Add libavcodec to compiler include flags in order to simplify header
[ffmpeg.git] / libavcodec / i386 / snowdsp_mmx.c
1 /*
2  * MMX and SSE2 optimized snow DSP utils
3  * Copyright (c) 2005-2006 Robert Edele <yartrebo@earthlink.net>
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg 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  * FFmpeg 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 FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21
22 #include "avcodec.h"
23 #include "snow.h"
24 #include "x86_cpu.h"
25
26 void ff_snow_horizontal_compose97i_sse2(DWTELEM *b, int width){
27     const int w2= (width+1)>>1;
28     // SSE2 code runs faster with pointers aligned on a 32-byte boundary.
29     DWTELEM temp_buf[(width>>1) + 4];
30     DWTELEM * const temp = temp_buf + 4 - (((int)temp_buf & 0xF) >> 2);
31     const int w_l= (width>>1);
32     const int w_r= w2 - 1;
33     int i;
34
35     { // Lift 0
36         DWTELEM * const ref = b + w2 - 1;
37         DWTELEM b_0 = b[0]; //By allowing the first entry in b[0] to be calculated twice
38         // (the first time erroneously), we allow the SSE2 code to run an extra pass.
39         // The savings in code and time are well worth having to store this value and
40         // calculate b[0] correctly afterwards.
41
42         i = 0;
43         asm volatile(
44             "pcmpeqd   %%xmm7, %%xmm7         \n\t"
45             "pslld        $31, %%xmm7         \n\t"
46             "psrld        $29, %%xmm7         \n\t"
47         ::);
48         for(; i<w_l-7; i+=8){
49             asm volatile(
50                 "movdqu   (%1), %%xmm1        \n\t"
51                 "movdqu 16(%1), %%xmm5        \n\t"
52                 "movdqu  4(%1), %%xmm2        \n\t"
53                 "movdqu 20(%1), %%xmm6        \n\t"
54                 "paddd  %%xmm1, %%xmm2        \n\t"
55                 "paddd  %%xmm5, %%xmm6        \n\t"
56                 "movdqa %%xmm2, %%xmm0        \n\t"
57                 "movdqa %%xmm6, %%xmm4        \n\t"
58                 "paddd  %%xmm2, %%xmm2        \n\t"
59                 "paddd  %%xmm6, %%xmm6        \n\t"
60                 "paddd  %%xmm0, %%xmm2        \n\t"
61                 "paddd  %%xmm4, %%xmm6        \n\t"
62                 "paddd  %%xmm7, %%xmm2        \n\t"
63                 "paddd  %%xmm7, %%xmm6        \n\t"
64                 "psrad      $3, %%xmm2        \n\t"
65                 "psrad      $3, %%xmm6        \n\t"
66                 "movdqa   (%0), %%xmm0        \n\t"
67                 "movdqa 16(%0), %%xmm4        \n\t"
68                 "psubd  %%xmm2, %%xmm0        \n\t"
69                 "psubd  %%xmm6, %%xmm4        \n\t"
70                 "movdqa %%xmm0, (%0)          \n\t"
71                 "movdqa %%xmm4, 16(%0)        \n\t"
72                 :: "r"(&b[i]), "r"(&ref[i])
73                 : "memory"
74             );
75         }
76         snow_horizontal_compose_lift_lead_out(i, b, b, ref, width, w_l, 0, W_DM, W_DO, W_DS);
77         b[0] = b_0 - ((W_DM * 2 * ref[1]+W_DO)>>W_DS);
78     }
79
80     { // Lift 1
81         DWTELEM * const dst = b+w2;
82
83         i = 0;
84         for(; (((long)&dst[i]) & 0xF) && i<w_r; i++){
85             dst[i] = dst[i] - (b[i] + b[i + 1]);
86         }
87         for(; i<w_r-7; i+=8){
88             asm volatile(
89                 "movdqu   (%1), %%xmm1        \n\t"
90                 "movdqu 16(%1), %%xmm5        \n\t"
91                 "movdqu  4(%1), %%xmm2        \n\t"
92                 "movdqu 20(%1), %%xmm6        \n\t"
93                 "paddd  %%xmm1, %%xmm2        \n\t"
94                 "paddd  %%xmm5, %%xmm6        \n\t"
95                 "movdqa   (%0), %%xmm0        \n\t"
96                 "movdqa 16(%0), %%xmm4        \n\t"
97                 "psubd  %%xmm2, %%xmm0        \n\t"
98                 "psubd  %%xmm6, %%xmm4        \n\t"
99                 "movdqa %%xmm0, (%0)          \n\t"
100                 "movdqa %%xmm4, 16(%0)        \n\t"
101                 :: "r"(&dst[i]), "r"(&b[i])
102                 : "memory"
103             );
104         }
105         snow_horizontal_compose_lift_lead_out(i, dst, dst, b, width, w_r, 1, W_CM, W_CO, W_CS);
106     }
107
108     { // Lift 2
109         DWTELEM * const ref = b+w2 - 1;
110         DWTELEM b_0 = b[0];
111
112         i = 0;
113         asm volatile(
114             "pslld          $1, %%xmm7       \n\t" /* xmm7 already holds a '4' from 2 lifts ago. */
115         ::);
116         for(; i<w_l-7; i+=8){
117             asm volatile(
118                 "movdqu   (%1), %%xmm1        \n\t"
119                 "movdqu 16(%1), %%xmm5        \n\t"
120                 "movdqu  4(%1), %%xmm0        \n\t"
121                 "movdqu 20(%1), %%xmm4        \n\t"
122                 "paddd  %%xmm1, %%xmm0        \n\t"
123                 "paddd  %%xmm5, %%xmm4        \n\t"
124                 "movdqa %%xmm7, %%xmm1        \n\t"
125                 "movdqa %%xmm7, %%xmm5        \n\t"
126                 "psubd  %%xmm0, %%xmm1        \n\t"
127                 "psubd  %%xmm4, %%xmm5        \n\t"
128                 "movdqa   (%0), %%xmm0        \n\t"
129                 "movdqa 16(%0), %%xmm4        \n\t"
130                 "pslld      $2, %%xmm0        \n\t"
131                 "pslld      $2, %%xmm4        \n\t"
132                 "psubd  %%xmm0, %%xmm1        \n\t"
133                 "psubd  %%xmm4, %%xmm5        \n\t"
134                 "psrad      $4, %%xmm1        \n\t"
135                 "psrad      $4, %%xmm5        \n\t"
136                 "movdqa   (%0), %%xmm0        \n\t"
