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 <stdint.h>
23 #include "config.h"
24 #include "libavutil/attributes.h"
25 #include "libavutil/cpu.h"
26 #include "libavutil/x86/asm.h"
27 #include "libavcodec/snow.h"
28 #include "libavcodec/snow_dwt.h"
29
30 #if HAVE_INLINE_ASM
31
ff_snow_horizontal_compose97i_sse2(IDWTELEM *b, IDWTELEM *temp, int width)32 static void ff_snow_horizontal_compose97i_sse2(IDWTELEM *b, IDWTELEM *temp, int width){
33 const int w2= (width+1)>>1;
34 const int w_l= (width>>1);
35 const int w_r= w2 - 1;
36 int i;
37
38 { // Lift 0
39 IDWTELEM * const ref = b + w2 - 1;
40 IDWTELEM b_0 = b[0]; //By allowing the first entry in b[0] to be calculated twice
41 // (the first time erroneously), we allow the SSE2 code to run an extra pass.
42 // The savings in code and time are well worth having to store this value and
43 // calculate b[0] correctly afterwards.
44
45 i = 0;
46 __asm__ volatile(
47 "pcmpeqd %%xmm7, %%xmm7 \n\t"
48 "pcmpeqd %%xmm3, %%xmm3 \n\t"
49 "psllw $1, %%xmm3 \n\t"
50 "paddw %%xmm7, %%xmm3 \n\t"
51 "psllw $13, %%xmm3 \n\t"
52 ::);
53 for(; i<w_l-15; i+=16){
54 __asm__ volatile(
55 "movdqu (%1), %%xmm1 \n\t"
56 "movdqu 16(%1), %%xmm5 \n\t"
57 "movdqu 2(%1), %%xmm2 \n\t"
58 "movdqu 18(%1), %%xmm6 \n\t"
59 "paddw %%xmm1, %%xmm2 \n\t"
60 "paddw %%xmm5, %%xmm6 \n\t"
61 "paddw %%xmm7, %%xmm2 \n\t"
62 "paddw %%xmm7, %%xmm6 \n\t"
63 "pmulhw %%xmm3, %%xmm2 \n\t"
64 "pmulhw %%xmm3, %%xmm6 \n\t"
65 "paddw (%0), %%xmm2 \n\t"
66 "paddw 16(%0), %%xmm6 \n\t"
67 "movdqa %%xmm2, (%0) \n\t"
68 "movdqa %%xmm6, 16(%0) \n\t"
69 :: "r"(&b[i]), "r"(&ref[i])
70 : "memory"
71 );
72 }
73 snow_horizontal_compose_lift_lead_out(i, b, b, ref, width, w_l, 0, W_DM, W_DO, W_DS);
74 b[0] = b_0 - ((W_DM * 2 * ref[1]+W_DO)>>W_DS);
75 }
76
77 { // Lift 1
78 IDWTELEM * const dst = b+w2;
79
80 i = 0;
81 for(; (((x86_reg)&dst[i]) & 0x1F) && i<w_r; i++){
82 dst[i] = dst[i] - (b[i] + b[i + 1]);
83 }
84 for(; i<w_r-15; i+=16){
85 __asm__ volatile(
86 "movdqu (%1), %%xmm1 \n\t"
87 "movdqu 16(%1), %%xmm5 \n\t"
88 "movdqu 2(%1), %%xmm2 \n\t"
89 "movdqu 18(%1), %%xmm6 \n\t"
90 "paddw %%xmm1, %%xmm2 \n\t"
91 "paddw %%xmm5, %%xmm6 \n\t"
92 "movdqa (%0), %%xmm0 \n\t"
93 "movdqa 16(%0), %%xmm4 \n\t"
94 "psubw %%xmm2, %%xmm0 \n\t"
95 "psubw %%xmm6, %%xmm4 \n\t"
96 "movdqa %%xmm0, (%0) \n\t"
97 "movdqa %%xmm4, 16(%0) \n\t"
98 :: "r"(&dst[i]), "r"(&b[i])
99 : "memory"
100 );
101 }
102 snow_horizontal_compose_lift_lead_out(i, dst, dst, b, width, w_r, 1, W_CM, W_CO, W_CS);
103 }
104
105 { // Lift 2
106 IDWTELEM * const ref = b+w2 - 1;
107 IDWTELEM b_0 = b[0];
108
109 i = 0;
110 __asm__ volatile(
111 "psllw $15, %%xmm7 \n\t"
112 "pcmpeqw %%xmm6, %%xmm6 \n\t"
113 "psrlw $13, %%xmm6 \n\t"
114 "paddw %%xmm7, %%xmm6 \n\t"
115 ::);
116 for(; i<w_l-15; i+=16){
117 __asm__ volatile(
118 "movdqu (%1), %%xmm0 \n\t"
119 "movdqu 16(%1), %%xmm4 \n\t"
120 "movdqu 2(%1), %%xmm1 \n\t"
