #define FCMPEQQ(d, s) (d == s ? -1 : 0)
1674 SSE_HELPER_Q(helper_pcmpeqq, FCMPEQQ)
1675 
1676 void glue(helper_packusdw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
1677 {
1678     Reg *v = d;
1679     uint16_t r[8];
1680     int i, j, k;
1681 
1682     for (i = 0, j = 0; i <= 2 << SHIFT; i += 8, j += 4) {
1683         r[0] = satuw(v->L(j));
1684         r[1] = satuw(v->L(j + 1));
1685         r[2] = satuw(v->L(j + 2));
1686         r[3] = satuw(v->L(j + 3));
1687         r[4] = satuw(s->L(j));
1688         r[5] = satuw(s->L(j + 1));
1689         r[6] = satuw(s->L(j + 2));
1690         r[7] = satuw(s->L(j + 3));
1691         for (k = 0; k < 8; k++) {
1692             d->W(i + k) = r[k];
1693         }
1694     }
1695 }
1696 
1697 #define FMINSB(d, s) MIN((int8_t)d, (int8_t)s)
1698 #define FMINSD(d, s) MIN((int32_t)d, (int32_t)s)
1699 #define FMAXSB(d, s) MAX((int8_t)d, (int8_t)s)
1700 #define FMAXSD(d, s) MAX((int32_t)d, (int32_t)s)
1701 SSE_HELPER_B(helper_pminsb, FMINSB)
1702 SSE_HELPER_L(helper_pminsd, FMINSD)
1703 SSE_HELPER_W(helper_pminuw, MIN)
1704 SSE_HELPER_L(helper_pminud, MIN)
1705 SSE_HELPER_B(helper_pmaxsb, FMAXSB)
1706 SSE_HELPER_L(helper_pmaxsd, FMAXSD)
1707 SSE_HELPER_W(helper_pmaxuw, MAX)
1708 SSE_HELPER_L(helper_pmaxud, MAX)
1709 
1710 #define FMULLD(d, s) ((int32_t)d * (int32_t)s)
1711 SSE_HELPER_L(helper_pmulld, FMULLD)
1712 
1713 #if SHIFT == 1
1714 void glue(helper_phminposuw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
1715 {
1716     int idx = 0;
1717 
1718     if (s->W(1) < s->W(idx)) {
1719         idx = 1;
1720     }
1721     if (s->W(2) < s->W(idx)) {
1722         idx = 2;
1723     }
1724     if (s->W(3) < s->W(idx)) {
1725         idx = 3;
1726     }
1727     if (s->W(4) < s->W(idx)) {
1728         idx = 4;
1729     }
1730     if (s->W(5) < s->W(idx)) {
1731         idx = 5;
1732     }
1733     if (s->W(6) < s->W(idx)) {
1734         idx = 6;
1735     }
1736     if (s->W(7) < s->W(idx)) {
1737         idx = 7;
1738     }
1739 
1740     d->W(0) = s->W(idx);
1741     d->W(1) = idx;
1742     d->L(1) = 0;
1743     d->Q(1) = 0;
1744 }
1745 #endif
1746 
1747 void glue(helper_roundps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
1748                                   uint32_t mode)
1749 {
1750     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
1751     signed char prev_rounding_mode;
1752     int i;
1753 
1754     prev_rounding_mode = env->sse_status.float_rounding_mode;
1755     if (!(mode & (1 << 2))) {
1756         switch (mode & 3) {
1757         case 0:
1758             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1759             break;
1760         case 1:
1761             set_float_rounding_mode(float_round_down, &env->sse_status);
1762             break;
1763         case 2:
1764             set_float_rounding_mode(float_round_up, &env->sse_status);
1765             break;
1766         case 3:
1767             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1768             break;
1769         }
1770     }
1771 
1772     for (i = 0; i < 2 << SHIFT; i++) {
1773         d->ZMM_S(i) = float32_round_to_int(s->ZMM_S(i), &env->sse_status);
1774     }
1775 
1776     if (mode & (1 << 3) && !(old_flags & float_flag_inexact)) {
1777         set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1778                                   ~float_flag_inexact,
1779                                   &env->sse_status);
1780     }
1781     env->sse_status.float_rounding_mode = prev_rounding_mode;
1782 }
1783 
1784 void glue(helper_roundpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
1785                                   uint32_t mode)
1786 {