137                 "movdqa 16(%0), %%xmm4        \n\t"
138                 "psubd  %%xmm1, %%xmm0        \n\t"
139                 "psubd  %%xmm5, %%xmm4        \n\t"
140                 "movdqa %%xmm0, (%0)          \n\t"
141                 "movdqa %%xmm4, 16(%0)        \n\t"
142                 :: "r"(&b[i]), "r"(&ref[i])
143                 : "memory"
144             );
145         }
146         snow_horizontal_compose_liftS_lead_out(i, b, b, ref, width, w_l);
147         b[0] = b_0 - (((-2 * ref[1] + W_BO) - 4 * b_0) >> W_BS);
148     }
149
150     { // Lift 3
151         DWTELEM * const src = b+w2;
152
153         i = 0;
154         for(; (((long)&temp[i]) & 0xF) && i<w_r; i++){
155             temp[i] = src[i] - ((-W_AM*(b[i] + b[i+1]))>>W_AS);
156         }
157         for(; i<w_r-7; i+=8){
158             asm volatile(
159                 "movdqu  4(%1), %%xmm2        \n\t"
160                 "movdqu 20(%1), %%xmm6        \n\t"
161                 "paddd    (%1), %%xmm2        \n\t"
162                 "paddd  16(%1), %%xmm6        \n\t"
163                 "movdqa %%xmm2, %%xmm0        \n\t"
164                 "movdqa %%xmm6, %%xmm4        \n\t"
165                 "pslld      $2, %%xmm2        \n\t"
166                 "pslld      $2, %%xmm6        \n\t"
167                 "psubd  %%xmm2, %%xmm0        \n\t"
168                 "psubd  %%xmm6, %%xmm4        \n\t"
169                 "psrad      $1, %%xmm0        \n\t"
170                 "psrad      $1, %%xmm4        \n\t"
171                 "movdqu   (%0), %%xmm2        \n\t"
172                 "movdqu 16(%0), %%xmm6        \n\t"
173                 "psubd  %%xmm0, %%xmm2        \n\t"
174                 "psubd  %%xmm4, %%xmm6        \n\t"
175                 "movdqa %%xmm2, (%2)          \n\t"
176                 "movdqa %%xmm6, 16(%2)        \n\t"
177                 :: "r"(&src[i]), "r"(&b[i]), "r"(&temp[i])
178                  : "memory"
179                );
180         }
181         snow_horizontal_compose_lift_lead_out(i, temp, src, b, width, w_r, 1, -W_AM, W_AO, W_AS);
182     }
183
184     {
185         snow_interleave_line_header(&i, width, b, temp);
186
187         for (; (i & 0x1E) != 0x1E; i-=2){
188             b[i+1] = temp[i>>1];
189             b[i] = b[i>>1];
190         }
191         for (i-=30; i>=0; i-=32){
192             asm volatile(
193                 "movdqa      (%1), %%xmm0       \n\t"
194                 "movdqa    16(%1), %%xmm2       \n\t"
195                 "movdqa    32(%1), %%xmm4       \n\t"
196                 "movdqa    48(%1), %%xmm6       \n\t"
197                 "movdqa      (%1), %%xmm1       \n\t"
198                 "movdqa    16(%1), %%xmm3       \n\t"
199                 "movdqa    32(%1), %%xmm5       \n\t"
200                 "movdqa    48(%1), %%xmm7       \n\t"
201                 "punpckldq   (%2), %%xmm0       \n\t"
202                 "punpckldq 16(%2), %%xmm2       \n\t"
203                 "punpckldq 32(%2), %%xmm4       \n\t"
204                 "punpckldq 48(%2), %%xmm6       \n\t"
205                 "movdqa    %%xmm0, (%0)         \n\t"
206                 "movdqa    %%xmm2, 32(%0)       \n\t"
207                 "movdqa    %%xmm4, 64(%0)       \n\t"
208                 "movdqa    %%xmm6, 96(%0)       \n\t"
209                 "punpckhdq   (%2), %%xmm1       \n\t"
210                 "punpckhdq 16(%2), %%xmm3       \n\t"
211                 "punpckhdq 32(%2), %%xmm5       \n\t"
212                 "punpckhdq 48(%2), %%xmm7       \n\t"
213                 "movdqa    %%xmm1, 16(%0)       \n\t"
214                 "movdqa    %%xmm3, 48(%0)       \n\t"
215                 "movdqa    %%xmm5, 80(%0)       \n\t"
216                 "movdqa    %%xmm7, 112(%0)      \n\t"
217                 :: "r"(&(b)[i]), "r"(&(b)[i>>1]), "r"(&(temp)[i>>1])
218                  : "memory"
219                );
220         }
221     }
222 }
223
224 void ff_snow_horizontal_compose97i_mmx(DWTELEM *b, int width){
225     const int w2= (width+1)>>1;
226     DWTELEM temp[width >> 1];
227     const int w_l= (width>>1);
228     const int w_r= w2 - 1;
229     int i;
230
231     { // Lift 0
232         DWTELEM * const ref = b + w2 - 1;
233
234         i = 1;
235         b[0] = b[0] - ((W_DM * 2 * ref[1]+W_DO)>>W_DS);
236         asm volatile(
237             "pcmpeqd    %%mm7, %%mm7         \n\t"
238             "pslld        $31, %%mm7         \n\t"
239             "psrld        $29, %%mm7         \n\t"
240            ::);
241         for(; i<w_l-3; i+=4){
242             asm volatile(
243                 "movq     (%1), %%mm2        \n\t"
244                 "movq    8(%1), %%mm6        \n\t"
245                 "paddd   4(%1), %%mm2        \n\t"
246                 "paddd  12(%1), %%mm6        \n\t"
247                 "movq    %%mm2, %%mm0        \n\t"
248                 "movq    %%mm6, %%mm4        \n\t"
249                 "paddd   %%mm2, %%mm2        \n\t"
250                 "paddd   %%mm6, %%mm6        \n\t"
251                 "paddd   %%mm0, %%mm2        \n\t"
252                 "paddd   %%mm4, %%mm6        \n\t"
253                 "paddd   %%mm7, %%mm2        \n\t"
254                 "paddd   %%mm7, %%mm6        \n\t"
255                 "psrad      $3, %%mm2        \n\t"
256                 "psrad      $3, %%mm6        \n\t"
257                 "movq     (%0), %%mm0        \n\t"
258                 "movq    8(%0), %%mm4        \n\t"
259                 "psubd   %%mm2, %%mm0        \n\t"
260                 "psubd   %%mm6, %%mm4        \n\t"
