xref: /openbmc/qemu/target/i386/ops_sse.h (revision 2b74dd91)
1 /*
2  *  MMX/3DNow!/SSE/SSE2/SSE3/SSSE3/SSE4/PNI support
3  *
4  *  Copyright (c) 2005 Fabrice Bellard
5  *  Copyright (c) 2008 Intel Corporation  <andrew.zaborowski@intel.com>
6  *
7  * This library is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * This library is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
19  */
20 
21 #include "crypto/aes.h"
22 #include "crypto/aes-round.h"
23 #include "crypto/clmul.h"
24 
25 #if SHIFT == 0
26 #define Reg MMXReg
27 #define XMM_ONLY(...)
28 #define B(n) MMX_B(n)
29 #define W(n) MMX_W(n)
30 #define L(n) MMX_L(n)
31 #define Q(n) MMX_Q(n)
32 #define SUFFIX _mmx
33 #else
34 #define Reg ZMMReg
35 #define XMM_ONLY(...) __VA_ARGS__
36 #define B(n) ZMM_B(n)
37 #define W(n) ZMM_W(n)
38 #define L(n) ZMM_L(n)
39 #define Q(n) ZMM_Q(n)
40 #if SHIFT == 1
41 #define SUFFIX _xmm
42 #else
43 #define SUFFIX _ymm
44 #endif
45 #endif
46 
47 #define LANE_WIDTH (SHIFT ? 16 : 8)
48 #define PACK_WIDTH (LANE_WIDTH / 2)
49 
50 #if SHIFT == 0
51 #define FPSRL(x, c) ((x) >> shift)
52 #define FPSRAW(x, c) ((int16_t)(x) >> shift)
53 #define FPSRAL(x, c) ((int32_t)(x) >> shift)
54 #define FPSLL(x, c) ((x) << shift)
55 #endif
56 
57 void glue(helper_psrlw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
58 {
59     int shift;
60     if (c->Q(0) > 15) {
61         for (int i = 0; i < 1 << SHIFT; i++) {
62             d->Q(i) = 0;
63         }
64     } else {
65         shift = c->B(0);
66         for (int i = 0; i < 4 << SHIFT; i++) {
67             d->W(i) = FPSRL(s->W(i), shift);
68         }
69     }
70 }
71 
72 void glue(helper_psllw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
73 {
74     int shift;
75     if (c->Q(0) > 15) {
76         for (int i = 0; i < 1 << SHIFT; i++) {
77             d->Q(i) = 0;
78         }
79     } else {
80         shift = c->B(0);
81         for (int i = 0; i < 4 << SHIFT; i++) {
82             d->W(i) = FPSLL(s->W(i), shift);
83         }
84     }
85 }
86 
87 void glue(helper_psraw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
88 {
89     int shift;
90     if (c->Q(0) > 15) {
91         shift = 15;
92     } else {
93         shift = c->B(0);
94     }
95     for (int i = 0; i < 4 << SHIFT; i++) {
96         d->W(i) = FPSRAW(s->W(i), shift);
97     }
98 }
99 
100 void glue(helper_psrld, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
101 {
102     int shift;
103     if (c->Q(0) > 31) {
104         for (int i = 0; i < 1 << SHIFT; i++) {
105             d->Q(i) = 0;
106         }
107     } else {
108         shift = c->B(0);
109         for (int i = 0; i < 2 << SHIFT; i++) {
110             d->L(i) = FPSRL(s->L(i), shift);
111         }
112     }
113 }
114 
115 void glue(helper_pslld, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
116 {
117     int shift;
118     if (c->Q(0) > 31) {
119         for (int i = 0; i < 1 << SHIFT; i++) {
120             d->Q(i) = 0;
121         }
122     } else {
123         shift = c->B(0);
124         for (int i = 0; i < 2 << SHIFT; i++) {
125             d->L(i) = FPSLL(s->L(i), shift);
126         }
127     }
128 }
129 
130 void glue(helper_psrad, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
131 {
132     int shift;
133     if (c->Q(0) > 31) {
134         shift = 31;
135     } else {
136         shift = c->B(0);
137     }
138     for (int i = 0; i < 2 << SHIFT; i++) {
139         d->L(i) = FPSRAL(s->L(i), shift);
140     }
141 }
142 
143 void glue(helper_psrlq, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
144 {
145     int shift;
146     if (c->Q(0) > 63) {
147         for (int i = 0; i < 1 << SHIFT; i++) {
148             d->Q(i) = 0;
149         }
150     } else {
151         shift = c->B(0);
152         for (int i = 0; i < 1 << SHIFT; i++) {
153             d->Q(i) = FPSRL(s->Q(i), shift);
154         }
155     }
156 }
157 
158 void glue(helper_psllq, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
159 {
160     int shift;
161     if (c->Q(0) > 63) {
162         for (int i = 0; i < 1 << SHIFT; i++) {
163             d->Q(i) = 0;
164         }
165     } else {
166         shift = c->B(0);
167         for (int i = 0; i < 1 << SHIFT; i++) {
168             d->Q(i) = FPSLL(s->Q(i), shift);
169         }
170     }
171 }
172 
173 #if SHIFT >= 1
174 void glue(helper_psrldq, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
175 {
176     int shift, i, j;
177 
178     shift = c->L(0);
179     if (shift > 16) {
180         shift = 16;
181     }
182     for (j = 0; j < 8 << SHIFT; j += LANE_WIDTH) {
183         for (i = 0; i < 16 - shift; i++) {
184             d->B(j + i) = s->B(j + i + shift);
185         }
186         for (i = 16 - shift; i < 16; i++) {
187             d->B(j + i) = 0;
188         }
189     }
190 }
191 
192 void glue(helper_pslldq, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, Reg *c)
193 {
194     int shift, i, j;
195 
196     shift = c->L(0);
197     if (shift > 16) {
198         shift = 16;
199     }
200     for (j = 0; j < 8 << SHIFT; j += LANE_WIDTH) {
201         for (i = 15; i >= shift; i--) {
202             d->B(j + i) = s->B(j + i - shift);
203         }
204         for (i = 0; i < shift; i++) {
205             d->B(j + i) = 0;
206         }
207     }
208 }
209 #endif
210 
211 #define SSE_HELPER_1(name, elem, num, F)                        \
212     void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)   \
213     {                                                           \
214         int n = num;                                            \
215         for (int i = 0; i < n; i++) {                           \
216             d->elem(i) = F(s->elem(i));                         \
217         }                                                       \
218     }
219 
220 #define SSE_HELPER_2(name, elem, num, F)                        \
221     void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)   \
222     {                                                           \
223         int n = num;                                            \
224         for (int i = 0; i < n; i++) {                           \
225             d->elem(i) = F(v->elem(i), s->elem(i));             \
226         }                                                       \
227     }
228 
229 #define SSE_HELPER_B(name, F)                                   \
230     SSE_HELPER_2(name, B, 8 << SHIFT, F)
231 
232 #define SSE_HELPER_W(name, F)                                   \
233     SSE_HELPER_2(name, W, 4 << SHIFT, F)
234 
235 #define SSE_HELPER_L(name, F)                                   \
236     SSE_HELPER_2(name, L, 2 << SHIFT, F)
237 
238 #define SSE_HELPER_Q(name, F)                                   \
239     SSE_HELPER_2(name, Q, 1 << SHIFT, F)
240 
241 #if SHIFT == 0
242 static inline int satub(int x)
243 {
244     if (x < 0) {
245         return 0;
246     } else if (x > 255) {
247         return 255;
248     } else {
249         return x;
250     }
251 }
252 
253 static inline int satuw(int x)
254 {
255     if (x < 0) {
256         return 0;
257     } else if (x > 65535) {
258         return 65535;
259     } else {
260         return x;
261     }
262 }
263 
264 static inline int satsb(int x)
265 {
266     if (x < -128) {
267         return -128;
268     } else if (x > 127) {
269         return 127;
270     } else {
271         return x;
272     }
273 }
274 
275 static inline int satsw(int x)
276 {
277     if (x < -32768) {
278         return -32768;
279     } else if (x > 32767) {
280         return 32767;
281     } else {
282         return x;
283     }
284 }
285 
286 #define FADD(a, b) ((a) + (b))
287 #define FADDUB(a, b) satub((a) + (b))
288 #define FADDUW(a, b) satuw((a) + (b))
289 #define FADDSB(a, b) satsb((int8_t)(a) + (int8_t)(b))
290 #define FADDSW(a, b) satsw((int16_t)(a) + (int16_t)(b))
291 
292 #define FSUB(a, b) ((a) - (b))
293 #define FSUBUB(a, b) satub((a) - (b))
294 #define FSUBUW(a, b) satuw((a) - (b))
295 #define FSUBSB(a, b) satsb((int8_t)(a) - (int8_t)(b))
296 #define FSUBSW(a, b) satsw((int16_t)(a) - (int16_t)(b))
297 #define FMINUB(a, b) ((a) < (b)) ? (a) : (b)
298 #define FMINSW(a, b) ((int16_t)(a) < (int16_t)(b)) ? (a) : (b)
299 #define FMAXUB(a, b) ((a) > (b)) ? (a) : (b)
300 #define FMAXSW(a, b) ((int16_t)(a) > (int16_t)(b)) ? (a) : (b)
301 
302 #define FMULHRW(a, b) (((int16_t)(a) * (int16_t)(b) + 0x8000) >> 16)
303 #define FMULHUW(a, b) ((a) * (b) >> 16)
304 #define FMULHW(a, b) ((int16_t)(a) * (int16_t)(b) >> 16)
305 
306 #define FAVG(a, b) (((a) + (b) + 1) >> 1)
307 #endif
308 
309 SSE_HELPER_W(helper_pmulhuw, FMULHUW)
310 SSE_HELPER_W(helper_pmulhw, FMULHW)
311 
312 #if SHIFT == 0
313 void glue(helper_pmulhrw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
314 {
315     d->W(0) = FMULHRW(d->W(0), s->W(0));
316     d->W(1) = FMULHRW(d->W(1), s->W(1));
317     d->W(2) = FMULHRW(d->W(2), s->W(2));
318     d->W(3) = FMULHRW(d->W(3), s->W(3));
319 }
320 #endif
321 
322 SSE_HELPER_B(helper_pavgb, FAVG)
323 SSE_HELPER_W(helper_pavgw, FAVG)
324 
325 void glue(helper_pmuludq, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
326 {
327     int i;
328 
329     for (i = 0; i < (1 << SHIFT); i++) {
330         d->Q(i) = (uint64_t)s->L(i * 2) * (uint64_t)v->L(i * 2);
331     }
332 }
333 
334 void glue(helper_pmaddwd, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
335 {
336     int i;
337 
338     for (i = 0; i < (2 << SHIFT); i++) {
339         d->L(i) = (int16_t)s->W(2 * i) * (int16_t)v->W(2 * i) +
340             (int16_t)s->W(2 * i + 1) * (int16_t)v->W(2 * i + 1);
341     }
342 }
343 
344 #if SHIFT == 0
345 static inline int abs1(int a)
346 {
347     if (a < 0) {
348         return -a;
349     } else {
350         return a;
351     }
352 }
353 #endif
354 void glue(helper_psadbw, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
355 {
356     int i;
357 
358     for (i = 0; i < (1 << SHIFT); i++) {
359         unsigned int val = 0;
360         val += abs1(v->B(8 * i + 0) - s->B(8 * i + 0));
361         val += abs1(v->B(8 * i + 1) - s->B(8 * i + 1));
362         val += abs1(v->B(8 * i + 2) - s->B(8 * i + 2));
363         val += abs1(v->B(8 * i + 3) - s->B(8 * i + 3));
364         val += abs1(v->B(8 * i + 4) - s->B(8 * i + 4));
365         val += abs1(v->B(8 * i + 5) - s->B(8 * i + 5));
366         val += abs1(v->B(8 * i + 6) - s->B(8 * i + 6));
367         val += abs1(v->B(8 * i + 7) - s->B(8 * i + 7));
368         d->Q(i) = val;
369     }
370 }
371 
372 #if SHIFT < 2
373 void glue(helper_maskmov, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
374                                   target_ulong a0)
375 {
376     int i;
377 
378     for (i = 0; i < (8 << SHIFT); i++) {
379         if (s->B(i) & 0x80) {
380             cpu_stb_data_ra(env, a0 + i, d->B(i), GETPC());
381         }
382     }
383 }
384 #endif
385 
386 #define SHUFFLE4(F, a, b, offset) do {      \
387     r0 = a->F((order & 3) + offset);        \
388     r1 = a->F(((order >> 2) & 3) + offset); \
389     r2 = b->F(((order >> 4) & 3) + offset); \
390     r3 = b->F(((order >> 6) & 3) + offset); \
391     d->F(offset) = r0;                      \
392     d->F(offset + 1) = r1;                  \
393     d->F(offset + 2) = r2;                  \
394     d->F(offset + 3) = r3;                  \
395     } while (0)
396 
397 #if SHIFT == 0
398 void glue(helper_pshufw, SUFFIX)(Reg *d, Reg *s, int order)
399 {
400     uint16_t r0, r1, r2, r3;
401 
402     SHUFFLE4(W, s, s, 0);
403 }
404 #else
405 void glue(helper_shufps, SUFFIX)(Reg *d, Reg *v, Reg *s, int order)
406 {
407     uint32_t r0, r1, r2, r3;
408     int i;
409 
410     for (i = 0; i < 2 << SHIFT; i += 4) {
411         SHUFFLE4(L, v, s, i);
412     }
413 }
414 
415 void glue(helper_shufpd, SUFFIX)(Reg *d, Reg *v, Reg *s, int order)
416 {
417     uint64_t r0, r1;
418     int i;
419 
420     for (i = 0; i < 1 << SHIFT; i += 2) {
421         r0 = v->Q(((order & 1) & 1) + i);
422         r1 = s->Q(((order >> 1) & 1) + i);
423         d->Q(i) = r0;
424         d->Q(i + 1) = r1;
425         order >>= 2;
426     }
427 }
428 
429 void glue(helper_pshufd, SUFFIX)(Reg *d, Reg *s, int order)
430 {
431     uint32_t r0, r1, r2, r3;
432     int i;
433 
434     for (i = 0; i < 2 << SHIFT; i += 4) {
435         SHUFFLE4(L, s, s, i);
436     }
437 }
438 
439 void glue(helper_pshuflw, SUFFIX)(Reg *d, Reg *s, int order)
440 {
441     uint16_t r0, r1, r2, r3;
442     int i, j;
443 
444     for (i = 0, j = 1; j < 1 << SHIFT; i += 8, j += 2) {
445         SHUFFLE4(W, s, s, i);
446         d->Q(j) = s->Q(j);
447     }
448 }
449 
450 void glue(helper_pshufhw, SUFFIX)(Reg *d, Reg *s, int order)
451 {
452     uint16_t r0, r1, r2, r3;
453     int i, j;
454 
455     for (i = 4, j = 0; j < 1 << SHIFT; i += 8, j += 2) {
456         d->Q(j) = s->Q(j);
457         SHUFFLE4(W, s, s, i);
458     }
459 }
460 #endif
461 
462 #if SHIFT >= 1
463 /* FPU ops */
464 /* XXX: not accurate */
465 
466 #define SSE_HELPER_P(name, F)                                           \
467     void glue(helper_ ## name ## ps, SUFFIX)(CPUX86State *env,          \
468             Reg *d, Reg *v, Reg *s)                                     \
469     {                                                                   \
470         int i;                                                          \
471         for (i = 0; i < 2 << SHIFT; i++) {                              \
472             d->ZMM_S(i) = F(32, v->ZMM_S(i), s->ZMM_S(i));              \
473         }                                                               \
474     }                                                                   \
475                                                                         \
476     void glue(helper_ ## name ## pd, SUFFIX)(CPUX86State *env,          \
477             Reg *d, Reg *v, Reg *s)                                     \
478     {                                                                   \
479         int i;                                                          \
480         for (i = 0; i < 1 << SHIFT; i++) {                              \
481             d->ZMM_D(i) = F(64, v->ZMM_D(i), s->ZMM_D(i));              \
482         }                                                               \
483     }
484 
485 #if SHIFT == 1
486 
487 #define SSE_HELPER_S(name, F)                                           \
488     SSE_HELPER_P(name, F)                                               \
489                                                                         \
490     void helper_ ## name ## ss(CPUX86State *env, Reg *d, Reg *v, Reg *s)\
491     {                                                                   \
492         int i;                                                          \
493         d->ZMM_S(0) = F(32, v->ZMM_S(0), s->ZMM_S(0));                  \
494         for (i = 1; i < 2 << SHIFT; i++) {                              \
495             d->ZMM_L(i) = v->ZMM_L(i);                                  \
496         }                                                               \
497     }                                                                   \
498                                                                         \
499     void helper_ ## name ## sd(CPUX86State *env, Reg *d, Reg *v, Reg *s)\
500     {                                                                   \
501         int i;                                                          \
502         d->ZMM_D(0) = F(64, v->ZMM_D(0), s->ZMM_D(0));                  \
503         for (i = 1; i < 1 << SHIFT; i++) {                              \
504             d->ZMM_Q(i) = v->ZMM_Q(i);                                  \
505         }                                                               \
506     }
507 
508 #else
509 
510 #define SSE_HELPER_S(name, F) SSE_HELPER_P(name, F)
511 
512 #endif
513 
514 #define FPU_ADD(size, a, b) float ## size ## _add(a, b, &env->sse_status)
515 #define FPU_SUB(size, a, b) float ## size ## _sub(a, b, &env->sse_status)
516 #define FPU_MUL(size, a, b) float ## size ## _mul(a, b, &env->sse_status)
517 #define FPU_DIV(size, a, b) float ## size ## _div(a, b, &env->sse_status)
518 
519 /* Note that the choice of comparison op here is important to get the
520  * special cases right: for min and max Intel specifies that (-0,0),
521  * (NaN, anything) and (anything, NaN) return the second argument.