121 "movdqu 18(%1), %%xmm5 \n\t" //FIXME try aligned reads and shifts
122 "paddw %%xmm6, %%xmm0 \n\t"
123 "paddw %%xmm6, %%xmm4 \n\t"
124 "paddw %%xmm7, %%xmm1 \n\t"
125 "paddw %%xmm7, %%xmm5 \n\t"
126 "pavgw %%xmm1, %%xmm0 \n\t"
127 "pavgw %%xmm5, %%xmm4 \n\t"
128 "psubw %%xmm7, %%xmm0 \n\t"
129 "psubw %%xmm7, %%xmm4 \n\t"
130 "psraw $1, %%xmm0 \n\t"
131 "psraw $1, %%xmm4 \n\t"
132 "movdqa (%0), %%xmm1 \n\t"
133 "movdqa 16(%0), %%xmm5 \n\t"
134 "paddw %%xmm1, %%xmm0 \n\t"
135 "paddw %%xmm5, %%xmm4 \n\t"
136 "psraw $2, %%xmm0 \n\t"
137 "psraw $2, %%xmm4 \n\t"
138 "paddw %%xmm1, %%xmm0 \n\t"
139 "paddw %%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-1 + 4 * b_0) >> W_BS);
148 }
149
150 { // Lift 3
151 IDWTELEM * const src = b+w2;
152
153 i = 0;
154 for(; (((x86_reg)&temp[i]) & 0x1F) && 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 2(%1), %%xmm2 \n\t"
160 "movdqu 18(%1), %%xmm6 \n\t"
161 "paddw (%1), %%xmm2 \n\t"
162 "paddw 16(%1), %%xmm6 \n\t"
163 "movdqu (%0), %%xmm0 \n\t"
164 "movdqu 16(%0), %%xmm4 \n\t"
165 "paddw %%xmm2, %%xmm0 \n\t"
166 "paddw %%xmm6, %%xmm4 \n\t"
167 "psraw $1, %%xmm2 \n\t"
168 "psraw $1, %%xmm6 \n\t"
169 "paddw %%xmm0, %%xmm2 \n\t"
170 "paddw %%xmm4, %%xmm6 \n\t"
171 "movdqa %%xmm2, (%2) \n\t"
172 "movdqa %%xmm6, 16(%2) \n\t"
173 :: "r"(&src[i]), "r"(&b[i]), "r"(&temp[i])
174 : "memory"
175 );
176 }
177 snow_horizontal_compose_lift_lead_out(i, temp, src, b, width, w_r, 1, -W_AM, W_AO+1, W_AS);
178 }
179
180 {
181 snow_interleave_line_header(&i, width, b, temp);
182
183 for (; (i & 0x3E) != 0x3E; i-=2){
184 b[i+1] = temp[i>>1];
185 b[i] = b[i>>1];
186 }
187 for (i-=62; i>=0; i-=64){
188 __asm__ volatile(
189 "movdqa (%1), %%xmm0 \n\t"
190 "movdqa 16(%1), %%xmm2 \n\t"
191 "movdqa 32(%1), %%xmm4 \n\t"
192 "movdqa 48(%1), %%xmm6 \n\t"
193 "movdqa (%1), %%xmm1 \n\t"
194 "movdqa 16(%1), %%xmm3 \n\t"
195 "movdqa 32(%1), %%xmm5 \n\t"
196 "movdqa 48(%1), %%xmm7 \n\t"
197 "punpcklwd (%2), %%xmm0 \n\t"
198 "punpcklwd 16(%2), %%xmm2 \n\t"
199 "punpcklwd 32(%2), %%xmm4 \n\t"
200 "punpcklwd 48(%2), %%xmm6 \n\t"
201 "movdqa %%xmm0, (%0) \n\t"
202 "movdqa %%xmm2, 32(%0) \n\t"
203 "movdqa %%xmm4, 64(%0) \n\t"
204 "movdqa %%xmm6, 96(%0) \n\t"
205 "punpckhwd (%2), %%xmm1 \n\t"
206 "punpckhwd 16(%2), %%xmm3 \n\t"
207 "punpckhwd 32(%2), %%xmm5 \n\t"
208 "punpckhwd 48(%2), %%xmm7 \n\t"
209 "movdqa %%xmm1, 16(%0) \n\t"
210 "movdqa %%xmm3, 48(%0) \n\t"
211 "movdqa %%xmm5, 80(%0) \n\t"
212 "movdqa %%xmm7, 112(%0) \n\t"
213 :: "r"(&(b)[i]), "r"(&(b)[i>>1]), "r"(&(temp)[i>>1])
214 : "memory"
215 );
216 }
217 }
218 }
219
ff_snow_horizontal_compose97i_mmx(IDWTELEM *b, IDWTELEM *temp, int width)220 static void ff_snow_horizontal_compose97i_mmx(IDWTELEM *b, IDWTELEM *temp, int width){
221 const int w2= (width+1)>>1;
222 const int w_l= (width>>1);
223 const int w_r= w2 - 1;
224 int i;
225
226 { // Lift 0
227 IDWTELEM * const ref = b + w2 - 1;
228
229 i = 1;
230 b[0] = b[0] - ((W_DM * 2 * ref[1]+W_DO)>>W_DS);