261                 "movq    %%mm0, (%0)         \n\t"
262                 "movq    %%mm4, 8(%0)        \n\t"
263                 :: "r"(&b[i]), "r"(&ref[i])
264                  : "memory"
265                );
266         }
267         snow_horizontal_compose_lift_lead_out(i, b, b, ref, width, w_l, 0, W_DM, W_DO, W_DS);
268     }
269
270     { // Lift 1
271         DWTELEM * const dst = b+w2;
272
273         i = 0;
274         for(; i<w_r-3; i+=4){
275             asm volatile(
276                 "movq     (%1), %%mm2        \n\t"
277                 "movq    8(%1), %%mm6        \n\t"
278                 "paddd   4(%1), %%mm2        \n\t"
279                 "paddd  12(%1), %%mm6        \n\t"
280                 "movq     (%0), %%mm0        \n\t"
281                 "movq    8(%0), %%mm4        \n\t"
282                 "psubd   %%mm2, %%mm0        \n\t"
283                 "psubd   %%mm6, %%mm4        \n\t"
284                 "movq    %%mm0, (%0)         \n\t"
285                 "movq    %%mm4, 8(%0)        \n\t"
286                 :: "r"(&dst[i]), "r"(&b[i])
287                  : "memory"
288                );
289         }
290         snow_horizontal_compose_lift_lead_out(i, dst, dst, b, width, w_r, 1, W_CM, W_CO, W_CS);
291     }
292
293     { // Lift 2
294         DWTELEM * const ref = b+w2 - 1;
295
296         i = 1;
297         b[0] = b[0] - (((-2 * ref[1] + W_BO) - 4 * b[0]) >> W_BS);
298         asm volatile(
299             "pslld          $1, %%mm7       \n\t" /* xmm7 already holds a '4' from 2 lifts ago. */
300            ::);
301         for(; i<w_l-3; i+=4){
302             asm volatile(
303                 "movq     (%1), %%mm0        \n\t"
304                 "movq    8(%1), %%mm4        \n\t"
305                 "paddd   4(%1), %%mm0        \n\t"
306                 "paddd  12(%1), %%mm4        \n\t"
307                 "movq    %%mm7, %%mm1        \n\t"
308                 "movq    %%mm7, %%mm5        \n\t"
309                 "psubd   %%mm0, %%mm1        \n\t"
310                 "psubd   %%mm4, %%mm5        \n\t"
311                 "movq     (%0), %%mm0        \n\t"
312                 "movq    8(%0), %%mm4        \n\t"
313                 "pslld      $2, %%mm0        \n\t"
314                 "pslld      $2, %%mm4        \n\t"
315                 "psubd   %%mm0, %%mm1        \n\t"
316                 "psubd   %%mm4, %%mm5        \n\t"
317                 "psrad      $4, %%mm1        \n\t"
318                 "psrad      $4, %%mm5        \n\t"
319                 "movq     (%0), %%mm0        \n\t"
320                 "movq    8(%0), %%mm4        \n\t"
321                 "psubd   %%mm1, %%mm0        \n\t"
322                 "psubd   %%mm5, %%mm4        \n\t"
323                 "movq    %%mm0, (%0)         \n\t"
324                 "movq    %%mm4, 8(%0)        \n\t"
325                 :: "r"(&b[i]), "r"(&ref[i])
326                  : "memory"
327                );
328         }
329         snow_horizontal_compose_liftS_lead_out(i, b, b, ref, width, w_l);
330     }
331
332     { // Lift 3
333         DWTELEM * const src = b+w2;
334         i = 0;
335
336         for(; i<w_r-3; i+=4){
337             asm volatile(
338                 "movq    4(%1), %%mm2        \n\t"
339                 "movq   12(%1), %%mm6        \n\t"
340                 "paddd    (%1), %%mm2        \n\t"
341                 "paddd   8(%1), %%mm6        \n\t"
342                 "movq    %%mm2, %%mm0        \n\t"
343                 "movq    %%mm6, %%mm4        \n\t"
344                 "pslld      $2, %%mm2        \n\t"
345                 "pslld      $2, %%mm6        \n\t"
346                 "psubd   %%mm2, %%mm0        \n\t"
347                 "psubd   %%mm6, %%mm4        \n\t"
348                 "psrad      $1, %%mm0        \n\t"
349                 "psrad      $1, %%mm4        \n\t"
350                 "movq     (%0), %%mm2        \n\t"
351                 "movq    8(%0), %%mm6        \n\t"
352                 "psubd   %%mm0, %%mm2        \n\t"
353                 "psubd   %%mm4, %%mm6        \n\t"
354                 "movq    %%mm2, (%2)         \n\t"
355                 "movq    %%mm6, 8(%2)        \n\t"
356                 :: "r"(&src[i]), "r"(&b[i]), "r"(&temp[i])
357                  : "memory"
358                );
359         }
360         snow_horizontal_compose_lift_lead_out(i, temp, src, b, width, w_r, 1, -W_AM, W_AO, W_AS);
361     }
362
363     {
364         snow_interleave_line_header(&i, width, b, temp);
365
366         for (; (i & 0xE) != 0xE; i-=2){
367             b[i+1] = temp[i>>1];
368             b[i] = b[i>>1];
369         }
370         for (i-=14; i>=0; i-=16){
371             asm volatile(
372                 "movq        (%1), %%mm0       \n\t"
373                 "movq       8(%1), %%mm2       \n\t"
374                 "movq      16(%1), %%mm4       \n\t"
375                 "movq      24(%1), %%mm6       \n\t"
376                 "movq        (%1), %%mm1       \n\t"
377                 "movq       8(%1), %%mm3       \n\t"
378                 "movq      16(%1), %%mm5       \n\t"
379                 "movq      24(%1), %%mm7       \n\t"
380                 "punpckldq   (%2), %%mm0       \n\t"
381                 "punpckldq  8(%2), %%mm2       \n\t"
382                 "punpckldq 16(%2), %%mm4       \n\t"
383                 "punpckldq 24(%2), %%mm6       \n\t"
384                 "movq       %%mm0, (%0)        \n\t"
385                 "movq       %%mm2, 16(%0)      \n\t"
386                 "movq       %%mm4, 32(%0)      \n\t"