522  */
523 #define FPU_MIN(size, a, b)                                     \
524     (float ## size ## _lt(a, b, &env->sse_status) ? (a) : (b))
525 #define FPU_MAX(size, a, b)                                     \
526     (float ## size ## _lt(b, a, &env->sse_status) ? (a) : (b))
527 
528 SSE_HELPER_S(add, FPU_ADD)
529 SSE_HELPER_S(sub, FPU_SUB)
530 SSE_HELPER_S(mul, FPU_MUL)
531 SSE_HELPER_S(div, FPU_DIV)
532 SSE_HELPER_S(min, FPU_MIN)
533 SSE_HELPER_S(max, FPU_MAX)
534 
535 void glue(helper_sqrtps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
536 {
537     int i;
538     for (i = 0; i < 2 << SHIFT; i++) {
539         d->ZMM_S(i) = float32_sqrt(s->ZMM_S(i), &env->sse_status);
540     }
541 }
542 
543 void glue(helper_sqrtpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
544 {
545     int i;
546     for (i = 0; i < 1 << SHIFT; i++) {
547         d->ZMM_D(i) = float64_sqrt(s->ZMM_D(i), &env->sse_status);
548     }
549 }
550 
551 #if SHIFT == 1
552 void helper_sqrtss(CPUX86State *env, Reg *d, Reg *v, Reg *s)
553 {
554     int i;
555     d->ZMM_S(0) = float32_sqrt(s->ZMM_S(0), &env->sse_status);
556     for (i = 1; i < 2 << SHIFT; i++) {
557         d->ZMM_L(i) = v->ZMM_L(i);
558     }
559 }
560 
561 void helper_sqrtsd(CPUX86State *env, Reg *d, Reg *v, Reg *s)
562 {
563     int i;
564     d->ZMM_D(0) = float64_sqrt(s->ZMM_D(0), &env->sse_status);
565     for (i = 1; i < 1 << SHIFT; i++) {
566         d->ZMM_Q(i) = v->ZMM_Q(i);
567     }
568 }
569 #endif
570 
571 /* float to float conversions */
572 void glue(helper_cvtps2pd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
573 {
574     int i;
575     for (i = 1 << SHIFT; --i >= 0; ) {
576         d->ZMM_D(i) = float32_to_float64(s->ZMM_S(i), &env->sse_status);
577     }
578 }
579 
580 void glue(helper_cvtpd2ps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
581 {
582     int i;
583     for (i = 0; i < 1 << SHIFT; i++) {
584          d->ZMM_S(i) = float64_to_float32(s->ZMM_D(i), &env->sse_status);
585     }
586     for (i >>= 1; i < 1 << SHIFT; i++) {
587          d->Q(i) = 0;
588     }
589 }
590 
591 #if SHIFT >= 1
592 void glue(helper_cvtph2ps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
593 {
594     int i;
595 
596     for (i = 2 << SHIFT; --i >= 0; ) {
597          d->ZMM_S(i) = float16_to_float32(s->ZMM_H(i), true, &env->sse_status);
598     }
599 }
600 
601 void glue(helper_cvtps2ph, SUFFIX)(CPUX86State *env, Reg *d, Reg *s, int mode)
602 {
603     int i;
604     FloatRoundMode prev_rounding_mode = env->sse_status.float_rounding_mode;
605     if (!(mode & (1 << 2))) {
606         set_x86_rounding_mode(mode & 3, &env->sse_status);
607     }
608 
609     for (i = 0; i < 2 << SHIFT; i++) {
610         d->ZMM_H(i) = float32_to_float16(s->ZMM_S(i), true, &env->sse_status);
611     }
612     for (i >>= 2; i < 1 << SHIFT; i++) {
613         d->Q(i) = 0;
614     }
615 
616     env->sse_status.float_rounding_mode = prev_rounding_mode;
617 }
618 #endif
619 
620 #if SHIFT == 1
621 void helper_cvtss2sd(CPUX86State *env, Reg *d, Reg *v, Reg *s)
622 {
623     int i;
624     d->ZMM_D(0) = float32_to_float64(s->ZMM_S(0), &env->sse_status);
625     for (i = 1; i < 1 << SHIFT; i++) {
626         d->ZMM_Q(i) = v->ZMM_Q(i);
627     }
628 }
629 
630 void helper_cvtsd2ss(CPUX86State *env, Reg *d, Reg *v, Reg *s)
631 {
632     int i;
633     d->ZMM_S(0) = float64_to_float32(s->ZMM_D(0), &env->sse_status);
634     for (i = 1; i < 2 << SHIFT; i++) {
635         d->ZMM_L(i) = v->ZMM_L(i);
636     }
637 }
638 #endif
639 
640 /* integer to float */
641 void glue(helper_cvtdq2ps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
642 {
643     int i;
644     for (i = 0; i < 2 << SHIFT; i++) {
645         d->ZMM_S(i) = int32_to_float32(s->ZMM_L(i), &env->sse_status);
646     }
647 }
648 
649 void glue(helper_cvtdq2pd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
650 {
651     int i;
652     for (i = 1 << SHIFT; --i >= 0; ) {
653         int32_t l = s->ZMM_L(i);
654         d->ZMM_D(i) = int32_to_float64(l, &env->sse_status);
655     }
656 }
657 
658 #if SHIFT == 1
659 void helper_cvtpi2ps(CPUX86State *env, ZMMReg *d, MMXReg *s)
660 {
661     d->ZMM_S(0) = int32_to_float32(s->MMX_L(0), &env->sse_status);
662     d->ZMM_S(1) = int32_to_float32(s->MMX_L(1), &env->sse_status);
663 }
664 
665 void helper_cvtpi2pd(CPUX86State *env, ZMMReg *d, MMXReg *s)
666 {
667     d->ZMM_D(0) = int32_to_float64(s->MMX_L(0), &env->sse_status);
668     d->ZMM_D(1) = int32_to_float64(s->MMX_L(1), &env->sse_status);
669 }
670 
671 void helper_cvtsi2ss(CPUX86State *env, ZMMReg *d, uint32_t val)
672 {
673     d->ZMM_S(0) = int32_to_float32(val, &env->sse_status);
674 }
675 
676 void helper_cvtsi2sd(CPUX86State *env, ZMMReg *d, uint32_t val)
677 {
678     d->ZMM_D(0) = int32_to_float64(val, &env->sse_status);
679 }
680 
681 #ifdef TARGET_X86_64
682 void helper_cvtsq2ss(CPUX86State *env, ZMMReg *d, uint64_t val)
683 {
684     d->ZMM_S(0) = int64_to_float32(val, &env->sse_status);
685 }
686 
687 void helper_cvtsq2sd(CPUX86State *env, ZMMReg *d, uint64_t val)
688 {
689     d->ZMM_D(0) = int64_to_float64(val, &env->sse_status);
690 }
691 #endif
692 
693 #endif
694 
695 /* float to integer */
696 
697 #if SHIFT == 1
698 /*
699  * x86 mandates that we return the indefinite integer value for the result
700  * of any float-to-integer conversion that raises the 'invalid' exception.
701  * Wrap the softfloat functions to get this behaviour.