1787     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
1788     signed char prev_rounding_mode;
1789     int i;
1790 
1791     prev_rounding_mode = env->sse_status.float_rounding_mode;
1792     if (!(mode & (1 << 2))) {
1793         switch (mode & 3) {
1794         case 0:
1795             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1796             break;
1797         case 1:
1798             set_float_rounding_mode(float_round_down, &env->sse_status);
1799             break;
1800         case 2:
1801             set_float_rounding_mode(float_round_up, &env->sse_status);
1802             break;
1803         case 3:
1804             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1805             break;
1806         }
1807     }
1808 
1809     for (i = 0; i < 1 << SHIFT; i++) {
1810         d->ZMM_D(i) = float64_round_to_int(s->ZMM_D(i), &env->sse_status);
1811     }
1812 
1813     if (mode & (1 << 3) && !(old_flags & float_flag_inexact)) {
1814         set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1815                                   ~float_flag_inexact,
1816                                   &env->sse_status);
1817     }
1818     env->sse_status.float_rounding_mode = prev_rounding_mode;
1819 }
1820 
1821 #if SHIFT == 1
1822 void glue(helper_roundss, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
1823                                   uint32_t mode)
1824 {
1825     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
1826     signed char prev_rounding_mode;
1827 
1828     prev_rounding_mode = env->sse_status.float_rounding_mode;
1829     if (!(mode & (1 << 2))) {
1830         switch (mode & 3) {
1831         case 0:
1832             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1833             break;
1834         case 1:
1835             set_float_rounding_mode(float_round_down, &env->sse_status);
1836             break;
1837         case 2:
1838             set_float_rounding_mode(float_round_up, &env->sse_status);
1839             break;
1840         case 3:
1841             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1842             break;
1843         }
1844     }
1845 
1846     d->ZMM_S(0) = float32_round_to_int(s->ZMM_S(0), &env->sse_status);
1847 
1848     if (mode & (1 << 3) && !(old_flags & float_flag_inexact)) {
1849         set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1850                                   ~float_flag_inexact,
1851                                   &env->sse_status);
1852     }
1853     env->sse_status.float_rounding_mode = prev_rounding_mode;
1854 }
1855 
1856 void glue(helper_roundsd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
1857                                   uint32_t mode)
1858 {
1859     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
1860     signed char prev_rounding_mode;
1861 
1862     prev_rounding_mode = env->sse_status.float_rounding_mode;
1863     if (!(mode & (1 << 2))) {
1864         switch (mode & 3) {
1865         case 0:
1866             set_float_rounding_mode(float_round_nearest_even, &env->sse_status);
1867             break;
1868         case 1:
1869             set_float_rounding_mode(float_round_down, &env->sse_status);
1870             break;
1871         case 2:
1872             set_float_rounding_mode(float_round_up, &env->sse_status);
1873             break;
1874         case 3:
1875             set_float_rounding_mode(float_round_to_zero, &env->sse_status);
1876             break;
1877         }
1878     }
1879 
1880     d->ZMM_D(0) = float64_round_to_int(s->ZMM_D(0), &env->sse_status);
1881 
1882     if (mode & (1 << 3) && !(old_flags & float_flag_inexact)) {