231 __asm__ volatile(
232 "pcmpeqw %%mm7, %%mm7 \n\t"
233 "pcmpeqw %%mm3, %%mm3 \n\t"
234 "psllw $1, %%mm3 \n\t"
235 "paddw %%mm7, %%mm3 \n\t"
236 "psllw $13, %%mm3 \n\t"
237 ::);
238 for(; i<w_l-7; i+=8){
239 __asm__ volatile(
240 "movq (%1), %%mm2 \n\t"
241 "movq 8(%1), %%mm6 \n\t"
242 "paddw 2(%1), %%mm2 \n\t"
243 "paddw 10(%1), %%mm6 \n\t"
244 "paddw %%mm7, %%mm2 \n\t"
245 "paddw %%mm7, %%mm6 \n\t"
246 "pmulhw %%mm3, %%mm2 \n\t"
247 "pmulhw %%mm3, %%mm6 \n\t"
248 "paddw (%0), %%mm2 \n\t"
249 "paddw 8(%0), %%mm6 \n\t"
250 "movq %%mm2, (%0) \n\t"
251 "movq %%mm6, 8(%0) \n\t"
252 :: "r"(&b[i]), "r"(&ref[i])
253 : "memory"
254 );
255 }
256 snow_horizontal_compose_lift_lead_out(i, b, b, ref, width, w_l, 0, W_DM, W_DO, W_DS);
257 }
258
259 { // Lift 1
260 IDWTELEM * const dst = b+w2;
261
262 i = 0;
263 for(; i<w_r-7; i+=8){
264 __asm__ volatile(
265 "movq (%1), %%mm2 \n\t"
266 "movq 8(%1), %%mm6 \n\t"
267 "paddw 2(%1), %%mm2 \n\t"
268 "paddw 10(%1), %%mm6 \n\t"
269 "movq (%0), %%mm0 \n\t"
270 "movq 8(%0), %%mm4 \n\t"
271 "psubw %%mm2, %%mm0 \n\t"
272 "psubw %%mm6, %%mm4 \n\t"
273 "movq %%mm0, (%0) \n\t"
274 "movq %%mm4, 8(%0) \n\t"
275 :: "r"(&dst[i]), "r"(&b[i])
276 : "memory"
277 );
278 }
279 snow_horizontal_compose_lift_lead_out(i, dst, dst, b, width, w_r, 1, W_CM, W_CO, W_CS);
280 }
281
282 { // Lift 2
283 IDWTELEM * const ref = b+w2 - 1;
284
285 i = 1;
286 b[0] = b[0] + (((2 * ref[1] + W_BO) + 4 * b[0]) >> W_BS);
287 __asm__ volatile(
288 "psllw $15, %%mm7 \n\t"
289 "pcmpeqw %%mm6, %%mm6 \n\t"
290 "psrlw $13, %%mm6 \n\t"
291 "paddw %%mm7, %%mm6 \n\t"
292 ::);
293 for(; i<w_l-7; i+=8){
294 __asm__ volatile(
295 "movq (%1), %%mm0 \n\t"
296 "movq 8(%1), %%mm4 \n\t"
297 "movq 2(%1), %%mm1 \n\t"
298 "movq 10(%1), %%mm5 \n\t"
299 "paddw %%mm6, %%mm0 \n\t"
300 "paddw %%mm6, %%mm4 \n\t"
301 "paddw %%mm7, %%mm1 \n\t"
302 "paddw %%mm7, %%mm5 \n\t"
303 "pavgw %%mm1, %%mm0 \n\t"
304 "pavgw %%mm5, %%mm4 \n\t"
305 "psubw %%mm7, %%mm0 \n\t"
306 "psubw %%mm7, %%mm4 \n\t"
307 "psraw $1, %%mm0 \n\t"
308 "psraw $1, %%mm4 \n\t"
309 "movq (%0), %%mm1 \n\t"
310 "movq 8(%0), %%mm5 \n\t"
311 "paddw %%mm1, %%mm0 \n\t"
312 "paddw %%mm5, %%mm4 \n\t"
313 "psraw $2, %%mm0 \n\t"
314 "psraw $2, %%mm4 \n\t"
315 "paddw %%mm1, %%mm0 \n\t"
316 "paddw %%mm5, %%mm4 \n\t"
317 "movq %%mm0, (%0) \n\t"
318 "movq %%mm4, 8(%0) \n\t"
319 :: "r"(&b[i]), "r"(&ref[i])
320 : "memory"
321 );
322 }
323 snow_horizontal_compose_liftS_lead_out(i, b, b, ref, width, w_l);
324 }
325
326 { // Lift 3
327 IDWTELEM * const src = b+w2;
328 i = 0;
329
330 for(; i<w_r-7; i+=8){
331 __asm__ volatile(
332 "movq 2(%1), %%mm2 \n\t"
333 "movq 10(%1), %%mm6 \n\t"
334 "paddw (%1), %%mm2 \n\t"
335 "paddw 8(%1), %%mm6 \n\t"
336 "movq (%0), %%mm0 \n\t"
337 "movq 8(%0), %%mm4 \n\t"
338 "paddw %%mm2, %%mm0 \n\t"
339 "paddw %%mm6, %%mm4 \n\t"
340 "psraw $1, %%mm2 \n\t"
341 "psraw $1, %%mm6 \n\t"
342 "paddw %%mm0, %%mm2 \n\t"