387                 "movq       %%mm6, 48(%0)      \n\t"
388                 "punpckhdq   (%2), %%mm1       \n\t"
389                 "punpckhdq  8(%2), %%mm3       \n\t"
390                 "punpckhdq 16(%2), %%mm5       \n\t"
391                 "punpckhdq 24(%2), %%mm7       \n\t"
392                 "movq       %%mm1, 8(%0)       \n\t"
393                 "movq       %%mm3, 24(%0)      \n\t"
394                 "movq       %%mm5, 40(%0)      \n\t"
395                 "movq       %%mm7, 56(%0)      \n\t"
396                 :: "r"(&b[i]), "r"(&b[i>>1]), "r"(&temp[i>>1])
397                  : "memory"
398                );
399         }
400     }
401 }
402
403 #define snow_vertical_compose_sse2_load_add(op,r,t0,t1,t2,t3)\
404         ""op" (%%"r",%%"REG_d",4), %%"t0"      \n\t"\
405         ""op" 16(%%"r",%%"REG_d",4), %%"t1"    \n\t"\
406         ""op" 32(%%"r",%%"REG_d",4), %%"t2"    \n\t"\
407         ""op" 48(%%"r",%%"REG_d",4), %%"t3"    \n\t"
408
409 #define snow_vertical_compose_sse2_load(r,t0,t1,t2,t3)\
410         snow_vertical_compose_sse2_load_add("movdqa",r,t0,t1,t2,t3)
411
412 #define snow_vertical_compose_sse2_add(r,t0,t1,t2,t3)\
413         snow_vertical_compose_sse2_load_add("paddd",r,t0,t1,t2,t3)
414
415 #define snow_vertical_compose_sse2_sub(s0,s1,s2,s3,t0,t1,t2,t3)\
416         "psubd %%"s0", %%"t0" \n\t"\
417         "psubd %%"s1", %%"t1" \n\t"\
418         "psubd %%"s2", %%"t2" \n\t"\
419         "psubd %%"s3", %%"t3" \n\t"
420
421 #define snow_vertical_compose_sse2_store(w,s0,s1,s2,s3)\
422         "movdqa %%"s0", (%%"w",%%"REG_d",4)      \n\t"\
423         "movdqa %%"s1", 16(%%"w",%%"REG_d",4)    \n\t"\
424         "movdqa %%"s2", 32(%%"w",%%"REG_d",4)    \n\t"\
425         "movdqa %%"s3", 48(%%"w",%%"REG_d",4)    \n\t"
426
427 #define snow_vertical_compose_sse2_sra(n,t0,t1,t2,t3)\
428         "psrad $"n", %%"t0" \n\t"\
429         "psrad $"n", %%"t1" \n\t"\
430         "psrad $"n", %%"t2" \n\t"\
431         "psrad $"n", %%"t3" \n\t"
432
433 #define snow_vertical_compose_sse2_r2r_add(s0,s1,s2,s3,t0,t1,t2,t3)\
434         "paddd %%"s0", %%"t0" \n\t"\
435         "paddd %%"s1", %%"t1" \n\t"\
436         "paddd %%"s2", %%"t2" \n\t"\
437         "paddd %%"s3", %%"t3" \n\t"
438
439 #define snow_vertical_compose_sse2_sll(n,t0,t1,t2,t3)\
440         "pslld $"n", %%"t0" \n\t"\
441         "pslld $"n", %%"t1" \n\t"\
442         "pslld $"n", %%"t2" \n\t"\
443         "pslld $"n", %%"t3" \n\t"
444
445 #define snow_vertical_compose_sse2_move(s0,s1,s2,s3,t0,t1,t2,t3)\
446         "movdqa %%"s0", %%"t0" \n\t"\
447         "movdqa %%"s1", %%"t1" \n\t"\
448         "movdqa %%"s2", %%"t2" \n\t"\
449         "movdqa %%"s3", %%"t3" \n\t"
450
451 void ff_snow_vertical_compose97i_sse2(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width){
452     long i = width;
453
454     while(i & 0xF)
455     {
456         i--;
457         b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;
458         b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS;
459         b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS;
460         b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
461     }
462
463          asm volatile (
464         "jmp 2f                                      \n\t"
465         "1:                                          \n\t"
466
467         "mov %6, %%"REG_a"                           \n\t"
468         "mov %4, %%"REG_S"                           \n\t"
469
470         snow_vertical_compose_sse2_load(REG_S,"xmm0","xmm2","xmm4","xmm6")
471         snow_vertical_compose_sse2_add(REG_a,"xmm0","xmm2","xmm4","xmm6")
472         snow_vertical_compose_sse2_move("xmm0","xmm2","xmm4","xmm6","xmm1","xmm3","xmm5","xmm7")
473         snow_vertical_compose_sse2_sll("1","xmm0","xmm2","xmm4","xmm6")\
474         snow_vertical_compose_sse2_r2r_add("xmm1","xmm3","xmm5","xmm7","xmm0","xmm2","xmm4","xmm6")
475
476         "pcmpeqd %%xmm1, %%xmm1                      \n\t"
477         "pslld $31, %%xmm1                           \n\t"
478         "psrld $29, %%xmm1                           \n\t"
479         "mov %5, %%"REG_a"                           \n\t"
480
481         snow_vertical_compose_sse2_r2r_add("xmm1","xmm1","xmm1","xmm1","xmm0","xmm2","xmm4","xmm6")
482         snow_vertical_compose_sse2_sra("3","xmm0","xmm2","xmm4","xmm6")
483         snow_vertical_compose_sse2_load(REG_a,"xmm1","xmm3","xmm5","xmm7")
484         snow_vertical_compose_sse2_sub("xmm0","xmm2","xmm4","xmm6","xmm1","xmm3","xmm5","xmm7")
485         snow_vertical_compose_sse2_store(REG_a,"xmm1","xmm3","xmm5","xmm7")
486         "mov %3, %%"REG_c"                           \n\t"
487         snow_vertical_compose_sse2_load(REG_S,"xmm0","xmm2","xmm4","xmm6")
488         snow_vertical_compose_sse2_add(REG_c,"xmm1","xmm3","xmm5","xmm7")
489         snow_vertical_compose_sse2_sub("xmm1","xmm3","xmm5","xmm7","xmm0","xmm2","xmm4","xmm6")
490         snow_vertical_compose_sse2_store(REG_S,"xmm0","xmm2","xmm4","xmm6")
491         "mov %2, %%"REG_a"                           \n\t"
492         snow_vertical_compose_sse2_load(REG_c,"xmm1","xmm3","xmm5","xmm7")
493         snow_vertical_compose_sse2_add(REG_a,"xmm0","xmm2","xmm4","xmm6")
494         snow_vertical_compose_sse2_sll("2","xmm1","xmm3","xmm5","xmm7")\
495         snow_vertical_compose_sse2_r2r_add("xmm1","xmm3","xmm5","xmm7","xmm0","xmm2","xmm4","xmm6")
496
497         "pcmpeqd %%xmm1, %%xmm1                      \n\t"
498         "pslld $31, %%xmm1                           \n\t"
499         "psrld $28, %%xmm1                           \n\t"