702  */
703 #define WRAP_FLOATCONV(RETTYPE, FN, FLOATTYPE, INDEFVALUE)              \
704     static inline RETTYPE x86_##FN(FLOATTYPE a, float_status *s)        \
705     {                                                                   \
706         int oldflags, newflags;                                         \
707         RETTYPE r;                                                      \
708                                                                         \
709         oldflags = get_float_exception_flags(s);                        \
710         set_float_exception_flags(0, s);                                \
711         r = FN(a, s);                                                   \
712         newflags = get_float_exception_flags(s);                        \
713         if (newflags & float_flag_invalid) {                            \
714             r = INDEFVALUE;                                             \
715         }                                                               \
716         set_float_exception_flags(newflags | oldflags, s);              \
717         return r;                                                       \
718     }
719 
720 WRAP_FLOATCONV(int32_t, float32_to_int32, float32, INT32_MIN)
721 WRAP_FLOATCONV(int32_t, float32_to_int32_round_to_zero, float32, INT32_MIN)
722 WRAP_FLOATCONV(int32_t, float64_to_int32, float64, INT32_MIN)
723 WRAP_FLOATCONV(int32_t, float64_to_int32_round_to_zero, float64, INT32_MIN)
724 WRAP_FLOATCONV(int64_t, float32_to_int64, float32, INT64_MIN)
725 WRAP_FLOATCONV(int64_t, float32_to_int64_round_to_zero, float32, INT64_MIN)
726 WRAP_FLOATCONV(int64_t, float64_to_int64, float64, INT64_MIN)
727 WRAP_FLOATCONV(int64_t, float64_to_int64_round_to_zero, float64, INT64_MIN)
728 #endif
729 
730 void glue(helper_cvtps2dq, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
731 {
732     int i;
733     for (i = 0; i < 2 << SHIFT; i++) {
734         d->ZMM_L(i) = x86_float32_to_int32(s->ZMM_S(i), &env->sse_status);
735     }
736 }
737 
738 void glue(helper_cvtpd2dq, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
739 {
740     int i;
741     for (i = 0; i < 1 << SHIFT; i++) {
742         d->ZMM_L(i) = x86_float64_to_int32(s->ZMM_D(i), &env->sse_status);
743     }
744     for (i >>= 1; i < 1 << SHIFT; i++) {
745          d->Q(i) = 0;
746     }
747 }
748 
749 #if SHIFT == 1
750 void helper_cvtps2pi(CPUX86State *env, MMXReg *d, ZMMReg *s)
751 {
752     d->MMX_L(0) = x86_float32_to_int32(s->ZMM_S(0), &env->sse_status);
753     d->MMX_L(1) = x86_float32_to_int32(s->ZMM_S(1), &env->sse_status);
754 }
755 
756 void helper_cvtpd2pi(CPUX86State *env, MMXReg *d, ZMMReg *s)
757 {
758     d->MMX_L(0) = x86_float64_to_int32(s->ZMM_D(0), &env->sse_status);
759     d->MMX_L(1) = x86_float64_to_int32(s->ZMM_D(1), &env->sse_status);
760 }
761 
762 int32_t helper_cvtss2si(CPUX86State *env, ZMMReg *s)
763 {
764     return x86_float32_to_int32(s->ZMM_S(0), &env->sse_status);
765 }
766 
767 int32_t helper_cvtsd2si(CPUX86State *env, ZMMReg *s)
768 {
769     return x86_float64_to_int32(s->ZMM_D(0), &env->sse_status);
770 }
771 
772 #ifdef TARGET_X86_64
773 int64_t helper_cvtss2sq(CPUX86State *env, ZMMReg *s)
774 {
775     return x86_float32_to_int64(s->ZMM_S(0), &env->sse_status);
776 }
777 
778 int64_t helper_cvtsd2sq(CPUX86State *env, ZMMReg *s)
779 {
780     return x86_float64_to_int64(s->ZMM_D(0), &env->sse_status);
781 }
782 #endif
783 #endif
784 
785 /* float to integer truncated */
786 void glue(helper_cvttps2dq, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
787 {
788     int i;
789     for (i = 0; i < 2 << SHIFT; i++) {
790         d->ZMM_L(i) = x86_float32_to_int32_round_to_zero(s->ZMM_S(i),
791                                                          &env->sse_status);
792     }
793 }
794 
795 void glue(helper_cvttpd2dq, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
796 {
797     int i;
798     for (i = 0; i < 1 << SHIFT; i++) {
799         d->ZMM_L(i) = x86_float64_to_int32_round_to_zero(s->ZMM_D(i),
800                                                          &env->sse_status);
801     }
802     for (i >>= 1; i < 1 << SHIFT; i++) {
803          d->Q(i) = 0;
804     }
805 }
806 
807 #if SHIFT == 1
808 void helper_cvttps2pi(CPUX86State *env, MMXReg *d, ZMMReg *s)
809 {
810     d->MMX_L(0) = x86_float32_to_int32_round_to_zero(s->ZMM_S(0), &env->sse_status);
811     d->MMX_L(1) = x86_float32_to_int32_round_to_zero(s->ZMM_S(1), &env->sse_status);
812 }
813 
814 void helper_cvttpd2pi(CPUX86State *env, MMXReg *d, ZMMReg *s)
815 {
816     d->MMX_L(0) = x86_float64_to_int32_round_to_zero(s->ZMM_D(0), &env->sse_status);
817     d->MMX_L(1) = x86_float64_to_int32_round_to_zero(s->ZMM_D(1), &env->sse_status);
818 }
819 
820 int32_t helper_cvttss2si(CPUX86State *env, ZMMReg *s)
821 {
822     return x86_float32_to_int32_round_to_zero(s->ZMM_S(0), &env->sse_status);
823 }
824 
825 int32_t helper_cvttsd2si(CPUX86State *env, ZMMReg *s)
826 {
827     return x86_float64_to_int32_round_to_zero(s->ZMM_D(0), &env->sse_status);
828 }
829 
830 #ifdef TARGET_X86_64
831 int64_t helper_cvttss2sq(CPUX86State *env, ZMMReg *s)
832 {
833     return x86_float32_to_int64_round_to_zero(s->ZMM_S(0), &env->sse_status);
834 }
835 
836 int64_t helper_cvttsd2sq(CPUX86State *env, ZMMReg *s)
837 {
838     return x86_float64_to_int64_round_to_zero(s->ZMM_D(0), &env->sse_status);
839 }
840 #endif
841 #endif
842 
843 void glue(helper_rsqrtps, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
844 {
845     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
846     int i;
847     for (i = 0; i < 2 << SHIFT; i++) {
848         d->ZMM_S(i) = float32_div(float32_one,
849                                   float32_sqrt(s->ZMM_S(i), &env->sse_status),
850                                   &env->sse_status);
851     }
852     set_float_exception_flags(old_flags, &env->sse_status);
853 }
854 
855 #if SHIFT == 1
856 void helper_rsqrtss(CPUX86State *env, ZMMReg *d, ZMMReg *v, ZMMReg *s)
857 {
858     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
859     int i;
860     d->ZMM_S(0) = float32_div(float32_one,
861                               float32_sqrt(s->ZMM_S(0), &env->sse_status),
862                               &env->sse_status);
863     set_float_exception_flags(old_flags, &env->sse_status);
864     for (i = 1; i < 2 << SHIFT; i++) {
865         d->ZMM_L(i) = v->ZMM_L(i);
866     }
867 }
868 #endif
869 
870 void glue(helper_rcpps, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
871 {
872     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
873     int i;
874     for (i = 0; i < 2 << SHIFT; i++) {
875         d->ZMM_S(i) = float32_div(float32_one, s->ZMM_S(i), &env->sse_status);
876     }
877     set_float_exception_flags(old_flags, &env->sse_status);
878 }
879 
880 #if SHIFT == 1
881 void helper_rcpss(CPUX86State *env, ZMMReg *d, ZMMReg *v, ZMMReg *s)
882 {
883     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
884     int i;
885     d->ZMM_S(0) = float32_div(float32_one, s->ZMM_S(0), &env->sse_status);
886     for (i = 1; i < 2 << SHIFT; i++) {
887         d->ZMM_L(i) = v->ZMM_L(i);
888     }
889     set_float_exception_flags(old_flags, &env->sse_status);
890 }
891 #endif
892 
893 #if SHIFT == 1
894 static inline uint64_t helper_extrq(uint64_t src, int shift, int len)
895 {
896     uint64_t mask;
897 
898     if (len == 0) {
899         mask = ~0LL;
900     } else {
901         mask = (1ULL << len) - 1;
902     }
903     return (src >> shift) & mask;
904 }
905 
906 void helper_extrq_r(CPUX86State *env, ZMMReg *d, ZMMReg *s)
907 {
908     d->ZMM_Q(0) = helper_extrq(d->ZMM_Q(0), s->ZMM_B(1) & 63, s->ZMM_B(0) & 63);
909 }
910 
911 void helper_extrq_i(CPUX86State *env, ZMMReg *d, int index, int length)
912 {
913     d->ZMM_Q(0) = helper_extrq(d->ZMM_Q(0), index, length);
914 }
915 
916 static inline uint64_t helper_insertq(uint64_t dest, uint64_t src, int shift, int len)
917 {
918     uint64_t mask;
919 
920     if (len == 0) {
921         mask = ~0ULL;
922     } else {
923         mask = (1ULL << len) - 1;
924     }
925     return (dest & ~(mask << shift)) | ((src & mask) << shift);
926 }
927 
928 void helper_insertq_r(CPUX86State *env, ZMMReg *d, ZMMReg *s)
929 {
930     d->ZMM_Q(0) = helper_insertq(d->ZMM_Q(0), s->ZMM_Q(0), s->ZMM_B(9) & 63, s->ZMM_B(8) & 63);
931 }
932 
933 void helper_insertq_i(CPUX86State *env, ZMMReg *d, ZMMReg *s, int index, int length)
934 {
935     d->ZMM_Q(0) = helper_insertq(d->ZMM_Q(0), s->ZMM_Q(0), index, length);
936 }
937 #endif
938 
939 #define SSE_HELPER_HPS(name, F)  \
940 void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s) \
941 {                                                                 \
942     float32 r[2 << SHIFT];                                        \
943     int i, j, k;                                                  \
944     for (k = 0; k < 2 << SHIFT; k += LANE_WIDTH / 4) {            \
945         for (i = j = 0; j < 4; i++, j += 2) {                     \
946             r[i + k] = F(v->ZMM_S(j + k), v->ZMM_S(j + k + 1), &env->sse_status); \
947         }                                                         \
948         for (j = 0; j < 4; i++, j += 2) {                         \
949             r[i + k] = F(s->ZMM_S(j + k), s->ZMM_S(j + k + 1), &env->sse_status); \
950         }                                                         \
951     }                                                             \
952     for (i = 0; i < 2 << SHIFT; i++) {                            \
953         d->ZMM_S(i) = r[i];                                       \
954     }                                                             \
955 }
956 
957 SSE_HELPER_HPS(haddps, float32_add)
958 SSE_HELPER_HPS(hsubps, float32_sub)
959 
960 #define SSE_HELPER_HPD(name, F)  \
961 void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s) \
962 {                                                                 \
963     float64 r[1 << SHIFT];                                        \
964     int i, j, k;                                                  \
965     for (k = 0; k < 1 << SHIFT; k += LANE_WIDTH / 8) {            \
966         for (i = j = 0; j < 2; i++, j += 2) {                     \
967             r[i + k] = F(v->ZMM_D(j + k), v->ZMM_D(j + k + 1), &env->sse_status); \
968         }                                                         \
969         for (j = 0; j < 2; i++, j += 2) {                         \
970             r[i + k] = F(s->ZMM_D(j + k), s->ZMM_D(j + k + 1), &env->sse_status); \
971         }                                                         \
972     }                                                             \
973     for (i = 0; i < 1 << SHIFT; i++) {                            \
974         d->ZMM_D(i) = r[i];                                       \
975     }                                                             \
976 }
977 
978 SSE_HELPER_HPD(haddpd, float64_add)
979 SSE_HELPER_HPD(hsubpd, float64_sub)
980 
981 void glue(helper_addsubps, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
982 {
983     int i;
984     for (i = 0; i < 2 << SHIFT; i += 2) {
985         d->ZMM_S(i) = float32_sub(v->ZMM_S(i), s->ZMM_S(i), &env->sse_status);
986         d->ZMM_S(i+1) = float32_add(v->ZMM_S(i+1), s->ZMM_S(i+1), &env->sse_status);
987     }
988 }
989 
990 void glue(helper_addsubpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
991 {
992     int i;
993     for (i = 0; i < 1 << SHIFT; i += 2) {
994         d->ZMM_D(i) = float64_sub(v->ZMM_D(i), s->ZMM_D(i), &env->sse_status);
995         d->ZMM_D(i+1) = float64_add(v->ZMM_D(i+1), s->ZMM_D(i+1), &env->sse_status);
996     }
997 }
998 
999 #define SSE_HELPER_CMP_P(name, F, C)                                    \
1000     void glue(helper_ ## name ## ps, SUFFIX)(CPUX86State *env,          \
1001                                              Reg *d, Reg *v, Reg *s)    \
1002     {                                                                   \
1003         int i;                                                          \
1004         for (i = 0; i < 2 << SHIFT; i++) {                              \
1005             d->ZMM_L(i) = C(F(32, v->ZMM_S(i), s->ZMM_S(i))) ? -1 : 0;  \
1006         }                                                               \
1007     }                                                                   \
1008                                                                         \
1009     void glue(helper_ ## name ## pd, SUFFIX)(CPUX86State *env,          \