343 "paddw %%mm4, %%mm6 \n\t"
344 "movq %%mm2, (%2) \n\t"
345 "movq %%mm6, 8(%2) \n\t"
346 :: "r"(&src[i]), "r"(&b[i]), "r"(&temp[i])
347 : "memory"
348 );
349 }
350 snow_horizontal_compose_lift_lead_out(i, temp, src, b, width, w_r, 1, -W_AM, W_AO+1, W_AS);
351 }
352
353 {
354 snow_interleave_line_header(&i, width, b, temp);
355
356 for (; (i & 0x1E) != 0x1E; i-=2){
357 b[i+1] = temp[i>>1];
358 b[i] = b[i>>1];
359 }
360 for (i-=30; i>=0; i-=32){
361 __asm__ volatile(
362 "movq (%1), %%mm0 \n\t"
363 "movq 8(%1), %%mm2 \n\t"
364 "movq 16(%1), %%mm4 \n\t"
365 "movq 24(%1), %%mm6 \n\t"
366 "movq (%1), %%mm1 \n\t"
367 "movq 8(%1), %%mm3 \n\t"
368 "movq 16(%1), %%mm5 \n\t"
369 "movq 24(%1), %%mm7 \n\t"
370 "punpcklwd (%2), %%mm0 \n\t"
371 "punpcklwd 8(%2), %%mm2 \n\t"
372 "punpcklwd 16(%2), %%mm4 \n\t"
373 "punpcklwd 24(%2), %%mm6 \n\t"
374 "movq %%mm0, (%0) \n\t"
375 "movq %%mm2, 16(%0) \n\t"
376 "movq %%mm4, 32(%0) \n\t"
377 "movq %%mm6, 48(%0) \n\t"
378 "punpckhwd (%2), %%mm1 \n\t"
379 "punpckhwd 8(%2), %%mm3 \n\t"
380 "punpckhwd 16(%2), %%mm5 \n\t"
381 "punpckhwd 24(%2), %%mm7 \n\t"
382 "movq %%mm1, 8(%0) \n\t"
383 "movq %%mm3, 24(%0) \n\t"
384 "movq %%mm5, 40(%0) \n\t"
385 "movq %%mm7, 56(%0) \n\t"
386 :: "r"(&b[i]), "r"(&b[i>>1]), "r"(&temp[i>>1])
387 : "memory"
388 );
389 }
390 }
391 }
392
393 #if HAVE_7REGS
394 #define snow_vertical_compose_sse2_load_add(op,r,t0,t1,t2,t3)\
395 ""op" ("r",%%"FF_REG_d"), %%"t0" \n\t"\
396 ""op" 16("r",%%"FF_REG_d"), %%"t1" \n\t"\
397 ""op" 32("r",%%"FF_REG_d"), %%"t2" \n\t"\
398 ""op" 48("r",%%"FF_REG_d"), %%"t3" \n\t"
399
400 #define snow_vertical_compose_sse2_load(r,t0,t1,t2,t3)\
401 snow_vertical_compose_sse2_load_add("movdqa",r,t0,t1,t2,t3)
402
403 #define snow_vertical_compose_sse2_add(r,t0,t1,t2,t3)\
404 snow_vertical_compose_sse2_load_add("paddw",r,t0,t1,t2,t3)
405
406 #define snow_vertical_compose_r2r_sub(s0,s1,s2,s3,t0,t1,t2,t3)\
407 "psubw %%"s0", %%"t0" \n\t"\
408 "psubw %%"s1", %%"t1" \n\t"\
409 "psubw %%"s2", %%"t2" \n\t"\
410 "psubw %%"s3", %%"t3" \n\t"
411
412 #define snow_vertical_compose_sse2_store(w,s0,s1,s2,s3)\
413 "movdqa %%"s0", ("w",%%"FF_REG_d") \n\t"\
414 "movdqa %%"s1", 16("w",%%"FF_REG_d") \n\t"\
415 "movdqa %%"s2", 32("w",%%"FF_REG_d") \n\t"\
416 "movdqa %%"s3", 48("w",%%"FF_REG_d") \n\t"
417
418 #define snow_vertical_compose_sra(n,t0,t1,t2,t3)\
419 "psraw $"n", %%"t0" \n\t"\
420 "psraw $"n", %%"t1" \n\t"\
421 "psraw $"n", %%"t2" \n\t"\
422 "psraw $"n", %%"t3" \n\t"
423
424 #define snow_vertical_compose_r2r_add(s0,s1,s2,s3,t0,t1,t2,t3)\
425 "paddw %%"s0", %%"t0" \n\t"\
426 "paddw %%"s1", %%"t1" \n\t"\
427 "paddw %%"s2", %%"t2" \n\t"\
428 "paddw %%"s3", %%"t3" \n\t"
429
430 #define snow_vertical_compose_r2r_pmulhw(s0,s1,s2,s3,t0,t1,t2,t3)\
431 "pmulhw %%"s0", %%"t0" \n\t"\
432 "pmulhw %%"s1", %%"t1" \n\t"\
433 "pmulhw %%"s2", %%"t2" \n\t"\
434 "pmulhw %%"s3", %%"t3" \n\t"
435
436 #define snow_vertical_compose_sse2_move(s0,s1,s2,s3,t0,t1,t2,t3)\