500         "mov %1, %%"REG_S"                           \n\t"
501
502         snow_vertical_compose_sse2_r2r_add("xmm1","xmm1","xmm1","xmm1","xmm0","xmm2","xmm4","xmm6")
503         snow_vertical_compose_sse2_sra("4","xmm0","xmm2","xmm4","xmm6")
504         snow_vertical_compose_sse2_add(REG_c,"xmm0","xmm2","xmm4","xmm6")
505         snow_vertical_compose_sse2_store(REG_c,"xmm0","xmm2","xmm4","xmm6")
506         snow_vertical_compose_sse2_add(REG_S,"xmm0","xmm2","xmm4","xmm6")
507         snow_vertical_compose_sse2_move("xmm0","xmm2","xmm4","xmm6","xmm1","xmm3","xmm5","xmm7")
508         snow_vertical_compose_sse2_sll("1","xmm0","xmm2","xmm4","xmm6")\
509         snow_vertical_compose_sse2_r2r_add("xmm1","xmm3","xmm5","xmm7","xmm0","xmm2","xmm4","xmm6")
510         snow_vertical_compose_sse2_sra("1","xmm0","xmm2","xmm4","xmm6")
511         snow_vertical_compose_sse2_add(REG_a,"xmm0","xmm2","xmm4","xmm6")
512         snow_vertical_compose_sse2_store(REG_a,"xmm0","xmm2","xmm4","xmm6")
513
514         "2:                                          \n\t"
515         "sub $16, %%"REG_d"                          \n\t"
516         "jge 1b                                      \n\t"
517         :"+d"(i)
518         :
519         "m"(b0),"m"(b1),"m"(b2),"m"(b3),"m"(b4),"m"(b5):
520         "%"REG_a"","%"REG_S"","%"REG_c"");
521 }
522
523 #define snow_vertical_compose_mmx_load_add(op,r,t0,t1,t2,t3)\
524         ""op" (%%"r",%%"REG_d",4), %%"t0"   \n\t"\
525         ""op" 8(%%"r",%%"REG_d",4), %%"t1"  \n\t"\
526         ""op" 16(%%"r",%%"REG_d",4), %%"t2" \n\t"\
527         ""op" 24(%%"r",%%"REG_d",4), %%"t3" \n\t"
528
529 #define snow_vertical_compose_mmx_load(r,t0,t1,t2,t3)\
530         snow_vertical_compose_mmx_load_add("movq",r,t0,t1,t2,t3)
531
532 #define snow_vertical_compose_mmx_add(r,t0,t1,t2,t3)\
533         snow_vertical_compose_mmx_load_add("paddd",r,t0,t1,t2,t3)
534
535 #define snow_vertical_compose_mmx_sub(s0,s1,s2,s3,t0,t1,t2,t3)\
536         snow_vertical_compose_sse2_sub(s0,s1,s2,s3,t0,t1,t2,t3)
537
538 #define snow_vertical_compose_mmx_store(w,s0,s1,s2,s3)\
539         "movq %%"s0", (%%"w",%%"REG_d",4)   \n\t"\
540         "movq %%"s1", 8(%%"w",%%"REG_d",4)  \n\t"\
541         "movq %%"s2", 16(%%"w",%%"REG_d",4) \n\t"\
542         "movq %%"s3", 24(%%"w",%%"REG_d",4) \n\t"
543
544 #define snow_vertical_compose_mmx_sra(n,t0,t1,t2,t3)\
545         snow_vertical_compose_sse2_sra(n,t0,t1,t2,t3)
546
547 #define snow_vertical_compose_mmx_r2r_add(s0,s1,s2,s3,t0,t1,t2,t3)\
548         snow_vertical_compose_sse2_r2r_add(s0,s1,s2,s3,t0,t1,t2,t3)
549
550 #define snow_vertical_compose_mmx_sll(n,t0,t1,t2,t3)\
551         snow_vertical_compose_sse2_sll(n,t0,t1,t2,t3)
552
553 #define snow_vertical_compose_mmx_move(s0,s1,s2,s3,t0,t1,t2,t3)\
554         "movq %%"s0", %%"t0" \n\t"\
555         "movq %%"s1", %%"t1" \n\t"\
556         "movq %%"s2", %%"t2" \n\t"\
557         "movq %%"s3", %%"t3" \n\t"
558
559 void ff_snow_vertical_compose97i_mmx(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width){
560     long i = width;
561     while(i & 0x7)
562     {
563         i--;
564         b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;
565         b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS;
566         b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS;
567         b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
568     }
569
570     asm volatile(
571         "jmp 2f                                      \n\t"
572         "1:                                          \n\t"
573
574         "mov %6, %%"REG_a"                           \n\t"
575         "mov %4, %%"REG_S"                           \n\t"
576
577         snow_vertical_compose_mmx_load(REG_S,"mm0","mm2","mm4","mm6")
578         snow_vertical_compose_mmx_add(REG_a,"mm0","mm2","mm4","mm6")
579         snow_vertical_compose_mmx_move("mm0","mm2","mm4","mm6","mm1","mm3","mm5","mm7")
580         snow_vertical_compose_mmx_sll("1","mm0","mm2","mm4","mm6")
581         snow_vertical_compose_mmx_r2r_add("mm1","mm3","mm5","mm7","mm0","mm2","mm4","mm6")
582
583         "pcmpeqd %%mm1, %%mm1                        \n\t"
584         "pslld $31, %%mm1                            \n\t"
585         "psrld $29, %%mm1                            \n\t"
586         "mov %5, %%"REG_a"                           \n\t"
587
588         snow_vertical_compose_mmx_r2r_add("mm1","mm1","mm1","mm1","mm0","mm2","mm4","mm6")
589         snow_vertical_compose_mmx_sra("3","mm0","mm2","mm4","mm6")
590         snow_vertical_compose_mmx_load(REG_a,"mm1","mm3","mm5","mm7")
591         snow_vertical_compose_mmx_sub("mm0","mm2","mm4","mm6","mm1","mm3","mm5","mm7")
592         snow_vertical_compose_mmx_store(REG_a,"mm1","mm3","mm5","mm7")
593         "mov %3, %%"REG_c"                           \n\t"
594         snow_vertical_compose_mmx_load(REG_S,"mm0","mm2","mm4","mm6")
595         snow_vertical_compose_mmx_add(REG_c,"mm1","mm3","mm5","mm7")
596         snow_vertical_compose_mmx_sub("mm1","mm3","mm5","mm7","mm0","mm2","mm4","mm6")
597         snow_vertical_compose_mmx_store(REG_S,"mm0","mm2","mm4","mm6")
598         "mov %2, %%"REG_a"                           \n\t"
599         snow_vertical_compose_mmx_load(REG_c,"mm1","mm3","mm5","mm7")
600         snow_vertical_compose_mmx_add(REG_a,"mm0","mm2","mm4","mm6")
601         snow_vertical_compose_mmx_sll("2","mm1","mm3","mm5","mm7")
602         snow_vertical_compose_mmx_r2r_add("mm1","mm3","mm5","mm7","mm0","mm2","mm4","mm6")
603
604         "pcmpeqd %%mm1, %%mm1                        \n\t"
605         "pslld $31, %%mm1                            \n\t"