1883         set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1884                                   ~float_flag_inexact,
1885                                   &env->sse_status);
1886     }
1887     env->sse_status.float_rounding_mode = prev_rounding_mode;
1888 }
1889 #endif
1890 
1891 #define FBLENDP(v, s, m) (m ? s : v)
1892 SSE_HELPER_I(helper_blendps, L, 2 << SHIFT, FBLENDP)
1893 SSE_HELPER_I(helper_blendpd, Q, 1 << SHIFT, FBLENDP)
1894 SSE_HELPER_I(helper_pblendw, W, 4 << SHIFT, FBLENDP)
1895 
1896 void glue(helper_dpps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
1897                                uint32_t mask)
1898 {
1899     Reg *v = d;
1900     float32 prod1, prod2, temp2, temp3, temp4;
1901     int i;
1902 
1903     for (i = 0; i < 2 << SHIFT; i += 4) {
1904         /*
1905          * We must evaluate (A+B)+(C+D), not ((A+B)+C)+D
1906          * to correctly round the intermediate results
1907          */
1908         if (mask & (1 << 4)) {
1909             prod1 = float32_mul(v->ZMM_S(i), s->ZMM_S(i), &env->sse_status);
1910         } else {
1911             prod1 = float32_zero;
1912         }
1913         if (mask & (1 << 5)) {
1914             prod2 = float32_mul(v->ZMM_S(i+1), s->ZMM_S(i+1), &env->sse_status);
1915         } else {
1916             prod2 = float32_zero;
1917         }
1918         temp2 = float32_add(prod1, prod2, &env->sse_status);
1919         if (mask & (1 << 6)) {
1920             prod1 = float32_mul(v->ZMM_S(i+2), s->ZMM_S(i+2), &env->sse_status);
1921         } else {
1922             prod1 = float32_zero;
1923         }
1924         if (mask & (1 << 7)) {
1925             prod2 = float32_mul(v->ZMM_S(i+3), s->ZMM_S(i+3), &env->sse_status);
1926         } else {
1927             prod2 = float32_zero;
1928         }
1929         temp3 = float32_add(prod1, prod2, &env->sse_status);
1930         temp4 = float32_add(temp2, temp3, &env->sse_status);
1931 
1932         d->ZMM_S(i) = (mask & (1 << 0)) ? temp4 : float32_zero;
1933         d->ZMM_S(i+1) = (mask & (1 << 1)) ? temp4 : float32_zero;
1934         d->ZMM_S(i+2) = (mask & (1 << 2)) ? temp4 : float32_zero;
1935         d->ZMM_S(i+3) = (mask & (1 << 3)) ? temp4 : float32_zero;
1936     }
1937 }
1938 
1939 #if SHIFT == 1
1940 /* Oddly, there is no ymm version of dppd */
1941 void glue(helper_dppd, SUFFIX)(CPUX86State *env,
1942                                Reg *d, Reg *s, uint32_t mask)
1943 {
1944     Reg *v = d;
1945     float64 prod1, prod2, temp2;
1946 
1947     if (mask & (1 << 4)) {
1948         prod1 = float64_mul(v->ZMM_D(0), s->ZMM_D(0), &env->sse_status);
1949     } else {
1950         prod1 = float64_zero;
1951     }
1952     if (mask & (1 << 5)) {
1953         prod2 = float64_mul(v->ZMM_D(1), s->ZMM_D(1), &env->sse_status);
1954     } else {
1955         prod2 = float64_zero;
1956     }
1957     temp2 = float64_add(prod1, prod2, &env->sse_status);
1958     d->ZMM_D(0) = (mask & (1 << 0)) ? temp2 : float64_zero;
1959     d->ZMM_D(1) = (mask & (1 << 1)) ? temp2 : float64_zero;
1960 }
1961 #endif
1962 
1963 void glue(helper_mpsadbw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
1964                                   uint32_t offset)
1965 {
1966     Reg *v = d;
1967     int i, j;
1968     uint16_t r[8];
1969 
1970     for (j = 0; j < 4 << SHIFT; ) {
1971         int s0 = (j * 2) + ((offset & 3) << 2);
1972         int d0 = (j * 2) + ((offset & 4) << 0);
1973         for (i = 0; i < LANE_WIDTH / 2; i++, d0++) {
1974             r[i] = 0;
1975             r[i] += abs1(v->B(d0 + 0) - s->B(s0 + 0));
1976             r[i] += abs1(v->B(d0 + 1) - s->B(s0 + 1));
1977             r[i] += abs1(v->B(d0 + 2) - s->B(s0 + 2));
1978             r[i] += abs1(v->B(d0 + 3) - s->B(s0 + 3));