437 "movdqa %%"s0", %%"t0" \n\t"\
438 "movdqa %%"s1", %%"t1" \n\t"\
439 "movdqa %%"s2", %%"t2" \n\t"\
440 "movdqa %%"s3", %%"t3" \n\t"
441
ff_snow_vertical_compose97i_sse2(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width)442 static void ff_snow_vertical_compose97i_sse2(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){
443 x86_reg i = width;
444
445 while(i & 0x1F)
446 {
447 i--;
448 b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;
449 b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS;
450 b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS;
451 b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
452 }
453 i+=i;
454
455 __asm__ volatile (
456 "jmp 2f \n\t"
457 "1: \n\t"
458 snow_vertical_compose_sse2_load("%4","xmm0","xmm2","xmm4","xmm6")
459 snow_vertical_compose_sse2_add("%6","xmm0","xmm2","xmm4","xmm6")
460
461
462 "pcmpeqw %%xmm0, %%xmm0 \n\t"
463 "pcmpeqw %%xmm2, %%xmm2 \n\t"
464 "paddw %%xmm2, %%xmm2 \n\t"
465 "paddw %%xmm0, %%xmm2 \n\t"
466 "psllw $13, %%xmm2 \n\t"
467 snow_vertical_compose_r2r_add("xmm0","xmm0","xmm0","xmm0","xmm1","xmm3","xmm5","xmm7")
468 snow_vertical_compose_r2r_pmulhw("xmm2","xmm2","xmm2","xmm2","xmm1","xmm3","xmm5","xmm7")
469 snow_vertical_compose_sse2_add("%5","xmm1","xmm3","xmm5","xmm7")
470 snow_vertical_compose_sse2_store("%5","xmm1","xmm3","xmm5","xmm7")
471 snow_vertical_compose_sse2_load("%4","xmm0","xmm2","xmm4","xmm6")
472 snow_vertical_compose_sse2_add("%3","xmm1","xmm3","xmm5","xmm7")
473 snow_vertical_compose_r2r_sub("xmm1","xmm3","xmm5","xmm7","xmm0","xmm2","xmm4","xmm6")
474 snow_vertical_compose_sse2_store("%4","xmm0","xmm2","xmm4","xmm6")
475
476 "pcmpeqw %%xmm7, %%xmm7 \n\t"
477 "pcmpeqw %%xmm5, %%xmm5 \n\t"
478 "psllw $15, %%xmm7 \n\t"
479 "psrlw $13, %%xmm5 \n\t"
480 "paddw %%xmm7, %%xmm5 \n\t"
481 snow_vertical_compose_r2r_add("xmm5","xmm5","xmm5","xmm5","xmm0","xmm2","xmm4","xmm6")
482 "movq (%2,%%"FF_REG_d"), %%xmm1 \n\t"
483 "movq 8(%2,%%"FF_REG_d"), %%xmm3 \n\t"
484 "paddw %%xmm7, %%xmm1 \n\t"
485 "paddw %%xmm7, %%xmm3 \n\t"
486 "pavgw %%xmm1, %%xmm0 \n\t"
487 "pavgw %%xmm3, %%xmm2 \n\t"
488 "movq 16(%2,%%"FF_REG_d"), %%xmm1 \n\t"
489 "movq 24(%2,%%"FF_REG_d"), %%xmm3 \n\t"
490 "paddw %%xmm7, %%xmm1 \n\t"
491 "paddw %%xmm7, %%xmm3 \n\t"
492 "pavgw %%xmm1, %%xmm4 \n\t"
493 "pavgw %%xmm3, %%xmm6 \n\t"
494 snow_vertical_compose_r2r_sub("xmm7","xmm7","xmm7","xmm7","xmm0","xmm2","xmm4","xmm6")
495 snow_vertical_compose_sra("1","xmm0","xmm2","xmm4","xmm6")
496 snow_vertical_compose_sse2_add("%3","xmm0","xmm2","xmm4","xmm6")
497
498 snow_vertical_compose_sra("2","xmm0","xmm2","xmm4","xmm6")
499 snow_vertical_compose_sse2_add("%3","xmm0","xmm2","xmm4","xmm6")
500 snow_vertical_compose_sse2_store("%3","xmm0","xmm2","xmm4","xmm6")
501 snow_vertical_compose_sse2_add("%1","xmm0","xmm2","xmm4","xmm6")
502 snow_vertical_compose_sse2_move("xmm0","xmm2","xmm4","xmm6","xmm1","xmm3","xmm5","xmm7")