1010                                              Reg *d, Reg *v, Reg *s)    \
1011     {                                                                   \
1012         int i;                                                          \
1013         for (i = 0; i < 1 << SHIFT; i++) {                              \
1014             d->ZMM_Q(i) = C(F(64, v->ZMM_D(i), s->ZMM_D(i))) ? -1 : 0;  \
1015         }                                                               \
1016     }
1017 
1018 #if SHIFT == 1
1019 #define SSE_HELPER_CMP(name, F, C)                                          \
1020     SSE_HELPER_CMP_P(name, F, C)                                            \
1021     void helper_ ## name ## ss(CPUX86State *env, Reg *d, Reg *v, Reg *s)    \
1022     {                                                                       \
1023         int i;                                                              \
1024         d->ZMM_L(0) = C(F(32, v->ZMM_S(0), s->ZMM_S(0))) ? -1 : 0;          \
1025         for (i = 1; i < 2 << SHIFT; i++) {                                  \
1026             d->ZMM_L(i) = v->ZMM_L(i);                                      \
1027         }                                                                   \
1028     }                                                                       \
1029                                                                             \
1030     void helper_ ## name ## sd(CPUX86State *env, Reg *d, Reg *v, Reg *s)    \
1031     {                                                                       \
1032         int i;                                                              \
1033         d->ZMM_Q(0) = C(F(64, v->ZMM_D(0), s->ZMM_D(0))) ? -1 : 0;          \
1034         for (i = 1; i < 1 << SHIFT; i++) {                                  \
1035             d->ZMM_Q(i) = v->ZMM_Q(i);                                      \
1036         }                                                                   \
1037     }
1038 
1039 static inline bool FPU_EQU(FloatRelation x)
1040 {
1041     return (x == float_relation_equal || x == float_relation_unordered);
1042 }
1043 static inline bool FPU_GE(FloatRelation x)
1044 {
1045     return (x == float_relation_equal || x == float_relation_greater);
1046 }
1047 #define FPU_EQ(x) (x == float_relation_equal)
1048 #define FPU_LT(x) (x == float_relation_less)
1049 #define FPU_LE(x) (x <= float_relation_equal)
1050 #define FPU_GT(x) (x == float_relation_greater)
1051 #define FPU_UNORD(x) (x == float_relation_unordered)
1052 /* We must make sure we evaluate the argument in case it is a signalling NAN */
1053 #define FPU_FALSE(x) (x == float_relation_equal && 0)
1054 
1055 #define FPU_CMPQ(size, a, b) \
1056     float ## size ## _compare_quiet(a, b, &env->sse_status)
1057 #define FPU_CMPS(size, a, b) \
1058     float ## size ## _compare(a, b, &env->sse_status)
1059 
1060 #else
1061 #define SSE_HELPER_CMP(name, F, C) SSE_HELPER_CMP_P(name, F, C)
1062 #endif
1063 
1064 SSE_HELPER_CMP(cmpeq, FPU_CMPQ, FPU_EQ)
1065 SSE_HELPER_CMP(cmplt, FPU_CMPS, FPU_LT)
1066 SSE_HELPER_CMP(cmple, FPU_CMPS, FPU_LE)
1067 SSE_HELPER_CMP(cmpunord, FPU_CMPQ,  FPU_UNORD)
1068 SSE_HELPER_CMP(cmpneq, FPU_CMPQ, !FPU_EQ)
1069 SSE_HELPER_CMP(cmpnlt, FPU_CMPS, !FPU_LT)
1070 SSE_HELPER_CMP(cmpnle, FPU_CMPS, !FPU_LE)
1071 SSE_HELPER_CMP(cmpord, FPU_CMPQ, !FPU_UNORD)
1072 
1073 SSE_HELPER_CMP(cmpequ, FPU_CMPQ, FPU_EQU)
1074 SSE_HELPER_CMP(cmpnge, FPU_CMPS, !FPU_GE)
1075 SSE_HELPER_CMP(cmpngt, FPU_CMPS, !FPU_GT)
1076 SSE_HELPER_CMP(cmpfalse, FPU_CMPQ,  FPU_FALSE)
1077 SSE_HELPER_CMP(cmpnequ, FPU_CMPQ, !FPU_EQU)
1078 SSE_HELPER_CMP(cmpge, FPU_CMPS, FPU_GE)
1079 SSE_HELPER_CMP(cmpgt, FPU_CMPS, FPU_GT)
1080 SSE_HELPER_CMP(cmptrue, FPU_CMPQ,  !FPU_FALSE)
1081 
1082 SSE_HELPER_CMP(cmpeqs, FPU_CMPS, FPU_EQ)
1083 SSE_HELPER_CMP(cmpltq, FPU_CMPQ, FPU_LT)
1084 SSE_HELPER_CMP(cmpleq, FPU_CMPQ, FPU_LE)
1085 SSE_HELPER_CMP(cmpunords, FPU_CMPS,  FPU_UNORD)
1086 SSE_HELPER_CMP(cmpneqq, FPU_CMPS, !FPU_EQ)
1087 SSE_HELPER_CMP(cmpnltq, FPU_CMPQ, !FPU_LT)
1088 SSE_HELPER_CMP(cmpnleq, FPU_CMPQ, !FPU_LE)
1089 SSE_HELPER_CMP(cmpords, FPU_CMPS, !FPU_UNORD)
1090 
1091 SSE_HELPER_CMP(cmpequs, FPU_CMPS, FPU_EQU)
1092 SSE_HELPER_CMP(cmpngeq, FPU_CMPQ, !FPU_GE)
1093 SSE_HELPER_CMP(cmpngtq, FPU_CMPQ, !FPU_GT)
1094 SSE_HELPER_CMP(cmpfalses, FPU_CMPS,  FPU_FALSE)
1095 SSE_HELPER_CMP(cmpnequs, FPU_CMPS, !FPU_EQU)
1096 SSE_HELPER_CMP(cmpgeq, FPU_CMPQ, FPU_GE)
1097 SSE_HELPER_CMP(cmpgtq, FPU_CMPQ, FPU_GT)
1098 SSE_HELPER_CMP(cmptrues, FPU_CMPS,  !FPU_FALSE)
1099 
1100 #undef SSE_HELPER_CMP
1101 
1102 #if SHIFT == 1
1103 static const int comis_eflags[4] = {CC_C, CC_Z, 0, CC_Z | CC_P | CC_C};
1104 
1105 void helper_ucomiss(CPUX86State *env, Reg *d, Reg *s)
1106 {
1107     FloatRelation ret;
1108     float32 s0, s1;
1109 
1110     s0 = d->ZMM_S(0);
1111     s1 = s->ZMM_S(0);
1112     ret = float32_compare_quiet(s0, s1, &env->sse_status);
1113     CC_SRC = comis_eflags[ret + 1];
1114     CC_OP = CC_OP_EFLAGS;
1115 }
1116 
1117 void helper_comiss(CPUX86State *env, Reg *d, Reg *s)
1118 {
1119     FloatRelation ret;
1120     float32 s0, s1;
1121 
1122     s0 = d->ZMM_S(0);
1123     s1 = s->ZMM_S(0);
1124     ret = float32_compare(s0, s1, &env->sse_status);
1125     CC_SRC = comis_eflags[ret + 1];
1126     CC_OP = CC_OP_EFLAGS;
1127 }
1128 
1129 void helper_ucomisd(CPUX86State *env, Reg *d, Reg *s)
1130 {
1131     FloatRelation ret;
1132     float64 d0, d1;
1133 
1134     d0 = d->ZMM_D(0);
1135     d1 = s->ZMM_D(0);
1136     ret = float64_compare_quiet(d0, d1, &env->sse_status);
1137     CC_SRC = comis_eflags[ret + 1];
1138     CC_OP = CC_OP_EFLAGS;
1139 }
1140 
1141 void helper_comisd(CPUX86State *env, Reg *d, Reg *s)
1142 {
1143     FloatRelation ret;
1144     float64 d0, d1;
1145 
1146     d0 = d->ZMM_D(0);
1147     d1 = s->ZMM_D(0);
1148     ret = float64_compare(d0, d1, &env->sse_status);
1149     CC_SRC = comis_eflags[ret + 1];
1150     CC_OP = CC_OP_EFLAGS;
1151 }
1152 #endif
1153 
1154 uint32_t glue(helper_movmskps, SUFFIX)(CPUX86State *env, Reg *s)
1155 {
1156     uint32_t mask;
1157     int i;
1158 
1159     mask = 0;
1160     for (i = 0; i < 2 << SHIFT; i++) {
1161         mask |= (s->ZMM_L(i) >> (31 - i)) & (1 << i);
1162     }
1163     return mask;
1164 }
1165 
1166 uint32_t glue(helper_movmskpd, SUFFIX)(CPUX86State *env, Reg *s)
1167 {
1168     uint32_t mask;
1169     int i;
1170 
1171     mask = 0;
1172     for (i = 0; i < 1 << SHIFT; i++) {
1173         mask |= (s->ZMM_Q(i) >> (63 - i)) & (1 << i);
1174     }
1175     return mask;
1176 }
1177 
1178 #endif
1179 
1180 #define PACK_HELPER_B(name, F) \
1181 void glue(helper_pack ## name, SUFFIX)(CPUX86State *env,      \
1182         Reg *d, Reg *v, Reg *s)                               \
1183 {                                                             \
1184     uint8_t r[PACK_WIDTH * 2];                                \
1185     int j, k;                                                 \
1186     for (j = 0; j < 4 << SHIFT; j += PACK_WIDTH) {            \
1187         for (k = 0; k < PACK_WIDTH; k++) {                    \
1188             r[k] = F((int16_t)v->W(j + k));                   \
1189         }                                                     \
1190         for (k = 0; k < PACK_WIDTH; k++) {                    \
1191             r[PACK_WIDTH + k] = F((int16_t)s->W(j + k));      \
1192         }                                                     \
1193         for (k = 0; k < PACK_WIDTH * 2; k++) {                \
1194             d->B(2 * j + k) = r[k];                           \
1195         }                                                     \
1196     }                                                         \
1197 }
1198 
1199 PACK_HELPER_B(sswb, satsb)
1200 PACK_HELPER_B(uswb, satub)
1201 
1202 void glue(helper_packssdw, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
1203 {
1204     uint16_t r[PACK_WIDTH];
1205     int j, k;
1206 
1207     for (j = 0; j < 2 << SHIFT; j += PACK_WIDTH / 2) {
1208         for (k = 0; k < PACK_WIDTH / 2; k++) {
1209             r[k] = satsw(v->L(j + k));
1210         }
1211         for (k = 0; k < PACK_WIDTH / 2; k++) {
1212             r[PACK_WIDTH / 2 + k] = satsw(s->L(j + k));
1213         }
1214         for (k = 0; k < PACK_WIDTH; k++) {
1215             d->W(2 * j + k) = r[k];
1216         }
1217     }
1218 }
1219 
1220 #define UNPCK_OP(base_name, base)                                       \
1221                                                                         \
1222     void glue(helper_punpck ## base_name ## bw, SUFFIX)(CPUX86State *env,\
1223                                                 Reg *d, Reg *v, Reg *s) \
1224     {                                                                   \
1225         uint8_t r[PACK_WIDTH * 2];                                      \
1226         int j, i;                                                       \
1227                                                                         \
1228         for (j = 0; j < 8 << SHIFT; ) {                                 \
1229             int k = j + base * PACK_WIDTH;                              \
1230             for (i = 0; i < PACK_WIDTH; i++) {                          \
1231                 r[2 * i] = v->B(k + i);                                 \
1232                 r[2 * i + 1] = s->B(k + i);                             \
1233             }                                                           \
1234             for (i = 0; i < PACK_WIDTH * 2; i++, j++) {                 \
1235                 d->B(j) = r[i];                                         \
1236             }                                                           \
1237         }                                                               \
1238     }                                                                   \
1239                                                                         \
1240     void glue(helper_punpck ## base_name ## wd, SUFFIX)(CPUX86State *env,\
1241                                                 Reg *d, Reg *v, Reg *s) \
1242     {                                                                   \
1243         uint16_t r[PACK_WIDTH];                                         \
1244         int j, i;                                                       \
1245                                                                         \
1246         for (j = 0; j < 4 << SHIFT; ) {                                 \
1247             int k = j + base * PACK_WIDTH / 2;                          \
1248             for (i = 0; i < PACK_WIDTH / 2; i++) {                      \
1249                 r[2 * i] = v->W(k + i);                                 \
1250                 r[2 * i + 1] = s->W(k + i);                             \
1251             }                                                           \
1252             for (i = 0; i < PACK_WIDTH; i++, j++) {                     \
1253                 d->W(j) = r[i];                                         \
1254             }                                                           \
1255         }                                                               \
1256     }                                                                   \
1257                                                                         \
1258     void glue(helper_punpck ## base_name ## dq, SUFFIX)(CPUX86State *env,\
1259                                                 Reg *d, Reg *v, Reg *s) \
1260     {                                                                   \
1261         uint32_t r[PACK_WIDTH / 2];                                     \
1262         int j, i;                                                       \
1263                                                                         \
1264         for (j = 0; j < 2 << SHIFT; ) {                                 \
1265             int k = j + base * PACK_WIDTH / 4;                          \
1266             for (i = 0; i < PACK_WIDTH / 4; i++) {                      \
1267                 r[2 * i] = v->L(k + i);                                 \
1268                 r[2 * i + 1] = s->L(k + i);                             \
1269             }                                                           \
1270             for (i = 0; i < PACK_WIDTH / 2; i++, j++) {                 \
1271                 d->L(j) = r[i];                                         \
1272             }                                                           \