1979         }
1980         for (i = 0; i < LANE_WIDTH / 2; i++, j++) {
1981             d->W(j) = r[i];
1982         }
1983         offset >>= 3;
1984     }
1985 }
1986 
1987 /* SSE4.2 op helpers */
1988 #define FCMPGTQ(d, s) ((int64_t)d > (int64_t)s ? -1 : 0)
1989 SSE_HELPER_Q(helper_pcmpgtq, FCMPGTQ)
1990 
1991 #if SHIFT == 1
1992 static inline int pcmp_elen(CPUX86State *env, int reg, uint32_t ctrl)
1993 {
1994     target_long val, limit;
1995 
1996     /* Presence of REX.W is indicated by a bit higher than 7 set */
1997     if (ctrl >> 8) {
1998         val = (target_long)env->regs[reg];
1999     } else {
2000         val = (int32_t)env->regs[reg];
2001     }
2002     if (ctrl & 1) {
2003         limit = 8;
2004     } else {
2005         limit = 16;
2006     }
2007     if ((val > limit) || (val < -limit)) {
2008         return limit;
2009     }
2010     return abs1(val);
2011 }
2012 
2013 static inline int pcmp_ilen(Reg *r, uint8_t ctrl)
2014 {
2015     int val = 0;
2016 
2017     if (ctrl & 1) {
2018         while (val < 8 && r->W(val)) {
2019             val++;
2020         }
2021     } else {
2022         while (val < 16 && r->B(val)) {
2023             val++;
2024         }
2025     }
2026 
2027     return val;
2028 }
2029 
2030 static inline int pcmp_val(Reg *r, uint8_t ctrl, int i)
2031 {
2032     switch ((ctrl >> 0) & 3) {
2033     case 0:
2034         return r->B(i);
2035     case 1:
2036         return r->W(i);
2037     case 2:
2038         return (int8_t)r->B(i);
2039     case 3:
2040     default:
2041         return (int16_t)r->W(i);
2042     }
2043 }
2044 
2045 static inline unsigned pcmpxstrx(CPUX86State *env, Reg *d, Reg *s,
2046                                  int8_t ctrl, int valids, int validd)
2047 {
2048     unsigned int res = 0;
2049     int v;
2050     int j, i;
2051     int upper = (ctrl & 1) ? 7 : 15;
2052 
2053     valids--;
2054     validd--;
2055 
2056     CC_SRC = (valids < upper ? CC_Z : 0) | (validd < upper ? CC_S : 0);
2057 
2058     switch ((ctrl >> 2) & 3) {
2059     case 0:
2060         for (j = valids; j >= 0; j--) {
2061             res <<= 1;
2062             v = pcmp_val(s, ctrl, j);
2063             for (i = validd; i >= 0; i--) {
2064                 res |= (v == pcmp_val(d, ctrl, i));
2065             }
2066         }
2067         break;
2068     case 1:
2069         for (j = valids; j >= 0; j--) {
2070             res <<= 1;
2071             v = pcmp_val(s, ctrl, j);
2072             for (i = ((validd - 1) | 1); i >= 0; i -= 2) {
2073                 res |= (pcmp_val(d, ctrl, i - 0) >= v &&
2074                         pcmp_val(d, ctrl, i - 1) <= v);
2075             }
2076         }
2077         break;
2078     case 2:
2079         res = (1 << (upper - MAX(valids, validd))) - 1;
2080         res <<= MAX(valids, validd) - MIN(valids, validd);
2081         for (i = MIN(valids, validd); i >= 0; i--) {
2082             res <<= 1;
2083             v = pcmp_val(s, ctrl, i);
2084             res |= (v == pcmp_val(d, ctrl, i));
2085         }
2086         break;
2087     case 3:
2088         if (validd == -1) {
2089             res = (2 << upper) - 1;
2090             break;
2091         }
2092         for (j = valids == upper ? valids : valids - validd; j >= 0; j--) {
2093             res <<= 1;
2094             v = 1;
2095             for (i = MIN(valids - j, validd); i >= 0; i--) {
2096                 v &= (pcmp_val(s, ctrl, i + j) == pcmp_val(d, ctrl, i));
2097             }
2098             res |= v;
2099         }
2100         break;
2101     }
2102 
2103     switch ((ctrl >> 4) & 3) {
2104     case 1:
2105         res ^= (2 << upper) - 1;
2106         break;