503 snow_vertical_compose_sra("1","xmm0","xmm2","xmm4","xmm6")
504 snow_vertical_compose_r2r_add("xmm1","xmm3","xmm5","xmm7","xmm0","xmm2","xmm4","xmm6")
505 snow_vertical_compose_sse2_add("%2","xmm0","xmm2","xmm4","xmm6")
506 snow_vertical_compose_sse2_store("%2","xmm0","xmm2","xmm4","xmm6")
507
508 "2: \n\t"
509 "sub $64, %%"FF_REG_d" \n\t"
510 "jge 1b \n\t"
511 :"+d"(i)
512 :"r"(b0),"r"(b1),"r"(b2),"r"(b3),"r"(b4),"r"(b5));
513 }
514
515 #define snow_vertical_compose_mmx_load_add(op,r,t0,t1,t2,t3)\
516 ""op" ("r",%%"FF_REG_d"), %%"t0" \n\t"\
517 ""op" 8("r",%%"FF_REG_d"), %%"t1" \n\t"\
518 ""op" 16("r",%%"FF_REG_d"), %%"t2" \n\t"\
519 ""op" 24("r",%%"FF_REG_d"), %%"t3" \n\t"
520
521 #define snow_vertical_compose_mmx_load(r,t0,t1,t2,t3)\
522 snow_vertical_compose_mmx_load_add("movq",r,t0,t1,t2,t3)
523
524 #define snow_vertical_compose_mmx_add(r,t0,t1,t2,t3)\
525 snow_vertical_compose_mmx_load_add("paddw",r,t0,t1,t2,t3)
526
527 #define snow_vertical_compose_mmx_store(w,s0,s1,s2,s3)\
528 "movq %%"s0", ("w",%%"FF_REG_d") \n\t"\
529 "movq %%"s1", 8("w",%%"FF_REG_d") \n\t"\
530 "movq %%"s2", 16("w",%%"FF_REG_d") \n\t"\
531 "movq %%"s3", 24("w",%%"FF_REG_d") \n\t"
532
533 #define snow_vertical_compose_mmx_move(s0,s1,s2,s3,t0,t1,t2,t3)\
534 "movq %%"s0", %%"t0" \n\t"\
535 "movq %%"s1", %%"t1" \n\t"\
536 "movq %%"s2", %%"t2" \n\t"\
537 "movq %%"s3", %%"t3" \n\t"
538
539
ff_snow_vertical_compose97i_mmx(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width)540 static void ff_snow_vertical_compose97i_mmx(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){
541 x86_reg i = width;
542 while(i & 15)
543 {
544 i--;
545 b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;
546 b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS;
547 b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS;
548 b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
549 }
550 i+=i;
551 __asm__ volatile(
552 "jmp 2f \n\t"
553 "1: \n\t"
554
555 snow_vertical_compose_mmx_load("%4","mm1","mm3","mm5","mm7")
556 snow_vertical_compose_mmx_add("%6","mm1","mm3","mm5","mm7")
557 "pcmpeqw %%mm0, %%mm0 \n\t"
558 "pcmpeqw %%mm2, %%mm2 \n\t"
559 "paddw %%mm2, %%mm2 \n\t"
560 "paddw %%mm0, %%mm2 \n\t"
561 "psllw $13, %%mm2 \n\t"
562 snow_vertical_compose_r2r_add("mm0","mm0","mm0","mm0","mm1","mm3","mm5","mm7")
563 snow_vertical_compose_r2r_pmulhw("mm2","mm2","mm2","mm2","mm1","mm3","mm5","mm7")
564 snow_vertical_compose_mmx_add("%5","mm1","mm3","mm5","mm7")
565 snow_vertical_compose_mmx_store("%5","mm1","mm3","mm5","mm7")
566 snow_vertical_compose_mmx_load("%4","mm0","mm2","mm4","mm6")
567 snow_vertical_compose_mmx_add("%3","mm1","mm3","mm5","mm7")
568 snow_vertical_compose_r2r_sub("mm1","mm3","mm5","mm7","mm0","mm2","mm4","mm6")
569 snow_vertical_compose_mmx_store("%4","mm0","mm2","mm4","mm6")
570 "pcmpeqw %%mm7, %%mm7 \n\t"
571 "pcmpeqw %%mm5, %%mm5 \n\t"
572 "psllw $15, %%mm7 \n\t"
573 "psrlw $13, %%mm5 \n\t"