606         "psrld $28, %%mm1                            \n\t"
607         "mov %1, %%"REG_S"                           \n\t"
608
609         snow_vertical_compose_mmx_r2r_add("mm1","mm1","mm1","mm1","mm0","mm2","mm4","mm6")
610         snow_vertical_compose_mmx_sra("4","mm0","mm2","mm4","mm6")
611         snow_vertical_compose_mmx_add(REG_c,"mm0","mm2","mm4","mm6")
612         snow_vertical_compose_mmx_store(REG_c,"mm0","mm2","mm4","mm6")
613         snow_vertical_compose_mmx_add(REG_S,"mm0","mm2","mm4","mm6")
614         snow_vertical_compose_mmx_move("mm0","mm2","mm4","mm6","mm1","mm3","mm5","mm7")
615         snow_vertical_compose_mmx_sll("1","mm0","mm2","mm4","mm6")
616         snow_vertical_compose_mmx_r2r_add("mm1","mm3","mm5","mm7","mm0","mm2","mm4","mm6")
617         snow_vertical_compose_mmx_sra("1","mm0","mm2","mm4","mm6")
618         snow_vertical_compose_mmx_add(REG_a,"mm0","mm2","mm4","mm6")
619         snow_vertical_compose_mmx_store(REG_a,"mm0","mm2","mm4","mm6")
620
621         "2:                                          \n\t"
622         "sub $8, %%"REG_d"                           \n\t"
623         "jge 1b                                      \n\t"
624         :"+d"(i)
625         :
626         "m"(b0),"m"(b1),"m"(b2),"m"(b3),"m"(b4),"m"(b5):
627         "%"REG_a"","%"REG_S"","%"REG_c"");
628 }
629
630 #define snow_inner_add_yblock_sse2_header \
631     DWTELEM * * dst_array = sb->line + src_y;\
632     long tmp;\
633     asm volatile(\
634              "mov  %7, %%"REG_c"             \n\t"\
635              "mov  %6, %2                    \n\t"\
636              "mov  %4, %%"REG_S"             \n\t"\
637              "pxor %%xmm7, %%xmm7            \n\t" /* 0 */\
638              "pcmpeqd %%xmm3, %%xmm3         \n\t"\
639              "pslld $31, %%xmm3              \n\t"\
640              "psrld $24, %%xmm3              \n\t" /* FRAC_BITS >> 1 */\
641              "1:                             \n\t"\
642              "mov %1, %%"REG_D"              \n\t"\
643              "mov (%%"REG_D"), %%"REG_D"     \n\t"\
644              "add %3, %%"REG_D"              \n\t"
645
646 #define snow_inner_add_yblock_sse2_start_8(out_reg1, out_reg2, ptr_offset, s_offset)\
647              "mov "PTR_SIZE"*"ptr_offset"(%%"REG_a"), %%"REG_d"; \n\t"\
648              "movq (%%"REG_d"), %%"out_reg1" \n\t"\
649              "movq (%%"REG_d", %%"REG_c"), %%"out_reg2" \n\t"\
650              "punpcklbw %%xmm7, %%"out_reg1" \n\t"\
651              "punpcklbw %%xmm7, %%"out_reg2" \n\t"\
652              "movq "s_offset"(%%"REG_S"), %%xmm0 \n\t"\
653              "movq "s_offset"+16(%%"REG_S"), %%xmm4 \n\t"\
654              "punpcklbw %%xmm7, %%xmm0       \n\t"\
655              "punpcklbw %%xmm7, %%xmm4       \n\t"\
656              "pmullw %%xmm0, %%"out_reg1"    \n\t"\
657              "pmullw %%xmm4, %%"out_reg2"    \n\t"
658
659 #define snow_inner_add_yblock_sse2_start_16(out_reg1, out_reg2, ptr_offset, s_offset)\
660              "mov "PTR_SIZE"*"ptr_offset"(%%"REG_a"), %%"REG_d"; \n\t"\
661              "movq (%%"REG_d"), %%"out_reg1" \n\t"\
662              "movq 8(%%"REG_d"), %%"out_reg2" \n\t"\
663              "punpcklbw %%xmm7, %%"out_reg1" \n\t"\
664              "punpcklbw %%xmm7, %%"out_reg2" \n\t"\
665              "movq "s_offset"(%%"REG_S"), %%xmm0 \n\t"\
666              "movq "s_offset"+8(%%"REG_S"), %%xmm4 \n\t"\
667              "punpcklbw %%xmm7, %%xmm0       \n\t"\
668              "punpcklbw %%xmm7, %%xmm4       \n\t"\
669              "pmullw %%xmm0, %%"out_reg1"    \n\t"\
670              "pmullw %%xmm4, %%"out_reg2"    \n\t"
671
672 #define snow_inner_add_yblock_sse2_accum_8(ptr_offset, s_offset) \
673              snow_inner_add_yblock_sse2_start_8("xmm2", "xmm6", ptr_offset, s_offset)\
674              "paddusw %%xmm2, %%xmm1         \n\t"\
675              "paddusw %%xmm6, %%xmm5         \n\t"
676
677 #define snow_inner_add_yblock_sse2_accum_16(ptr_offset, s_offset) \
678              snow_inner_add_yblock_sse2_start_16("xmm2", "xmm6", ptr_offset, s_offset)\
679              "paddusw %%xmm2, %%xmm1         \n\t"\
680              "paddusw %%xmm6, %%xmm5         \n\t"
681
682 #define snow_inner_add_yblock_sse2_end_common1\
683              "add $32, %%"REG_S"             \n\t"\
684              "add %%"REG_c", %0              \n\t"\
685              "add %%"REG_c", "PTR_SIZE"*3(%%"REG_a");\n\t"\
686              "add %%"REG_c", "PTR_SIZE"*2(%%"REG_a");\n\t"\
687              "add %%"REG_c", "PTR_SIZE"*1(%%"REG_a");\n\t"\
688              "add %%"REG_c", (%%"REG_a")     \n\t"
689
690 #define snow_inner_add_yblock_sse2_end_common2\
691              "jnz 1b                         \n\t"\
692              :"+m"(dst8),"+m"(dst_array),"=&r"(tmp)\
693              :\
694              "rm"((long)(src_x<<2)),"m"(obmc),"a"(block),"m"((long)b_h),"m"((long)src_stride):\
695              "%"REG_c"","%"REG_S"","%"REG_D"","%"REG_d"");
696
697 #define snow_inner_add_yblock_sse2_end_8\
698              "sal $1, %%"REG_c"              \n\t"\
699              "add $"PTR_SIZE"*2, %1          \n\t"\
700              snow_inner_add_yblock_sse2_end_common1\
701              "sar $1, %%"REG_c"              \n\t"\
702              "sub $2, %2                     \n\t"\
703              snow_inner_add_yblock_sse2_end_common2
704
705 #define snow_inner_add_yblock_sse2_end_16\
706              "add $"PTR_SIZE"*1, %1          \n\t"\
707              snow_inner_add_yblock_sse2_end_common1\
708              "dec %2                         \n\t"\