2107     case 3:
2108         res ^= (1 << (valids + 1)) - 1;
2109         break;
2110     }
2111 
2112     if (res) {
2113         CC_SRC |= CC_C;
2114     }
2115     if (res & 1) {
2116         CC_SRC |= CC_O;
2117     }
2118 
2119     return res;
2120 }
2121 
2122 void glue(helper_pcmpestri, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2123                                     uint32_t ctrl)
2124 {
2125     unsigned int res = pcmpxstrx(env, d, s, ctrl,
2126                                  pcmp_elen(env, R_EDX, ctrl),
2127                                  pcmp_elen(env, R_EAX, ctrl));
2128 
2129     if (res) {
2130         env->regs[R_ECX] = (ctrl & (1 << 6)) ? 31 - clz32(res) : ctz32(res);
2131     } else {
2132         env->regs[R_ECX] = 16 >> (ctrl & (1 << 0));
2133     }
2134 }
2135 
2136 void glue(helper_pcmpestrm, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2137                                     uint32_t ctrl)
2138 {
2139     int i;
2140     unsigned int res = pcmpxstrx(env, d, s, ctrl,
2141                                  pcmp_elen(env, R_EDX, ctrl),
2142                                  pcmp_elen(env, R_EAX, ctrl));
2143 
2144     if ((ctrl >> 6) & 1) {
2145         if (ctrl & 1) {
2146             for (i = 0; i < 8; i++, res >>= 1) {
2147                 env->xmm_regs[0].W(i) = (res & 1) ? ~0 : 0;
2148             }
2149         } else {
2150             for (i = 0; i < 16; i++, res >>= 1) {
2151                 env->xmm_regs[0].B(i) = (res & 1) ? ~0 : 0;
2152             }
2153         }
2154     } else {
2155         env->xmm_regs[0].Q(1) = 0;
2156         env->xmm_regs[0].Q(0) = res;
2157     }
2158 }
2159 
2160 void glue(helper_pcmpistri, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2161                                     uint32_t ctrl)
2162 {
2163     unsigned int res = pcmpxstrx(env, d, s, ctrl,
2164                                  pcmp_ilen(s, ctrl),
2165                                  pcmp_ilen(d, ctrl));
2166 
2167     if (res) {
2168         env->regs[R_ECX] = (ctrl & (1 << 6)) ? 31 - clz32(res) : ctz32(res);
2169     } else {
2170         env->regs[R_ECX] = 16 >> (ctrl & (1 << 0));
2171     }
2172 }
2173 
2174 void glue(helper_pcmpistrm, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2175                                     uint32_t ctrl)
2176 {
2177     int i;
2178     unsigned int res = pcmpxstrx(env, d, s, ctrl,
2179                                  pcmp_ilen(s, ctrl),
2180                                  pcmp_ilen(d, ctrl));
2181 
2182     if ((ctrl >> 6) & 1) {
2183         if (ctrl & 1) {
2184             for (i = 0; i < 8; i++, res >>= 1) {
2185                 env->xmm_regs[0].W(i) = (res & 1) ? ~0 : 0;
2186             }
2187         } else {
2188             for (i = 0; i < 16; i++, res >>= 1) {
2189                 env->xmm_regs[0].B(i) = (res & 1) ? ~0 : 0;
2190             }
2191         }
2192     } else {
2193         env->xmm_regs[0].Q(1) = 0;
2194         env->xmm_regs[0].Q(0) = res;
2195     }
2196 }
2197 
2198 #define CRCPOLY        0x1edc6f41
2199 #define CRCPOLY_BITREV 0x82f63b78
2200 target_ulong helper_crc32(uint32_t crc1, target_ulong msg, uint32_t len)
2201 {
2202     target_ulong crc = (msg & ((target_ulong) -1 >>
2203                                (TARGET_LONG_BITS - len))) ^ crc1;
2204 
2205     while (len--) {
2206         crc = (crc >> 1) ^ ((crc & 1) ? CRCPOLY_BITREV : 0);
2207     }
2208 
2209     return crc;
2210 }
2211 
2212 #endif
2213 
2214 #if SHIFT == 1
2215 static void clmulq(uint64_t *dest_l, uint64_t *dest_h,
2216                           uint64_t a, uint64_t b)
2217 {
2218     uint64_t al, ah, resh, resl;
2219 
2220     ah = 0;