574 "paddw %%mm7, %%mm5 \n\t"
575 snow_vertical_compose_r2r_add("mm5","mm5","mm5","mm5","mm0","mm2","mm4","mm6")
576 "movq (%2,%%"FF_REG_d"), %%mm1 \n\t"
577 "movq 8(%2,%%"FF_REG_d"), %%mm3 \n\t"
578 "paddw %%mm7, %%mm1 \n\t"
579 "paddw %%mm7, %%mm3 \n\t"
580 "pavgw %%mm1, %%mm0 \n\t"
581 "pavgw %%mm3, %%mm2 \n\t"
582 "movq 16(%2,%%"FF_REG_d"), %%mm1 \n\t"
583 "movq 24(%2,%%"FF_REG_d"), %%mm3 \n\t"
584 "paddw %%mm7, %%mm1 \n\t"
585 "paddw %%mm7, %%mm3 \n\t"
586 "pavgw %%mm1, %%mm4 \n\t"
587 "pavgw %%mm3, %%mm6 \n\t"
588 snow_vertical_compose_r2r_sub("mm7","mm7","mm7","mm7","mm0","mm2","mm4","mm6")
589 snow_vertical_compose_sra("1","mm0","mm2","mm4","mm6")
590 snow_vertical_compose_mmx_add("%3","mm0","mm2","mm4","mm6")
591
592 snow_vertical_compose_sra("2","mm0","mm2","mm4","mm6")
593 snow_vertical_compose_mmx_add("%3","mm0","mm2","mm4","mm6")
594 snow_vertical_compose_mmx_store("%3","mm0","mm2","mm4","mm6")
595 snow_vertical_compose_mmx_add("%1","mm0","mm2","mm4","mm6")
596 snow_vertical_compose_mmx_move("mm0","mm2","mm4","mm6","mm1","mm3","mm5","mm7")
597 snow_vertical_compose_sra("1","mm0","mm2","mm4","mm6")
598 snow_vertical_compose_r2r_add("mm1","mm3","mm5","mm7","mm0","mm2","mm4","mm6")
599 snow_vertical_compose_mmx_add("%2","mm0","mm2","mm4","mm6")
600 snow_vertical_compose_mmx_store("%2","mm0","mm2","mm4","mm6")
601
602 "2: \n\t"
603 "sub $32, %%"FF_REG_d" \n\t"
604 "jge 1b \n\t"
605 :"+d"(i)
606 :"r"(b0),"r"(b1),"r"(b2),"r"(b3),"r"(b4),"r"(b5));
607 }
608 #endif //HAVE_7REGS
609
610 #if HAVE_6REGS
611 #define snow_inner_add_yblock_sse2_header \
612 IDWTELEM * * dst_array = sb->line + src_y;\
613 x86_reg tmp;\
614 __asm__ volatile(\
615 "mov %7, %%"FF_REG_c" \n\t"\
616 "mov %6, %2 \n\t"\
617 "mov %4, %%"FF_REG_S" \n\t"\
618 "pxor %%xmm7, %%xmm7 \n\t" /* 0 */\
619 "pcmpeqd %%xmm3, %%xmm3 \n\t"\
620 "psllw $15, %%xmm3 \n\t"\
621 "psrlw $12, %%xmm3 \n\t" /* FRAC_BITS >> 1 */\
622 "1: \n\t"\
623 "mov %1, %%"FF_REG_D" \n\t"\
624 "mov (%%"FF_REG_D"), %%"FF_REG_D" \n\t"\
625 "add %3, %%"FF_REG_D" \n\t"
626
627 #define snow_inner_add_yblock_sse2_start_8(out_reg1, out_reg2, ptr_offset, s_offset)\
628 "mov "FF_PTR_SIZE"*"ptr_offset"(%%"FF_REG_a"), %%"FF_REG_d"; \n\t"\
629 "movq (%%"FF_REG_d"), %%"out_reg1" \n\t"\
630 "movq (%%"FF_REG_d", %%"FF_REG_c"), %%"out_reg2" \n\t"\
631 "punpcklbw %%xmm7, %%"out_reg1" \n\t"\
632 "punpcklbw %%xmm7, %%"out_reg2" \n\t"\
633 "movq "s_offset"(%%"FF_REG_S"), %%xmm0 \n\t"\
634 "movq "s_offset"+16(%%"FF_REG_S"), %%xmm4 \n\t"\
635 "punpcklbw %%xmm7, %%xmm0 \n\t"\
636 "punpcklbw %%xmm7, %%xmm4 \n\t"\
637 "pmullw %%xmm0, %%"out_reg1" \n\t"\
638 "pmullw %%xmm4, %%"out_reg2" \n\t"
639
640 #define snow_inner_add_yblock_sse2_start_16(out_reg1, out_reg2, ptr_offset, s_offset)\
641 "mov "FF_PTR_SIZE"*"ptr_offset"(%%"FF_REG_a"), %%"FF_REG_d"; \n\t"\
642 "movq (%%"FF_REG_d"), %%"out_reg1" \n\t"\
643 "movq 8(%%"FF_REG_d"), %%"out_reg2" \n\t"\