1273         }                                                               \
1274     }                                                                   \
1275                                                                         \
1276     XMM_ONLY(                                                           \
1277              void glue(helper_punpck ## base_name ## qdq, SUFFIX)(      \
1278                         CPUX86State *env, Reg *d, Reg *v, Reg *s)       \
1279              {                                                          \
1280                  uint64_t r[2];                                         \
1281                  int i;                                                 \
1282                                                                         \
1283                  for (i = 0; i < 1 << SHIFT; i += 2) {                  \
1284                      r[0] = v->Q(base + i);                             \
1285                      r[1] = s->Q(base + i);                             \
1286                      d->Q(i) = r[0];                                    \
1287                      d->Q(i + 1) = r[1];                                \
1288                  }                                                      \
1289              }                                                          \
1290                                                                         )
1291 
1292 UNPCK_OP(l, 0)
1293 UNPCK_OP(h, 1)
1294 
1295 #undef PACK_WIDTH
1296 #undef PACK_HELPER_B
1297 #undef UNPCK_OP
1298 
1299 
1300 /* 3DNow! float ops */
1301 #if SHIFT == 0
1302 void helper_pi2fd(CPUX86State *env, MMXReg *d, MMXReg *s)
1303 {
1304     d->MMX_S(0) = int32_to_float32(s->MMX_L(0), &env->mmx_status);
1305     d->MMX_S(1) = int32_to_float32(s->MMX_L(1), &env->mmx_status);
1306 }
1307 
1308 void helper_pi2fw(CPUX86State *env, MMXReg *d, MMXReg *s)
1309 {
1310     d->MMX_S(0) = int32_to_float32((int16_t)s->MMX_W(0), &env->mmx_status);
1311     d->MMX_S(1) = int32_to_float32((int16_t)s->MMX_W(2), &env->mmx_status);
1312 }
1313 
1314 void helper_pf2id(CPUX86State *env, MMXReg *d, MMXReg *s)
1315 {
1316     d->MMX_L(0) = float32_to_int32_round_to_zero(s->MMX_S(0), &env->mmx_status);
1317     d->MMX_L(1) = float32_to_int32_round_to_zero(s->MMX_S(1), &env->mmx_status);
1318 }
1319 
1320 void helper_pf2iw(CPUX86State *env, MMXReg *d, MMXReg *s)
1321 {
1322     d->MMX_L(0) = satsw(float32_to_int32_round_to_zero(s->MMX_S(0),
1323                                                        &env->mmx_status));
1324     d->MMX_L(1) = satsw(float32_to_int32_round_to_zero(s->MMX_S(1),
1325                                                        &env->mmx_status));
1326 }
1327 
1328 void helper_pfacc(CPUX86State *env, MMXReg *d, MMXReg *s)
1329 {
1330     float32 r;
1331 
1332     r = float32_add(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1333     d->MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1334     d->MMX_S(0) = r;
1335 }
1336 
1337 void helper_pfadd(CPUX86State *env, MMXReg *d, MMXReg *s)
1338 {
1339     d->MMX_S(0) = float32_add(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1340     d->MMX_S(1) = float32_add(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1341 }
1342 
1343 void helper_pfcmpeq(CPUX86State *env, MMXReg *d, MMXReg *s)
1344 {
1345     d->MMX_L(0) = float32_eq_quiet(d->MMX_S(0), s->MMX_S(0),
1346                                    &env->mmx_status) ? -1 : 0;
1347     d->MMX_L(1) = float32_eq_quiet(d->MMX_S(1), s->MMX_S(1),
1348                                    &env->mmx_status) ? -1 : 0;
1349 }
1350 
1351 void helper_pfcmpge(CPUX86State *env, MMXReg *d, MMXReg *s)
1352 {
1353     d->MMX_L(0) = float32_le(s->MMX_S(0), d->MMX_S(0),
1354                              &env->mmx_status) ? -1 : 0;
1355     d->MMX_L(1) = float32_le(s->MMX_S(1), d->MMX_S(1),
1356                              &env->mmx_status) ? -1 : 0;
1357 }
1358 
1359 void helper_pfcmpgt(CPUX86State *env, MMXReg *d, MMXReg *s)
1360 {
1361     d->MMX_L(0) = float32_lt(s->MMX_S(0), d->MMX_S(0),
1362                              &env->mmx_status) ? -1 : 0;
1363     d->MMX_L(1) = float32_lt(s->MMX_S(1), d->MMX_S(1),
1364                              &env->mmx_status) ? -1 : 0;
1365 }
1366 
1367 void helper_pfmax(CPUX86State *env, MMXReg *d, MMXReg *s)
1368 {
1369     if (float32_lt(d->MMX_S(0), s->MMX_S(0), &env->mmx_status)) {
1370         d->MMX_S(0) = s->MMX_S(0);
1371     }
1372     if (float32_lt(d->MMX_S(1), s->MMX_S(1), &env->mmx_status)) {
1373         d->MMX_S(1) = s->MMX_S(1);
1374     }
1375 }
1376 
1377 void helper_pfmin(CPUX86State *env, MMXReg *d, MMXReg *s)
1378 {
1379     if (float32_lt(s->MMX_S(0), d->MMX_S(0), &env->mmx_status)) {
1380         d->MMX_S(0) = s->MMX_S(0);
1381     }
1382     if (float32_lt(s->MMX_S(1), d->MMX_S(1), &env->mmx_status)) {
1383         d->MMX_S(1) = s->MMX_S(1);
1384     }
1385 }
1386 
1387 void helper_pfmul(CPUX86State *env, MMXReg *d, MMXReg *s)
1388 {
1389     d->MMX_S(0) = float32_mul(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1390     d->MMX_S(1) = float32_mul(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1391 }
1392 
1393 void helper_pfnacc(CPUX86State *env, MMXReg *d, MMXReg *s)
1394 {
1395     float32 r;
1396 
1397     r = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1398     d->MMX_S(1) = float32_sub(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1399     d->MMX_S(0) = r;
1400 }
1401 
1402 void helper_pfpnacc(CPUX86State *env, MMXReg *d, MMXReg *s)
1403 {
1404     float32 r;
1405 
1406     r = float32_sub(d->MMX_S(0), d->MMX_S(1), &env->mmx_status);
1407     d->MMX_S(1) = float32_add(s->MMX_S(0), s->MMX_S(1), &env->mmx_status);
1408     d->MMX_S(0) = r;
1409 }
1410 
1411 void helper_pfrcp(CPUX86State *env, MMXReg *d, MMXReg *s)
1412 {
1413     d->MMX_S(0) = float32_div(float32_one, s->MMX_S(0), &env->mmx_status);
1414     d->MMX_S(1) = d->MMX_S(0);
1415 }
1416 
1417 void helper_pfrsqrt(CPUX86State *env, MMXReg *d, MMXReg *s)
1418 {
1419     d->MMX_L(1) = s->MMX_L(0) & 0x7fffffff;
1420     d->MMX_S(1) = float32_div(float32_one,
1421                               float32_sqrt(d->MMX_S(1), &env->mmx_status),
1422                               &env->mmx_status);
1423     d->MMX_L(1) |= s->MMX_L(0) & 0x80000000;
1424     d->MMX_L(0) = d->MMX_L(1);
1425 }
1426 
1427 void helper_pfsub(CPUX86State *env, MMXReg *d, MMXReg *s)
1428 {
1429     d->MMX_S(0) = float32_sub(d->MMX_S(0), s->MMX_S(0), &env->mmx_status);
1430     d->MMX_S(1) = float32_sub(d->MMX_S(1), s->MMX_S(1), &env->mmx_status);
1431 }
1432 
1433 void helper_pfsubr(CPUX86State *env, MMXReg *d, MMXReg *s)
1434 {
1435     d->MMX_S(0) = float32_sub(s->MMX_S(0), d->MMX_S(0), &env->mmx_status);
1436     d->MMX_S(1) = float32_sub(s->MMX_S(1), d->MMX_S(1), &env->mmx_status);
1437 }
1438 
1439 void helper_pswapd(CPUX86State *env, MMXReg *d, MMXReg *s)
1440 {
1441     uint32_t r;
1442 
1443     r = s->MMX_L(0);
1444     d->MMX_L(0) = s->MMX_L(1);
1445     d->MMX_L(1) = r;
1446 }
1447 #endif
1448 
1449 /* SSSE3 op helpers */
1450 void glue(helper_pshufb, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
1451 {
1452     int i;
1453 #if SHIFT == 0
1454     uint8_t r[8];
1455 
1456     for (i = 0; i < 8; i++) {
1457         r[i] = (s->B(i) & 0x80) ? 0 : (v->B(s->B(i) & 7));
1458     }
1459     for (i = 0; i < 8; i++) {
1460         d->B(i) = r[i];
1461     }
1462 #else
1463     uint8_t r[8 << SHIFT];
1464 
1465     for (i = 0; i < 8 << SHIFT; i++) {
1466         int j = i & ~0xf;
1467         r[i] = (s->B(i) & 0x80) ? 0 : v->B(j | (s->B(i) & 0xf));
1468     }
1469     for (i = 0; i < 8 << SHIFT; i++) {
1470         d->B(i) = r[i];
1471     }
1472 #endif
1473 }
1474 
1475 #define SSE_HELPER_HW(name, F)  \
1476 void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s) \
1477 {                                                          \
1478     uint16_t r[4 << SHIFT];                                \
1479     int i, j, k;                                           \
1480     for (k = 0; k < 4 << SHIFT; k += LANE_WIDTH / 2) {     \
1481         for (i = j = 0; j < LANE_WIDTH / 2; i++, j += 2) { \
1482             r[i + k] = F(v->W(j + k), v->W(j + k + 1));    \
1483         }                                                  \
1484         for (j = 0; j < LANE_WIDTH / 2; i++, j += 2) {     \
1485             r[i + k] = F(s->W(j + k), s->W(j + k + 1));    \
1486         }                                                  \
1487     }                                                      \
1488     for (i = 0; i < 4 << SHIFT; i++) {                     \
1489         d->W(i) = r[i];                                    \
1490     }                                                      \
1491 }
1492 
1493 #define SSE_HELPER_HL(name, F)  \
1494 void glue(helper_ ## name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s) \
1495 {                                                          \
1496     uint32_t r[2 << SHIFT];                                \
1497     int i, j, k;                                           \
1498     for (k = 0; k < 2 << SHIFT; k += LANE_WIDTH / 4) {     \
1499         for (i = j = 0; j < LANE_WIDTH / 4; i++, j += 2) { \
1500             r[i + k] = F(v->L(j + k), v->L(j + k + 1));    \
1501         }                                                  \
1502         for (j = 0; j < LANE_WIDTH / 4; i++, j += 2) {     \
1503             r[i + k] = F(s->L(j + k), s->L(j + k + 1));    \
1504         }                                                  \
1505     }                                                      \
1506     for (i = 0; i < 2 << SHIFT; i++) {                     \
1507         d->L(i) = r[i];                                    \
1508     }                                                      \
1509 }
1510 
1511 SSE_HELPER_HW(phaddw, FADD)
1512 SSE_HELPER_HW(phsubw, FSUB)
1513 SSE_HELPER_HW(phaddsw, FADDSW)
1514 SSE_HELPER_HW(phsubsw, FSUBSW)
1515 SSE_HELPER_HL(phaddd, FADD)
1516 SSE_HELPER_HL(phsubd, FSUB)
1517 
1518 #undef SSE_HELPER_HW
1519 #undef SSE_HELPER_HL
1520 
1521 void glue(helper_pmaddubsw, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
1522 {
1523     int i;
1524     for (i = 0; i < 4 << SHIFT; i++) {
1525         d->W(i) = satsw((int8_t)s->B(i * 2) * (uint8_t)v->B(i * 2) +
1526                         (int8_t)s->B(i * 2 + 1) * (uint8_t)v->B(i * 2 + 1));
1527     }
1528 }
1529 
1530 #define FMULHRSW(d, s) (((int16_t) d * (int16_t)s + 0x4000) >> 15)
1531 SSE_HELPER_W(helper_pmulhrsw, FMULHRSW)
1532 
1533 #define FSIGNB(d, s) (s <= INT8_MAX  ? s ? d : 0 : -(int8_t)d)
1534 #define FSIGNW(d, s) (s <= INT16_MAX ? s ? d : 0 : -(int16_t)d)
1535 #define FSIGNL(d, s) (s <= INT32_MAX ? s ? d : 0 : -(int32_t)d)
1536 SSE_HELPER_B(helper_psignb, FSIGNB)
1537 SSE_HELPER_W(helper_psignw, FSIGNW)
1538 SSE_HELPER_L(helper_psignd, FSIGNL)
1539 
1540 void glue(helper_palignr, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
1541                                   uint32_t imm)
1542 {
1543     int i;
1544 
1545     /* XXX could be checked during translation */
1546     if (imm >= (SHIFT ? 32 : 16)) {
1547         for (i = 0; i < (1 << SHIFT); i++) {
1548             d->Q(i) = 0;
1549         }
1550     } else {
1551         int shift = imm * 8;
1552 #define SHR(v, i) (i < 64 && i > -64 ? i > 0 ? v >> (i) : (v << -(i)) : 0)
1553 #if SHIFT == 0
1554         d->Q(0) = SHR(s->Q(0), shift - 0) |
1555             SHR(v->Q(0), shift -  64);
1556 #else
1557         for (i = 0; i < (1 << SHIFT); i += 2) {
1558             uint64_t r0, r1;
1559 
1560             r0 = SHR(s->Q(i), shift - 0) |
1561                  SHR(s->Q(i + 1), shift -  64) |
1562                  SHR(v->Q(i), shift - 128) |
1563                  SHR(v->Q(i + 1), shift - 192);
1564             r1 = SHR(s->Q(i), shift + 64) |
1565                  SHR(s->Q(i + 1), shift -   0) |
1566                  SHR(v->Q(i), shift -  64) |
1567                  SHR(v->Q(i + 1), shift - 128);
1568             d->Q(i) = r0;
1569             d->Q(i + 1) = r1;
1570         }
1571 #endif
1572 #undef SHR
1573     }
1574 }
1575 
1576 #if SHIFT >= 1
1577 
1578 #define SSE_HELPER_V(name, elem, num, F)                                \
1579     void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,   \
1580                             Reg *m)                                     \
1581     {                                                                   \
1582         int i;                                                          \