2221     al = a;
2222     resh = resl = 0;
2223 
2224     while (b) {
2225         if (b & 1) {
2226             resl ^= al;
2227             resh ^= ah;
2228         }
2229         ah = (ah << 1) | (al >> 63);
2230         al <<= 1;
2231         b >>= 1;
2232     }
2233 
2234     *dest_l = resl;
2235     *dest_h = resh;
2236 }
2237 #endif
2238 
2239 void glue(helper_pclmulqdq, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2240                                     uint32_t ctrl)
2241 {
2242     Reg *v = d;
2243     uint64_t a, b;
2244     int i;
2245 
2246     for (i = 0; i < 1 << SHIFT; i += 2) {
2247         a = v->Q(((ctrl & 1) != 0) + i);
2248         b = s->Q(((ctrl & 16) != 0) + i);
2249         clmulq(&d->Q(i), &d->Q(i + 1), a, b);
2250     }
2251 }
2252 
2253 void glue(helper_aesdec, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
2254 {
2255     int i;
2256     Reg st = *d;
2257     Reg rk = *s;
2258 
2259     for (i = 0 ; i < 2 << SHIFT ; i++) {
2260         int j = i & 3;
2261         d->L(i) = rk.L(i) ^ bswap32(AES_Td0[st.B(AES_ishifts[4 * j + 0])] ^
2262                                     AES_Td1[st.B(AES_ishifts[4 * j + 1])] ^
2263                                     AES_Td2[st.B(AES_ishifts[4 * j + 2])] ^
2264                                     AES_Td3[st.B(AES_ishifts[4 * j + 3])]);
2265     }
2266 }
2267 
2268 void glue(helper_aesdeclast, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
2269 {
2270     int i;
2271     Reg st = *d;
2272     Reg rk = *s;
2273 
2274     for (i = 0; i < 8 << SHIFT; i++) {
2275         d->B(i) = rk.B(i) ^ (AES_isbox[st.B(AES_ishifts[i & 15] + (i & ~15))]);
2276     }
2277 }
2278 
2279 void glue(helper_aesenc, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
2280 {
2281     int i;
2282     Reg st = *d;
2283     Reg rk = *s;
2284 
2285     for (i = 0 ; i < 2 << SHIFT ; i++) {
2286         int j = i & 3;
2287         d->L(i) = rk.L(i) ^ bswap32(AES_Te0[st.B(AES_shifts[4 * j + 0])] ^
2288                                     AES_Te1[st.B(AES_shifts[4 * j + 1])] ^
2289                                     AES_Te2[st.B(AES_shifts[4 * j + 2])] ^
2290                                     AES_Te3[st.B(AES_shifts[4 * j + 3])]);
2291     }
2292 }
2293 
2294 void glue(helper_aesenclast, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
2295 {
2296     int i;
2297     Reg st = *d;
2298     Reg rk = *s;
2299 
2300     for (i = 0; i < 8 << SHIFT; i++) {
2301         d->B(i) = rk.B(i) ^ (AES_sbox[st.B(AES_shifts[i & 15] + (i & ~15))]);
2302     }
2303 }
2304 
2305 #if SHIFT == 1
2306 void glue(helper_aesimc, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
2307 {
2308     int i;
2309     Reg tmp = *s;
2310 
2311     for (i = 0 ; i < 4 ; i++) {
2312         d->L(i) = bswap32(AES_imc[tmp.B(4 * i + 0)][0] ^
2313                           AES_imc[tmp.B(4 * i + 1)][1] ^
2314                           AES_imc[tmp.B(4 * i + 2)][2] ^
2315                           AES_imc[tmp.B(4 * i + 3)][3]);
2316     }
2317 }
2318 
2319 void glue(helper_aeskeygenassist, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2320                                           uint32_t ctrl)
2321 {
2322     int i;
2323     Reg tmp = *s;
2324 
2325     for (i = 0 ; i < 4 ; i++) {
2326         d->B(i) = AES_sbox[tmp.B(i + 4)];
2327         d->B(i + 8) = AES_sbox[tmp.B(i + 12)];
2328     }
2329     d->L(1) = (d->L(0) << 24 | d->L(0) >> 8) ^ ctrl;
2330     d->L(3) = (d->L(2) << 24 | d->L(2) >> 8) ^ ctrl;
2331 }
2332 #endif
2333 #endif
2334 
2335 #undef SSE_HELPER_S
2336 
2337 #undef SHIFT
2338 #undef XMM_ONLY
2339 #undef Reg
2340 #undef B
2341 #undef W
2342 #undef L
2343 #undef Q
2344 #undef SUFFIX
2345