709              snow_inner_add_yblock_sse2_end_common2
710
711 static void inner_add_yblock_bw_8_obmc_16_bh_even_sse2(const uint8_t *obmc, const long obmc_stride, uint8_t * * block, int b_w, long b_h,
712                       int src_x, int src_y, long src_stride, slice_buffer * sb, int add, uint8_t * dst8){
713 snow_inner_add_yblock_sse2_header
714 snow_inner_add_yblock_sse2_start_8("xmm1", "xmm5", "3", "0")
715 snow_inner_add_yblock_sse2_accum_8("2", "8")
716 snow_inner_add_yblock_sse2_accum_8("1", "128")
717 snow_inner_add_yblock_sse2_accum_8("0", "136")
718
719              "mov %0, %%"REG_d"              \n\t"
720              "movdqa (%%"REG_D"), %%xmm0     \n\t"
721              "movdqa %%xmm1, %%xmm2          \n\t"
722
723              "punpckhwd %%xmm7, %%xmm1       \n\t"
724              "punpcklwd %%xmm7, %%xmm2       \n\t"
725              "paddd %%xmm2, %%xmm0           \n\t"
726              "movdqa 16(%%"REG_D"), %%xmm2   \n\t"
727              "paddd %%xmm1, %%xmm2           \n\t"
728              "paddd %%xmm3, %%xmm0           \n\t"
729              "paddd %%xmm3, %%xmm2           \n\t"
730
731              "mov %1, %%"REG_D"              \n\t"
732              "mov "PTR_SIZE"(%%"REG_D"), %%"REG_D";\n\t"
733              "add %3, %%"REG_D"              \n\t"
734
735              "movdqa (%%"REG_D"), %%xmm4     \n\t"
736              "movdqa %%xmm5, %%xmm6          \n\t"
737              "punpckhwd %%xmm7, %%xmm5       \n\t"
738              "punpcklwd %%xmm7, %%xmm6       \n\t"
739              "paddd %%xmm6, %%xmm4           \n\t"
740              "movdqa 16(%%"REG_D"), %%xmm6   \n\t"
741              "paddd %%xmm5, %%xmm6           \n\t"
742              "paddd %%xmm3, %%xmm4           \n\t"
743              "paddd %%xmm3, %%xmm6           \n\t"
744
745              "psrad $8, %%xmm0               \n\t" /* FRAC_BITS. */
746              "psrad $8, %%xmm2               \n\t" /* FRAC_BITS. */
747              "packssdw %%xmm2, %%xmm0        \n\t"
748              "packuswb %%xmm7, %%xmm0        \n\t"
749              "movq %%xmm0, (%%"REG_d")       \n\t"
750
751              "psrad $8, %%xmm4               \n\t" /* FRAC_BITS. */
752              "psrad $8, %%xmm6               \n\t" /* FRAC_BITS. */
753              "packssdw %%xmm6, %%xmm4        \n\t"
754              "packuswb %%xmm7, %%xmm4        \n\t"
755              "movq %%xmm4, (%%"REG_d",%%"REG_c");\n\t"
756 snow_inner_add_yblock_sse2_end_8
757 }
758
759 static void inner_add_yblock_bw_16_obmc_32_sse2(const uint8_t *obmc, const long obmc_stride, uint8_t * * block, int b_w, long b_h,
760                       int src_x, int src_y, long src_stride, slice_buffer * sb, int add, uint8_t * dst8){
761 snow_inner_add_yblock_sse2_header
762 snow_inner_add_yblock_sse2_start_16("xmm1", "xmm5", "3", "0")
763 snow_inner_add_yblock_sse2_accum_16("2", "16")
764 snow_inner_add_yblock_sse2_accum_16("1", "512")
765 snow_inner_add_yblock_sse2_accum_16("0", "528")
766
767              "mov %0, %%"REG_d"              \n\t"
768              "movdqa %%xmm1, %%xmm0          \n\t"
769              "movdqa %%xmm5, %%xmm4          \n\t"
770              "punpcklwd %%xmm7, %%xmm0       \n\t"
771              "paddd (%%"REG_D"), %%xmm0      \n\t"
772              "punpckhwd %%xmm7, %%xmm1       \n\t"
773              "paddd 16(%%"REG_D"), %%xmm1    \n\t"
774              "punpcklwd %%xmm7, %%xmm4       \n\t"
775              "paddd 32(%%"REG_D"), %%xmm4    \n\t"
776              "punpckhwd %%xmm7, %%xmm5       \n\t"
777              "paddd 48(%%"REG_D"), %%xmm5    \n\t"
778              "paddd %%xmm3, %%xmm0           \n\t"
779              "paddd %%xmm3, %%xmm1           \n\t"
780              "paddd %%xmm3, %%xmm4           \n\t"
781              "paddd %%xmm3, %%xmm5           \n\t"
782              "psrad $8, %%xmm0               \n\t" /* FRAC_BITS. */
783              "psrad $8, %%xmm1               \n\t" /* FRAC_BITS. */
784              "psrad $8, %%xmm4               \n\t" /* FRAC_BITS. */
785              "psrad $8, %%xmm5               \n\t" /* FRAC_BITS. */
786
787              "packssdw %%xmm1, %%xmm0        \n\t"
788              "packssdw %%xmm5, %%xmm4        \n\t"
789              "packuswb %%xmm4, %%xmm0        \n\t"
790
791              "movdqu %%xmm0, (%%"REG_d")       \n\t"
792
793 snow_inner_add_yblock_sse2_end_16
794 }
795
796 #define snow_inner_add_yblock_mmx_header \
797     DWTELEM * * dst_array = sb->line + src_y;\
798     long tmp;\
799     asm volatile(\
800              "mov  %7, %%"REG_c"             \n\t"\
801              "mov  %6, %2                    \n\t"\
802              "mov  %4, %%"REG_S"             \n\t"\
803              "pxor %%mm7, %%mm7              \n\t" /* 0 */\
804              "pcmpeqd %%mm3, %%mm3           \n\t"\
805              "pslld $31, %%mm3               \n\t"\
806              "psrld $24, %%mm3               \n\t" /* FRAC_BITS >> 1 */\
807              "1:                             \n\t"\
808              "mov %1, %%"REG_D"              \n\t"\
809              "mov (%%"REG_D"), %%"REG_D"     \n\t"\
810              "add %3, %%"REG_D"              \n\t"
811
812 #define snow_inner_add_yblock_mmx_start(out_reg1, out_reg2, ptr_offset, s_offset, d_offset)\
813              "mov "PTR_SIZE"*"ptr_offset"(%%"REG_a"), %%"REG_d"; \n\t"\
814              "movd "d_offset"(%%"REG_d"), %%"out_reg1" \n\t"\
815              "movd "d_offset"+4(%%"REG_d"), %%"out_reg2" \n\t"\
816              "punpcklbw %%mm7, %%"out_reg1" \n\t"\
817              "punpcklbw %%mm7, %%"out_reg2" \n\t"\