1583         for (i = 0; i < num; i++) {                                     \
1584             d->elem(i) = F(v->elem(i), s->elem(i), m->elem(i));         \
1585         }                                                               \
1586     }
1587 
1588 #define SSE_HELPER_I(name, elem, num, F)                                \
1589     void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,   \
1590                             uint32_t imm)                               \
1591     {                                                                   \
1592         int i;                                                          \
1593         for (i = 0; i < num; i++) {                                     \
1594             int j = i & 7;                                              \
1595             d->elem(i) = F(v->elem(i), s->elem(i), (imm >> j) & 1);     \
1596         }                                                               \
1597     }
1598 
1599 /* SSE4.1 op helpers */
1600 #define FBLENDVB(v, s, m) ((m & 0x80) ? s : v)
1601 #define FBLENDVPS(v, s, m) ((m & 0x80000000) ? s : v)
1602 #define FBLENDVPD(v, s, m) ((m & 0x8000000000000000LL) ? s : v)
1603 SSE_HELPER_V(helper_pblendvb, B, 8 << SHIFT, FBLENDVB)
1604 SSE_HELPER_V(helper_blendvps, L, 2 << SHIFT, FBLENDVPS)
1605 SSE_HELPER_V(helper_blendvpd, Q, 1 << SHIFT, FBLENDVPD)
1606 
1607 void glue(helper_ptest, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
1608 {
1609     uint64_t zf = 0, cf = 0;
1610     int i;
1611 
1612     for (i = 0; i < 1 << SHIFT; i++) {
1613         zf |= (s->Q(i) &  d->Q(i));
1614         cf |= (s->Q(i) & ~d->Q(i));
1615     }
1616     CC_SRC = (zf ? 0 : CC_Z) | (cf ? 0 : CC_C);
1617     CC_OP = CC_OP_EFLAGS;
1618 }
1619 
1620 #define FMOVSLDUP(i) s->L((i) & ~1)
1621 #define FMOVSHDUP(i) s->L((i) | 1)
1622 #define FMOVDLDUP(i) s->Q((i) & ~1)
1623 
1624 #define SSE_HELPER_F(name, elem, num, F)                        \
1625     void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)   \
1626     {                                                           \
1627         int n = num;                                            \
1628         for (int i = n; --i >= 0; ) {                           \
1629             d->elem(i) = F(i);                                  \
1630         }                                                       \
1631     }
1632 
1633 #if SHIFT > 0
1634 SSE_HELPER_F(helper_pmovsxbw, W, 4 << SHIFT, (int8_t) s->B)
1635 SSE_HELPER_F(helper_pmovsxbd, L, 2 << SHIFT, (int8_t) s->B)
1636 SSE_HELPER_F(helper_pmovsxbq, Q, 1 << SHIFT, (int8_t) s->B)
1637 SSE_HELPER_F(helper_pmovsxwd, L, 2 << SHIFT, (int16_t) s->W)
1638 SSE_HELPER_F(helper_pmovsxwq, Q, 1 << SHIFT, (int16_t) s->W)
1639 SSE_HELPER_F(helper_pmovsxdq, Q, 1 << SHIFT, (int32_t) s->L)
1640 SSE_HELPER_F(helper_pmovzxbw, W, 4 << SHIFT, s->B)
1641 SSE_HELPER_F(helper_pmovzxbd, L, 2 << SHIFT, s->B)
1642 SSE_HELPER_F(helper_pmovzxbq, Q, 1 << SHIFT, s->B)
1643 SSE_HELPER_F(helper_pmovzxwd, L, 2 << SHIFT, s->W)
1644 SSE_HELPER_F(helper_pmovzxwq, Q, 1 << SHIFT, s->W)
1645 SSE_HELPER_F(helper_pmovzxdq, Q, 1 << SHIFT, s->L)
1646 SSE_HELPER_F(helper_pmovsldup, L, 2 << SHIFT, FMOVSLDUP)
1647 SSE_HELPER_F(helper_pmovshdup, L, 2 << SHIFT, FMOVSHDUP)
1648 SSE_HELPER_F(helper_pmovdldup, Q, 1 << SHIFT, FMOVDLDUP)
1649 #endif
1650 
1651 void glue(helper_pmuldq, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
1652 {
1653     int i;
1654 
1655     for (i = 0; i < 1 << SHIFT; i++) {
1656         d->Q(i) = (int64_t)(int32_t) v->L(2 * i) * (int32_t) s->L(2 * i);
1657     }
1658 }
1659 
1660 void glue(helper_packusdw, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
1661 {
1662     uint16_t r[8];
1663     int i, j, k;
1664 
1665     for (i = 0, j = 0; i <= 2 << SHIFT; i += 8, j += 4) {
1666         r[0] = satuw(v->L(j));
1667         r[1] = satuw(v->L(j + 1));
1668         r[2] = satuw(v->L(j + 2));
1669         r[3] = satuw(v->L(j + 3));
1670         r[4] = satuw(s->L(j));
1671         r[5] = satuw(s->L(j + 1));
1672         r[6] = satuw(s->L(j + 2));
1673         r[7] = satuw(s->L(j + 3));
1674         for (k = 0; k < 8; k++) {
1675             d->W(i + k) = r[k];
1676         }
1677     }
1678 }
1679 
1680 #if SHIFT == 1
1681 void glue(helper_phminposuw, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
1682 {
1683     int idx = 0;
1684 
1685     if (s->W(1) < s->W(idx)) {
1686         idx = 1;
1687     }
1688     if (s->W(2) < s->W(idx)) {
1689         idx = 2;
1690     }
1691     if (s->W(3) < s->W(idx)) {
1692         idx = 3;
1693     }
1694     if (s->W(4) < s->W(idx)) {
1695         idx = 4;
1696     }
1697     if (s->W(5) < s->W(idx)) {
1698         idx = 5;
1699     }
1700     if (s->W(6) < s->W(idx)) {
1701         idx = 6;
1702     }
1703     if (s->W(7) < s->W(idx)) {
1704         idx = 7;
1705     }
1706 
1707     d->W(0) = s->W(idx);
1708     d->W(1) = idx;
1709     d->L(1) = 0;
1710     d->Q(1) = 0;
1711 }
1712 #endif
1713 
1714 void glue(helper_roundps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
1715                                   uint32_t mode)
1716 {
1717     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
1718     signed char prev_rounding_mode;
1719     int i;
1720 
1721     prev_rounding_mode = env->sse_status.float_rounding_mode;
1722     if (!(mode & (1 << 2))) {
1723         set_x86_rounding_mode(mode & 3, &env->sse_status);
1724     }
1725 
1726     for (i = 0; i < 2 << SHIFT; i++) {
1727         d->ZMM_S(i) = float32_round_to_int(s->ZMM_S(i), &env->sse_status);
1728     }
1729 
1730     if (mode & (1 << 3) && !(old_flags & float_flag_inexact)) {
1731         set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1732                                   ~float_flag_inexact,
1733                                   &env->sse_status);
1734     }
1735     env->sse_status.float_rounding_mode = prev_rounding_mode;
1736 }
1737 
1738 void glue(helper_roundpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
1739                                   uint32_t mode)
1740 {
1741     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
1742     signed char prev_rounding_mode;
1743     int i;
1744 
1745     prev_rounding_mode = env->sse_status.float_rounding_mode;
1746     if (!(mode & (1 << 2))) {
1747         set_x86_rounding_mode(mode & 3, &env->sse_status);
1748     }
1749 
1750     for (i = 0; i < 1 << SHIFT; i++) {
1751         d->ZMM_D(i) = float64_round_to_int(s->ZMM_D(i), &env->sse_status);
1752     }
1753 
1754     if (mode & (1 << 3) && !(old_flags & float_flag_inexact)) {
1755         set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1756                                   ~float_flag_inexact,
1757                                   &env->sse_status);
1758     }
1759     env->sse_status.float_rounding_mode = prev_rounding_mode;
1760 }
1761 
1762 #if SHIFT == 1
1763 void glue(helper_roundss, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
1764                                   uint32_t mode)
1765 {
1766     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
1767     signed char prev_rounding_mode;
1768     int i;
1769 
1770     prev_rounding_mode = env->sse_status.float_rounding_mode;
1771     if (!(mode & (1 << 2))) {
1772         set_x86_rounding_mode(mode & 3, &env->sse_status);
1773     }
1774 
1775     d->ZMM_S(0) = float32_round_to_int(s->ZMM_S(0), &env->sse_status);
1776     for (i = 1; i < 2 << SHIFT; i++) {
1777         d->ZMM_L(i) = v->ZMM_L(i);
1778     }
1779 
1780     if (mode & (1 << 3) && !(old_flags & float_flag_inexact)) {
1781         set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1782                                   ~float_flag_inexact,
1783                                   &env->sse_status);
1784     }
1785     env->sse_status.float_rounding_mode = prev_rounding_mode;
1786 }
1787 
1788 void glue(helper_roundsd, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
1789                                   uint32_t mode)
1790 {
1791     uint8_t old_flags = get_float_exception_flags(&env->sse_status);
1792     signed char prev_rounding_mode;
1793     int i;
1794 
1795     prev_rounding_mode = env->sse_status.float_rounding_mode;
1796     if (!(mode & (1 << 2))) {
1797         set_x86_rounding_mode(mode & 3, &env->sse_status);
1798     }
1799 
1800     d->ZMM_D(0) = float64_round_to_int(s->ZMM_D(0), &env->sse_status);
1801     for (i = 1; i < 1 << SHIFT; i++) {
1802         d->ZMM_Q(i) = v->ZMM_Q(i);
1803     }
1804 
1805     if (mode & (1 << 3) && !(old_flags & float_flag_inexact)) {
1806         set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
1807                                   ~float_flag_inexact,
1808                                   &env->sse_status);
1809     }
1810     env->sse_status.float_rounding_mode = prev_rounding_mode;
1811 }
1812 #endif
1813 
1814 #define FBLENDP(v, s, m) (m ? s : v)
1815 SSE_HELPER_I(helper_blendps, L, 2 << SHIFT, FBLENDP)
1816 SSE_HELPER_I(helper_blendpd, Q, 1 << SHIFT, FBLENDP)
1817 SSE_HELPER_I(helper_pblendw, W, 4 << SHIFT, FBLENDP)
1818 
1819 void glue(helper_dpps, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
1820                                uint32_t mask)
1821 {
1822     float32 prod1, prod2, temp2, temp3, temp4;
1823     int i;
1824 
1825     for (i = 0; i < 2 << SHIFT; i += 4) {
1826         /*
1827          * We must evaluate (A+B)+(C+D), not ((A+B)+C)+D
1828          * to correctly round the intermediate results
1829          */
1830         if (mask & (1 << 4)) {
1831             prod1 = float32_mul(v->ZMM_S(i), s->ZMM_S(i), &env->sse_status);
1832         } else {
1833             prod1 = float32_zero;
1834         }
1835         if (mask & (1 << 5)) {
1836             prod2 = float32_mul(v->ZMM_S(i+1), s->ZMM_S(i+1), &env->sse_status);
1837         } else {
1838             prod2 = float32_zero;
1839         }
1840         temp2 = float32_add(prod1, prod2, &env->sse_status);
1841         if (mask & (1 << 6)) {
1842             prod1 = float32_mul(v->ZMM_S(i+2), s->ZMM_S(i+2), &env->sse_status);
1843         } else {
1844             prod1 = float32_zero;
1845         }
1846         if (mask & (1 << 7)) {
1847             prod2 = float32_mul(v->ZMM_S(i+3), s->ZMM_S(i+3), &env->sse_status);
1848         } else {
1849             prod2 = float32_zero;
1850         }
1851         temp3 = float32_add(prod1, prod2, &env->sse_status);
1852         temp4 = float32_add(temp2, temp3, &env->sse_status);
1853 
1854         d->ZMM_S(i) = (mask & (1 << 0)) ? temp4 : float32_zero;
1855         d->ZMM_S(i+1) = (mask & (1 << 1)) ? temp4 : float32_zero;
1856         d->ZMM_S(i+2) = (mask & (1 << 2)) ? temp4 : float32_zero;
1857         d->ZMM_S(i+3) = (mask & (1 << 3)) ? temp4 : float32_zero;
1858     }
1859 }
1860 
1861 #if SHIFT == 1
1862 /* Oddly, there is no ymm version of dppd */
1863 void glue(helper_dppd, SUFFIX)(CPUX86State *env,
1864                                Reg *d, Reg *v, Reg *s, uint32_t mask)
1865 {
1866     float64 prod1, prod2, temp2;
1867 
1868     if (mask & (1 << 4)) {
1869         prod1 = float64_mul(v->ZMM_D(0), s->ZMM_D(0), &env->sse_status);
1870     } else {
1871         prod1 = float64_zero;
1872     }
1873     if (mask & (1 << 5)) {
1874         prod2 = float64_mul(v->ZMM_D(1), s->ZMM_D(1), &env->sse_status);
1875     } else {
1876         prod2 = float64_zero;
1877     }
1878     temp2 = float64_add(prod1, prod2, &env->sse_status);
1879     d->ZMM_D(0) = (mask & (1 << 0)) ? temp2 : float64_zero;
1880     d->ZMM_D(1) = (mask & (1 << 1)) ? temp2 : float64_zero;
1881 }
1882 #endif
1883 
1884 void glue(helper_mpsadbw, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
1885                                   uint32_t offset)
1886 {
1887     int i, j;
1888     uint16_t r[8];
1889 
1890     for (j = 0; j < 4 << SHIFT; ) {
1891         int s0 = (j * 2) + ((offset & 3) << 2);
1892         int d0 = (j * 2) + ((offset & 4) << 0);
1893         for (i = 0; i < LANE_WIDTH / 2; i++, d0++) {
1894             r[i] = 0;
1895             r[i] += abs1(v->B(d0 + 0) - s->B(s0 + 0));
1896             r[i] += abs1(v->B(d0 + 1) - s->B(s0 + 1));
1897             r[i] += abs1(v->B(d0 + 2) - s->B(s0 + 2));
1898             r[i] += abs1(v->B(d0 + 3) - s->B(s0 + 3));
1899         }
1900         for (i = 0; i < LANE_WIDTH / 2; i++, j++) {
1901             d->W(j) = r[i];
1902         }
1903         offset >>= 3;
1904     }
1905 }
1906 
1907 /* SSE4.2 op helpers */
1908 #if SHIFT == 1
1909 static inline int pcmp_elen(CPUX86State *env, int reg, uint32_t ctrl)
1910 {
1911     target_long val, limit;
1912 
1913     /* Presence of REX.W is indicated by a bit higher than 7 set */
1914     if (ctrl >> 8) {
1915         val = (target_long)env->regs[reg];
1916     } else {
1917         val = (int32_t)env->regs[reg];
1918     }
1919     if (ctrl & 1) {
1920         limit = 8;
1921     } else {
1922         limit = 16;
1923     }
1924     if ((val > limit) || (val < -limit)) {
1925         return limit;
1926     }
1927     return abs1(val);
1928 }
1929 
1930 static inline int pcmp_ilen(Reg *r, uint8_t ctrl)
1931 {
1932     int val = 0;
1933 
1934     if (ctrl & 1) {
1935         while (val < 8 && r->W(val)) {
1936             val++;
1937         }
1938     } else {
1939         while (val < 16 && r->B(val)) {
1940             val++;
1941         }
1942     }
1943 
1944     return val;
1945 }
1946 
1947 static inline int pcmp_val(Reg *r, uint8_t ctrl, int i)
1948 {
1949     switch ((ctrl >> 0) & 3) {
1950     case 0:
1951         return r->B(i);
1952     case 1:
1953         return r->W(i);
1954     case 2:
1955         return (int8_t)r->B(i);
1956     case 3:
1957     default:
1958         return (int16_t)r->W(i);
1959     }
1960 }
1961 
1962 static inline unsigned pcmpxstrx(CPUX86State *env, Reg *d, Reg *s,
1963                                  uint8_t ctrl, int valids, int validd)
1964 {
1965     unsigned int res = 0;
1966     int v;
1967     int j, i;
1968     int upper = (ctrl & 1) ? 7 : 15;
1969 
1970     valids--;
1971     validd--;
1972 
1973     CC_SRC = (valids < upper ? CC_Z : 0) | (validd < upper ? CC_S : 0);
1974     CC_OP = CC_OP_EFLAGS;
1975 
1976     switch ((ctrl >> 2) & 3) {
1977     case 0:
1978         for (j = valids; j >= 0; j--) {
1979             res <<= 1;
1980             v = pcmp_val(s, ctrl, j);
1981             for (i = validd; i >= 0; i--) {
1982                 res |= (v == pcmp_val(d, ctrl, i));
1983             }
1984         }
1985         break;
1986     case 1:
1987         for (j = valids; j >= 0; j--) {
1988             res <<= 1;
1989             v = pcmp_val(s, ctrl, j);
1990             for (i = ((validd - 1) | 1); i >= 0; i -= 2) {
1991                 res |= (pcmp_val(d, ctrl, i - 0) >= v &&
1992                         pcmp_val(d, ctrl, i - 1) <= v);
1993             }
1994         }
1995         break;
1996     case 2:
1997         res = (1 << (upper - MAX(valids, validd))) - 1;
1998         res <<= MAX(valids, validd) - MIN(valids, validd);
1999         for (i = MIN(valids, validd); i >= 0; i--) {
2000             res <<= 1;
2001             v = pcmp_val(s, ctrl, i);
2002             res |= (v == pcmp_val(d, ctrl, i));
2003         }
2004         break;
2005     case 3:
2006         if (validd == -1) {
2007             res = (2 << upper) - 1;
2008             break;
2009         }
2010         for (j = valids == upper ? valids : valids - validd; j >= 0; j--) {
2011             res <<= 1;
2012             v = 1;
2013             for (i = MIN(valids - j, validd); i >= 0; i--) {
2014                 v &= (pcmp_val(s, ctrl, i + j) == pcmp_val(d, ctrl, i));
2015             }
2016             res |= v;
2017         }
2018         break;
2019     }
2020 
2021     switch ((ctrl >> 4) & 3) {
2022     case 1:
2023         res ^= (2 << upper) - 1;
2024         break;
2025     case 3:
2026         res ^= (1 << (valids + 1)) - 1;
2027         break;
2028     }
2029 
2030     if (res) {
2031         CC_SRC |= CC_C;
2032     }
2033     if (res & 1) {
2034         CC_SRC |= CC_O;
2035     }
2036 
2037     return res;
2038 }
2039 
2040 void glue(helper_pcmpestri, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2041                                     uint32_t ctrl)
2042 {
2043     unsigned int res = pcmpxstrx(env, d, s, ctrl,
2044                                  pcmp_elen(env, R_EDX, ctrl),
2045                                  pcmp_elen(env, R_EAX, ctrl));
2046 
2047     if (res) {
2048         env->regs[R_ECX] = (ctrl & (1 << 6)) ? 31 - clz32(res) : ctz32(res);
2049     } else {
2050         env->regs[R_ECX] = 16 >> (ctrl & (1 << 0));
2051     }
2052 }
2053 
2054 void glue(helper_pcmpestrm, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2055                                     uint32_t ctrl)
2056 {
2057     int i;
2058     unsigned int res = pcmpxstrx(env, d, s, ctrl,
2059                                  pcmp_elen(env, R_EDX, ctrl),
2060                                  pcmp_elen(env, R_EAX, ctrl));
2061 
2062     if ((ctrl >> 6) & 1) {
2063         if (ctrl & 1) {
2064             for (i = 0; i < 8; i++, res >>= 1) {
2065                 env->xmm_regs[0].W(i) = (res & 1) ? ~0 : 0;
2066             }
2067         } else {
2068             for (i = 0; i < 16; i++, res >>= 1) {
2069                 env->xmm_regs[0].B(i) = (res & 1) ? ~0 : 0;
2070             }
2071         }
2072     } else {
2073         env->xmm_regs[0].Q(1) = 0;
2074         env->xmm_regs[0].Q(0) = res;
2075     }
2076 }
2077 
2078 void glue(helper_pcmpistri, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2079                                     uint32_t ctrl)
2080 {
2081     unsigned int res = pcmpxstrx(env, d, s, ctrl,
2082                                  pcmp_ilen(s, ctrl),
2083                                  pcmp_ilen(d, ctrl));
2084 
2085     if (res) {
2086         env->regs[R_ECX] = (ctrl & (1 << 6)) ? 31 - clz32(res) : ctz32(res);
2087     } else {
2088         env->regs[R_ECX] = 16 >> (ctrl & (1 << 0));
2089     }
2090 }
2091 
2092 void glue(helper_pcmpistrm, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2093                                     uint32_t ctrl)
2094 {
2095     int i;
2096     unsigned int res = pcmpxstrx(env, d, s, ctrl,
2097                                  pcmp_ilen(s, ctrl),
2098                                  pcmp_ilen(d, ctrl));
2099 
2100     if ((ctrl >> 6) & 1) {
2101         if (ctrl & 1) {
2102             for (i = 0; i < 8; i++, res >>= 1) {
2103                 env->xmm_regs[0].W(i) = (res & 1) ? ~0 : 0;
2104             }
2105         } else {
2106             for (i = 0; i < 16; i++, res >>= 1) {
2107                 env->xmm_regs[0].B(i) = (res & 1) ? ~0 : 0;
2108             }
2109         }
2110     } else {
2111         env->xmm_regs[0].Q(1) = 0;
2112         env->xmm_regs[0].Q(0) = res;
2113     }
2114 }
2115 
2116 #define CRCPOLY        0x1edc6f41
2117 #define CRCPOLY_BITREV 0x82f63b78
2118 target_ulong helper_crc32(uint32_t crc1, target_ulong msg, uint32_t len)
2119 {
2120     target_ulong crc = (msg & ((target_ulong) -1 >>
2121                                (TARGET_LONG_BITS - len))) ^ crc1;
2122 
2123     while (len--) {
2124         crc = (crc >> 1) ^ ((crc & 1) ? CRCPOLY_BITREV : 0);
2125     }
2126 
2127     return crc;
2128 }
2129 
2130 #endif
2131 
2132 void glue(helper_pclmulqdq, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
2133                                     uint32_t ctrl)
2134 {
2135     int a_idx = (ctrl & 1) != 0;
2136     int b_idx = (ctrl & 16) != 0;
2137 
2138     for (int i = 0; i < SHIFT; i++) {
2139         uint64_t a = v->Q(2 * i + a_idx);
2140         uint64_t b = s->Q(2 * i + b_idx);
2141         Int128 *r = (Int128 *)&d->ZMM_X(i);
2142 
2143         *r = clmul_64(a, b);
2144     }
2145 }
2146 
2147 void glue(helper_aesdec, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2148 {
2149     for (int i = 0; i < SHIFT; i++) {
2150         AESState *ad = (AESState *)&d->ZMM_X(i);
2151         AESState *st = (AESState *)&v->ZMM_X(i);
2152         AESState *rk = (AESState *)&s->ZMM_X(i);
2153 
2154         aesdec_ISB_ISR_IMC_AK(ad, st, rk, false);
2155     }
2156 }
2157 
2158 void glue(helper_aesdeclast, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2159 {
2160     for (int i = 0; i < SHIFT; i++) {
2161         AESState *ad = (AESState *)&d->ZMM_X(i);
2162         AESState *st = (AESState *)&v->ZMM_X(i);
2163         AESState *rk = (AESState *)&s->ZMM_X(i);
2164 
2165         aesdec_ISB_ISR_AK(ad, st, rk, false);
2166     }
2167 }
2168 
2169 void glue(helper_aesenc, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2170 {
2171     for (int i = 0; i < SHIFT; i++) {
2172         AESState *ad = (AESState *)&d->ZMM_X(i);
2173         AESState *st = (AESState *)&v->ZMM_X(i);
2174         AESState *rk = (AESState *)&s->ZMM_X(i);
2175 
2176         aesenc_SB_SR_MC_AK(ad, st, rk, false);
2177     }
2178 }
2179 
2180 void glue(helper_aesenclast, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2181 {
2182     for (int i = 0; i < SHIFT; i++) {
2183         AESState *ad = (AESState *)&d->ZMM_X(i);
2184         AESState *st = (AESState *)&v->ZMM_X(i);
2185         AESState *rk = (AESState *)&s->ZMM_X(i);
2186 
2187         aesenc_SB_SR_AK(ad, st, rk, false);
2188     }
2189 }
2190 
2191 #if SHIFT == 1
2192 void glue(helper_aesimc, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
2193 {
2194     AESState *ad = (AESState *)&d->ZMM_X(0);
2195     AESState *st = (AESState *)&s->ZMM_X(0);
2196 
2197     aesdec_IMC(ad, st, false);
2198 }
2199 
2200 void glue(helper_aeskeygenassist, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
2201                                           uint32_t ctrl)
2202 {
2203     int i;
2204     Reg tmp = *s;
2205 
2206     for (i = 0 ; i < 4 ; i++) {
2207         d->B(i) = AES_sbox[tmp.B(i + 4)];
2208         d->B(i + 8) = AES_sbox[tmp.B(i + 12)];
2209     }
2210     d->L(1) = (d->L(0) << 24 | d->L(0) >> 8) ^ ctrl;
2211     d->L(3) = (d->L(2) << 24 | d->L(2) >> 8) ^ ctrl;
2212 }
2213 #endif
2214 #endif
2215 
2216 #if SHIFT >= 1
2217 void glue(helper_vpermilpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2218 {
2219     uint64_t r0, r1;
2220     int i;
2221 
2222     for (i = 0; i < 1 << SHIFT; i += 2) {
2223         r0 = v->Q(i + ((s->Q(i) >> 1) & 1));
2224         r1 = v->Q(i + ((s->Q(i+1) >> 1) & 1));
2225         d->Q(i) = r0;
2226         d->Q(i+1) = r1;
2227     }
2228 }
2229 
2230 void glue(helper_vpermilps, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2231 {
2232     uint32_t r0, r1, r2, r3;
2233     int i;
2234 
2235     for (i = 0; i < 2 << SHIFT; i += 4) {
2236         r0 = v->L(i + (s->L(i) & 3));
2237         r1 = v->L(i + (s->L(i+1) & 3));
2238         r2 = v->L(i + (s->L(i+2) & 3));
2239         r3 = v->L(i + (s->L(i+3) & 3));
2240         d->L(i) = r0;
2241         d->L(i+1) = r1;
2242         d->L(i+2) = r2;
2243         d->L(i+3) = r3;
2244     }
2245 }
2246 
2247 void glue(helper_vpermilpd_imm, SUFFIX)(Reg *d, Reg *s, uint32_t order)
2248 {
2249     uint64_t r0, r1;
2250     int i;
2251 
2252     for (i = 0; i < 1 << SHIFT; i += 2) {
2253         r0 = s->Q(i + ((order >> 0) & 1));
2254         r1 = s->Q(i + ((order >> 1) & 1));
2255         d->Q(i) = r0;
2256         d->Q(i+1) = r1;
2257 
2258         order >>= 2;
2259     }
2260 }
2261 
2262 void glue(helper_vpermilps_imm, SUFFIX)(Reg *d, Reg *s, uint32_t order)
2263 {
2264     uint32_t r0, r1, r2, r3;
2265     int i;
2266 
2267     for (i = 0; i < 2 << SHIFT; i += 4) {
2268         r0 = s->L(i + ((order >> 0) & 3));
2269         r1 = s->L(i + ((order >> 2) & 3));
2270         r2 = s->L(i + ((order >> 4) & 3));
2271         r3 = s->L(i + ((order >> 6) & 3));
2272         d->L(i) = r0;
2273         d->L(i+1) = r1;
2274         d->L(i+2) = r2;
2275         d->L(i+3) = r3;
2276     }
2277 }
2278 
2279 #if SHIFT == 1
2280 #define FPSRLVD(x, c) (c < 32 ? ((x) >> c) : 0)
2281 #define FPSRLVQ(x, c) (c < 64 ? ((x) >> c) : 0)
2282 #define FPSRAVD(x, c) ((int32_t)(x) >> (c < 32 ? c : 31))
2283 #define FPSRAVQ(x, c) ((int64_t)(x) >> (c < 64 ? c : 63))
2284 #define FPSLLVD(x, c) (c < 32 ? ((x) << c) : 0)
2285 #define FPSLLVQ(x, c) (c < 64 ? ((x) << c) : 0)
2286 #endif
2287 
2288 SSE_HELPER_L(helper_vpsrlvd, FPSRLVD)
2289 SSE_HELPER_L(helper_vpsravd, FPSRAVD)
2290 SSE_HELPER_L(helper_vpsllvd, FPSLLVD)
2291 
2292 SSE_HELPER_Q(helper_vpsrlvq, FPSRLVQ)
2293 SSE_HELPER_Q(helper_vpsravq, FPSRAVQ)
2294 SSE_HELPER_Q(helper_vpsllvq, FPSLLVQ)
2295 
2296 void glue(helper_vtestps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
2297 {
2298     uint32_t zf = 0, cf = 0;
2299     int i;
2300 
2301     for (i = 0; i < 2 << SHIFT; i++) {
2302         zf |= (s->L(i) &  d->L(i));
2303         cf |= (s->L(i) & ~d->L(i));
2304     }
2305     CC_SRC = ((zf >> 31) ? 0 : CC_Z) | ((cf >> 31) ? 0 : CC_C);
2306     CC_OP = CC_OP_EFLAGS;
2307 }
2308 
2309 void glue(helper_vtestpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
2310 {
2311     uint64_t zf = 0, cf = 0;
2312     int i;
2313 
2314     for (i = 0; i < 1 << SHIFT; i++) {
2315         zf |= (s->Q(i) &  d->Q(i));