818              "movd "s_offset"(%%"REG_S"), %%mm0 \n\t"\
819              "movd "s_offset"+4(%%"REG_S"), %%mm4 \n\t"\
820              "punpcklbw %%mm7, %%mm0       \n\t"\
821              "punpcklbw %%mm7, %%mm4       \n\t"\
822              "pmullw %%mm0, %%"out_reg1"    \n\t"\
823              "pmullw %%mm4, %%"out_reg2"    \n\t"
824
825 #define snow_inner_add_yblock_mmx_accum(ptr_offset, s_offset, d_offset) \
826              snow_inner_add_yblock_mmx_start("mm2", "mm6", ptr_offset, s_offset, d_offset)\
827              "paddusw %%mm2, %%mm1         \n\t"\
828              "paddusw %%mm6, %%mm5         \n\t"
829
830 #define snow_inner_add_yblock_mmx_mix(read_offset, write_offset)\
831              "mov %0, %%"REG_d"              \n\t"\
832              "movq %%mm1, %%mm0              \n\t"\
833              "movq %%mm5, %%mm4              \n\t"\
834              "punpcklwd %%mm7, %%mm0         \n\t"\
835              "paddd "read_offset"(%%"REG_D"), %%mm0 \n\t"\
836              "punpckhwd %%mm7, %%mm1         \n\t"\
837              "paddd "read_offset"+8(%%"REG_D"), %%mm1 \n\t"\
838              "punpcklwd %%mm7, %%mm4         \n\t"\
839              "paddd "read_offset"+16(%%"REG_D"), %%mm4 \n\t"\
840              "punpckhwd %%mm7, %%mm5         \n\t"\
841              "paddd "read_offset"+24(%%"REG_D"), %%mm5 \n\t"\
842              "paddd %%mm3, %%mm0             \n\t"\
843              "paddd %%mm3, %%mm1             \n\t"\
844              "paddd %%mm3, %%mm4             \n\t"\
845              "paddd %%mm3, %%mm5             \n\t"\
846              "psrad $8, %%mm0                \n\t"\
847              "psrad $8, %%mm1                \n\t"\
848              "psrad $8, %%mm4                \n\t"\
849              "psrad $8, %%mm5                \n\t"\
850 \
851              "packssdw %%mm1, %%mm0          \n\t"\
852              "packssdw %%mm5, %%mm4          \n\t"\
853              "packuswb %%mm4, %%mm0          \n\t"\
854              "movq %%mm0, "write_offset"(%%"REG_d") \n\t"
855
856 #define snow_inner_add_yblock_mmx_end(s_step)\
857              "add $"s_step", %%"REG_S"             \n\t"\
858              "add %%"REG_c", "PTR_SIZE"*3(%%"REG_a");\n\t"\
859              "add %%"REG_c", "PTR_SIZE"*2(%%"REG_a");\n\t"\
860              "add %%"REG_c", "PTR_SIZE"*1(%%"REG_a");\n\t"\
861              "add %%"REG_c", (%%"REG_a")     \n\t"\
862              "add $"PTR_SIZE"*1, %1          \n\t"\
863              "add %%"REG_c", %0              \n\t"\
864              "dec %2                         \n\t"\
865              "jnz 1b                         \n\t"\
866              :"+m"(dst8),"+m"(dst_array),"=&r"(tmp)\
867              :\
868              "rm"((long)(src_x<<2)),"m"(obmc),"a"(block),"m"((long)b_h),"m"((long)src_stride):\
869              "%"REG_c"","%"REG_S"","%"REG_D"","%"REG_d"");
870
871 static void inner_add_yblock_bw_8_obmc_16_mmx(const uint8_t *obmc, const long obmc_stride, uint8_t * * block, int b_w, long b_h,
872                       int src_x, int src_y, long src_stride, slice_buffer * sb, int add, uint8_t * dst8){
873 snow_inner_add_yblock_mmx_header
874 snow_inner_add_yblock_mmx_start("mm1", "mm5", "3", "0", "0")
875 snow_inner_add_yblock_mmx_accum("2", "8", "0")
876 snow_inner_add_yblock_mmx_accum("1", "128", "0")
877 snow_inner_add_yblock_mmx_accum("0", "136", "0")
878 snow_inner_add_yblock_mmx_mix("0", "0")
879 snow_inner_add_yblock_mmx_end("16")
880 }
881
882 static void inner_add_yblock_bw_16_obmc_32_mmx(const uint8_t *obmc, const long obmc_stride, uint8_t * * block, int b_w, long b_h,
883                       int src_x, int src_y, long src_stride, slice_buffer * sb, int add, uint8_t * dst8){
884 snow_inner_add_yblock_mmx_header
885 snow_inner_add_yblock_mmx_start("mm1", "mm5", "3", "0", "0")
886 snow_inner_add_yblock_mmx_accum("2", "16", "0")
887 snow_inner_add_yblock_mmx_accum("1", "512", "0")
888 snow_inner_add_yblock_mmx_accum("0", "528", "0")
889 snow_inner_add_yblock_mmx_mix("0", "0")
890
891 snow_inner_add_yblock_mmx_start("mm1", "mm5", "3", "8", "8")
892 snow_inner_add_yblock_mmx_accum("2", "24", "8")
893 snow_inner_add_yblock_mmx_accum("1", "520", "8")
894 snow_inner_add_yblock_mmx_accum("0", "536", "8")
895 snow_inner_add_yblock_mmx_mix("32", "8")
896 snow_inner_add_yblock_mmx_end("32")
897 }
898
899 void ff_snow_inner_add_yblock_sse2(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
900                            int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){
901
902     if (b_w == 16)
903         inner_add_yblock_bw_16_obmc_32_sse2(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
904     else if (b_w == 8 && obmc_stride == 16) {
905         if (!(b_h & 1))
906             inner_add_yblock_bw_8_obmc_16_bh_even_sse2(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
907         else
908             inner_add_yblock_bw_8_obmc_16_mmx(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
909     } else
910          ff_snow_inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
911 }
912
913 void ff_snow_inner_add_yblock_mmx(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
914                           int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){
915     if (b_w == 16)
916         inner_add_yblock_bw_16_obmc_32_mmx(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
917     else if (b_w == 8 && obmc_stride == 16)
918         inner_add_yblock_bw_8_obmc_16_mmx(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
919     else
920         ff_snow_inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
921 }