2316         cf |= (s->Q(i) & ~d->Q(i));
2317     }
2318     CC_SRC = ((zf >> 63) ? 0 : CC_Z) | ((cf >> 63) ? 0 : CC_C);
2319     CC_OP = CC_OP_EFLAGS;
2320 }
2321 
2322 void glue(helper_vpmaskmovd_st, SUFFIX)(CPUX86State *env,
2323                                         Reg *v, Reg *s, target_ulong a0)
2324 {
2325     int i;
2326 
2327     for (i = 0; i < (2 << SHIFT); i++) {
2328         if (v->L(i) >> 31) {
2329             cpu_stl_data_ra(env, a0 + i * 4, s->L(i), GETPC());
2330         }
2331     }
2332 }
2333 
2334 void glue(helper_vpmaskmovq_st, SUFFIX)(CPUX86State *env,
2335                                         Reg *v, Reg *s, target_ulong a0)
2336 {
2337     int i;
2338 
2339     for (i = 0; i < (1 << SHIFT); i++) {
2340         if (v->Q(i) >> 63) {
2341             cpu_stq_data_ra(env, a0 + i * 8, s->Q(i), GETPC());
2342         }
2343     }
2344 }
2345 
2346 void glue(helper_vpmaskmovd, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2347 {
2348     int i;
2349 
2350     for (i = 0; i < (2 << SHIFT); i++) {
2351         d->L(i) = (v->L(i) >> 31) ? s->L(i) : 0;
2352     }
2353 }
2354 
2355 void glue(helper_vpmaskmovq, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s)
2356 {
2357     int i;
2358 
2359     for (i = 0; i < (1 << SHIFT); i++) {
2360         d->Q(i) = (v->Q(i) >> 63) ? s->Q(i) : 0;
2361     }
2362 }
2363 
2364 void glue(helper_vpgatherdd, SUFFIX)(CPUX86State *env,
2365         Reg *d, Reg *v, Reg *s, target_ulong a0, unsigned scale)
2366 {
2367     int i;
2368     for (i = 0; i < (2 << SHIFT); i++) {
2369         if (v->L(i) >> 31) {
2370             target_ulong addr = a0
2371                 + ((target_ulong)(int32_t)s->L(i) << scale);
2372             d->L(i) = cpu_ldl_data_ra(env, addr, GETPC());
2373         }
2374         v->L(i) = 0;
2375     }
2376 }
2377 
2378 void glue(helper_vpgatherdq, SUFFIX)(CPUX86State *env,
2379         Reg *d, Reg *v, Reg *s, target_ulong a0, unsigned scale)
2380 {
2381     int i;
2382     for (i = 0; i < (1 << SHIFT); i++) {
2383         if (v->Q(i) >> 63) {
2384             target_ulong addr = a0
2385                 + ((target_ulong)(int32_t)s->L(i) << scale);
2386             d->Q(i) = cpu_ldq_data_ra(env, addr, GETPC());
2387         }
2388         v->Q(i) = 0;
2389     }
2390 }
2391 
2392 void glue(helper_vpgatherqd, 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->L(i) >> 31) {
2398             target_ulong addr = a0
2399                 + ((target_ulong)(int64_t)s->Q(i) << scale);
2400             d->L(i) = cpu_ldl_data_ra(env, addr, GETPC());
2401         }
2402         v->L(i) = 0;
2403     }
2404     for (i /= 2; i < 1 << SHIFT; i++) {
2405         d->Q(i) = 0;
2406         v->Q(i) = 0;
2407     }
2408 }
2409 
2410 void glue(helper_vpgatherqq, SUFFIX)(CPUX86State *env,
2411         Reg *d, Reg *v, Reg *s, target_ulong a0, unsigned scale)
2412 {
2413     int i;
2414     for (i = 0; i < (1 << SHIFT); i++) {
2415         if (v->Q(i) >> 63) {
2416             target_ulong addr = a0
2417                 + ((target_ulong)(int64_t)s->Q(i) << scale);
2418             d->Q(i) = cpu_ldq_data_ra(env, addr, GETPC());
2419         }
2420         v->Q(i) = 0;
2421     }
2422 }
2423 #endif
2424 
2425 #if SHIFT >= 2
2426 void helper_vpermdq_ymm(Reg *d, Reg *v, Reg *s, uint32_t order)
2427 {
2428     uint64_t r0, r1, r2, r3;
2429 
2430     switch (order & 3) {
2431     case 0:
2432         r0 = v->Q(0);
2433         r1 = v->Q(1);
2434         break;
2435     case 1:
2436         r0 = v->Q(2);
2437         r1 = v->Q(3);
2438         break;
2439     case 2:
2440         r0 = s->Q(0);
2441         r1 = s->Q(1);
2442         break;
2443     case 3:
2444         r0 = s->Q(2);
2445         r1 = s->Q(3);
2446         break;
2447     default: /* default case added to help the compiler to avoid warnings */
2448         g_assert_not_reached();
2449     }
2450     switch ((order >> 4) & 3) {
2451     case 0:
2452         r2 = v->Q(0);
2453         r3 = v->Q(1);
2454         break;
2455     case 1:
2456         r2 = v->Q(2);
2457         r3 = v->Q(3);
2458         break;
2459     case 2:
2460         r2 = s->Q(0);
2461         r3 = s->Q(1);
2462         break;
2463     case 3:
2464         r2 = s->Q(2);
2465         r3 = s->Q(3);
2466         break;
2467     default: /* default case added to help the compiler to avoid warnings */
2468         g_assert_not_reached();
2469     }
2470     d->Q(0) = r0;
2471     d->Q(1) = r1;
2472     d->Q(2) = r2;
2473     d->Q(3) = r3;
2474     if (order & 0x8) {
2475         d->Q(0) = 0;
2476         d->Q(1) = 0;
2477     }
2478     if (order & 0x80) {
2479         d->Q(2) = 0;
2480         d->Q(3) = 0;
2481     }
2482 }
2483 
2484 void helper_vpermq_ymm(Reg *d, Reg *s, uint32_t order)
2485 {
2486     uint64_t r0, r1, r2, r3;
2487     r0 = s->Q(order & 3);
2488     r1 = s->Q((order >> 2) & 3);
2489     r2 = s->Q((order >> 4) & 3);
2490     r3 = s->Q((order >> 6) & 3);
2491     d->Q(0) = r0;
2492     d->Q(1) = r1;
2493     d->Q(2) = r2;
2494     d->Q(3) = r3;
2495 }
2496 
2497 void helper_vpermd_ymm(Reg *d, Reg *v, Reg *s)
2498 {
2499     uint32_t r[8];
2500     int i;
2501 
2502     for (i = 0; i < 8; i++) {
2503         r[i] = s->L(v->L(i) & 7);
2504     }
2505     for (i = 0; i < 8; i++) {
2506         d->L(i) = r[i];
2507     }
2508 }
2509 #endif
2510 
2511 /* FMA3 op helpers */
2512 #if SHIFT == 1
2513 #define SSE_HELPER_FMAS(name, elem, F)                                         \
2514     void name(CPUX86State *env, Reg *d, Reg *a, Reg *b, Reg *c, int flags)     \
2515     {                                                                          \
2516         d->elem(0) = F(a->elem(0), b->elem(0), c->elem(0), flags, &env->sse_status); \
2517     }
2518 #define SSE_HELPER_FMAP(name, elem, num, F)                                    \
2519     void glue(name, SUFFIX)(CPUX86State *env, Reg *d, Reg *a, Reg *b, Reg *c,  \
2520                             int flags, int flip)                               \
2521     {                                                                          \
2522         int i;                                                                 \
2523         for (i = 0; i < num; i++) {                                            \
2524             d->elem(i) = F(a->elem(i), b->elem(i), c->elem(i), flags, &env->sse_status); \
2525             flags ^= flip;                                                     \
2526         }                                                                      \
2527     }
2528 
2529 SSE_HELPER_FMAS(helper_fma4ss,  ZMM_S, float32_muladd)
2530 SSE_HELPER_FMAS(helper_fma4sd,  ZMM_D, float64_muladd)
2531 #endif
2532 
2533 #if SHIFT >= 1
2534 SSE_HELPER_FMAP(helper_fma4ps,  ZMM_S, 2 << SHIFT, float32_muladd)
2535 SSE_HELPER_FMAP(helper_fma4pd,  ZMM_D, 1 << SHIFT, float64_muladd)
2536 #endif
2537 
2538 #if SHIFT == 1
2539 #define SSE_HELPER_SHA1RNDS4(name, F, K) \
2540     void name(Reg *d, Reg *a, Reg *b)                                       \
2541     {                                                                       \
2542         uint32_t A, B, C, D, E, t, i;                                       \
2543                                                                             \
2544         A = a->L(3);                                                        \
2545         B = a->L(2);                                                        \
2546         C = a->L(1);                                                        \
2547         D = a->L(0);                                                        \
2548         E = 0;                                                              \
2549                                                                             \
2550         for (i = 0; i <= 3; i++) {                                          \
2551             t = F(B, C, D) + rol32(A, 5) + b->L(3 - i) + E + K;             \
2552             E = D;                                                          \
2553             D = C;                                                          \
2554             C = rol32(B, 30);                                               \
2555             B = A;                                                          \
2556             A = t;                                                          \
2557         }                                                                   \
2558                                                                             \
2559         d->L(3) = A;                                                        \
2560         d->L(2) = B;                                                        \
2561         d->L(1) = C;                                                        \
2562         d->L(0) = D;                                                        \
2563     }
2564 
2565 #define SHA1_F0(b, c, d) (((b) & (c)) ^ (~(b) & (d)))
2566 #define SHA1_F1(b, c, d) ((b) ^ (c) ^ (d))
2567 #define SHA1_F2(b, c, d) (((b) & (c)) ^ ((b) & (d)) ^ ((c) & (d)))
2568 
2569 SSE_HELPER_SHA1RNDS4(helper_sha1rnds4_f0, SHA1_F0, 0x5A827999)
2570 SSE_HELPER_SHA1RNDS4(helper_sha1rnds4_f1, SHA1_F1, 0x6ED9EBA1)
2571 SSE_HELPER_SHA1RNDS4(helper_sha1rnds4_f2, SHA1_F2, 0x8F1BBCDC)
2572 SSE_HELPER_SHA1RNDS4(helper_sha1rnds4_f3, SHA1_F1, 0xCA62C1D6)
2573 
2574 void helper_sha1nexte(Reg *d, Reg *a, Reg *b)
2575 {
2576     d->L(3) = b->L(3) + rol32(a->L(3), 30);
2577     d->L(2) = b->L(2);
2578     d->L(1) = b->L(1);
2579     d->L(0) = b->L(0);
2580 }
2581 
2582 void helper_sha1msg1(Reg *d, Reg *a, Reg *b)
2583 {
2584     /* These could be overwritten by the first two assignments, save them.  */
2585     uint32_t b3 = b->L(3);
2586     uint32_t b2 = b->L(2);
2587 
2588     d->L(3) = a->L(3) ^ a->L(1);
2589     d->L(2) = a->L(2) ^ a->L(0);
2590     d->L(1) = a->L(1) ^ b3;
2591     d->L(0) = a->L(0) ^ b2;
2592 }
2593 
2594 void helper_sha1msg2(Reg *d, Reg *a, Reg *b)
2595 {
2596     d->L(3) = rol32(a->L(3) ^ b->L(2), 1);
2597     d->L(2) = rol32(a->L(2) ^ b->L(1), 1);
2598     d->L(1) = rol32(a->L(1) ^ b->L(0), 1);
2599     d->L(0) = rol32(a->L(0) ^ d->L(3), 1);
2600 }
2601 
2602 #define SHA256_CH(e, f, g)  (((e) & (f)) ^ (~(e) & (g)))
2603 #define SHA256_MAJ(a, b, c) (((a) & (b)) ^ ((a) & (c)) ^ ((b) & (c)))
2604 
2605 #define SHA256_RNDS0(w) (ror32((w), 2) ^ ror32((w), 13) ^ ror32((w), 22))
2606 #define SHA256_RNDS1(w) (ror32((w), 6) ^ ror32((w), 11) ^ ror32((w), 25))
2607 #define SHA256_MSGS0(w) (ror32((w), 7) ^ ror32((w), 18) ^ ((w) >> 3))
2608 #define SHA256_MSGS1(w) (ror32((w), 17) ^ ror32((w), 19) ^ ((w) >> 10))
2609 
2610 void helper_sha256rnds2(Reg *d, Reg *a, Reg *b, uint32_t wk0, uint32_t wk1)
2611 {
2612     uint32_t t, AA, EE;
2613 
2614     uint32_t A = b->L(3);
2615     uint32_t B = b->L(2);
2616     uint32_t C = a->L(3);
2617     uint32_t D = a->L(2);
2618     uint32_t E = b->L(1);
2619     uint32_t F = b->L(0);
2620     uint32_t G = a->L(1);
2621     uint32_t H = a->L(0);
2622 
2623     /* Even round */
2624     t = SHA256_CH(E, F, G) + SHA256_RNDS1(E) + wk0 + H;
2625     AA = t + SHA256_MAJ(A, B, C) + SHA256_RNDS0(A);
2626     EE = t + D;
2627 
2628     /* These will be B and F at the end of the odd round */
2629     d->L(2) = AA;
2630     d->L(0) = EE;
2631 
2632     D = C, C = B, B = A, A = AA;
2633     H = G, G = F, F = E, E = EE;
2634 
2635     /* Odd round */
2636     t = SHA256_CH(E, F, G) + SHA256_RNDS1(E) + wk1 + H;
2637     AA = t + SHA256_MAJ(A, B, C) + SHA256_RNDS0(A);
2638     EE = t + D;
2639 
2640     d->L(3) = AA;
2641     d->L(1) = EE;
2642 }
2643 
2644 void helper_sha256msg1(Reg *d, Reg *a, Reg *b)
2645 {
2646     /* b->L(0) could be overwritten by the first assignment, save it.  */
2647     uint32_t b0 = b->L(0);
2648 
2649     d->L(0) = a->L(0) + SHA256_MSGS0(a->L(1));
2650     d->L(1) = a->L(1) + SHA256_MSGS0(a->L(2));
2651     d->L(2) = a->L(2) + SHA256_MSGS0(a->L(3));
2652     d->L(3) = a->L(3) + SHA256_MSGS0(b0);
2653 }
2654 
2655 void helper_sha256msg2(Reg *d, Reg *a, Reg *b)
2656 {
2657     /* Earlier assignments cannot overwrite any of the two operands.  */
2658     d->L(0) = a->L(0) + SHA256_MSGS1(b->L(2));
2659     d->L(1) = a->L(1) + SHA256_MSGS1(b->L(3));
2660     /* Yes, this reuses the previously computed values.  */
2661     d->L(2) = a->L(2) + SHA256_MSGS1(d->L(0));
2662     d->L(3) = a->L(3) + SHA256_MSGS1(d->L(1));
2663 }
2664 #endif
2665 
2666 #undef SSE_HELPER_S
2667 
2668 #undef LANE_WIDTH
2669 #undef SHIFT
2670 #undef XMM_ONLY
2671 #undef Reg
2672 #undef B
2673 #undef W
2674 #undef L
2675 #undef Q
2676 #undef SUFFIX
2677