644 "punpcklbw %%xmm7, %%"out_reg1" \n\t"\
645 "punpcklbw %%xmm7, %%"out_reg2" \n\t"\
646 "movq "s_offset"(%%"FF_REG_S"), %%xmm0 \n\t"\
647 "movq "s_offset"+8(%%"FF_REG_S"), %%xmm4 \n\t"\
648 "punpcklbw %%xmm7, %%xmm0 \n\t"\
649 "punpcklbw %%xmm7, %%xmm4 \n\t"\
650 "pmullw %%xmm0, %%"out_reg1" \n\t"\
651 "pmullw %%xmm4, %%"out_reg2" \n\t"
652
653 #define snow_inner_add_yblock_sse2_accum_8(ptr_offset, s_offset) \
654 snow_inner_add_yblock_sse2_start_8("xmm2", "xmm6", ptr_offset, s_offset)\
655 "paddusw %%xmm2, %%xmm1 \n\t"\
656 "paddusw %%xmm6, %%xmm5 \n\t"
657
658 #define snow_inner_add_yblock_sse2_accum_16(ptr_offset, s_offset) \
659 snow_inner_add_yblock_sse2_start_16("xmm2", "xmm6", ptr_offset, s_offset)\
660 "paddusw %%xmm2, %%xmm1 \n\t"\
661 "paddusw %%xmm6, %%xmm5 \n\t"
662
663 #define snow_inner_add_yblock_sse2_end_common1\
664 "add $32, %%"FF_REG_S" \n\t"\
665 "add %%"FF_REG_c", %0 \n\t"\
666 "add %%"FF_REG_c", "FF_PTR_SIZE"*3(%%"FF_REG_a"); \n\t"\
667 "add %%"FF_REG_c", "FF_PTR_SIZE"*2(%%"FF_REG_a"); \n\t"\
668 "add %%"FF_REG_c", "FF_PTR_SIZE"*1(%%"FF_REG_a"); \n\t"\
669 "add %%"FF_REG_c", (%%"FF_REG_a") \n\t"
670
671 #define snow_inner_add_yblock_sse2_end_common2\
672 "jnz 1b \n\t"\
673 :"+m"(dst8),"+m"(dst_array),"=&r"(tmp)\
674 :\
675 "rm"((x86_reg)(src_x<<1)),"m"(obmc),"a"(block),"m"(b_h),"m"(src_stride):\
676 XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm4", "%xmm5", "%xmm6", "%xmm7", )\
677 "%"FF_REG_c"","%"FF_REG_S"","%"FF_REG_D"","%"FF_REG_d"");
678
679 #define snow_inner_add_yblock_sse2_end_8\
680 "sal $1, %%"FF_REG_c" \n\t"\
681 "add"FF_OPSIZE" $"FF_PTR_SIZE"*2, %1 \n\t"\
682 snow_inner_add_yblock_sse2_end_common1\
683 "sar $1, %%"FF_REG_c" \n\t"\
684 "sub $2, %2 \n\t"\
685 snow_inner_add_yblock_sse2_end_common2
686
687 #define snow_inner_add_yblock_sse2_end_16\
688 "add"FF_OPSIZE" $"FF_PTR_SIZE"*1, %1 \n\t"\
689 snow_inner_add_yblock_sse2_end_common1\
690 "dec %2 \n\t"\
691 snow_inner_add_yblock_sse2_end_common2
692
inner_add_yblock_bw_8_obmc_16_bh_even_sse2(const uint8_t *obmc, const x86_reg obmc_stride, uint8_t * * block, int b_w, x86_reg b_h, int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8)693 static void inner_add_yblock_bw_8_obmc_16_bh_even_sse2(const uint8_t *obmc, const x86_reg obmc_stride, uint8_t * * block, int b_w, x86_reg b_h,
694 int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8){
695 snow_inner_add_yblock_sse2_header
696 snow_inner_add_yblock_sse2_start_8("xmm1", "xmm5", "3", "0")
697 snow_inner_add_yblock_sse2_accum_8("2", "8")
698 snow_inner_add_yblock_sse2_accum_8("1", "128")
699 snow_inner_add_yblock_sse2_accum_8("0", "136")
700
701 "mov %0, %%"FF_REG_d" \n\t"
702 "movdqa (%%"FF_REG_D"), %%xmm0 \n\t"
703 "movdqa %%xmm1, %%xmm2 \n\t"
704
705 "punpckhwd %%xmm7, %%xmm1 \n\t"
706 "punpcklwd %%xmm7, %%xmm2 \n\t"
707 "paddd %%xmm2, %%xmm0 \n\t"
708 "movdqa 16(%%"FF_REG_D"), %%xmm2\n\t"
709 "paddd %%xmm1, %%xmm2 \n\t"
710 "paddd %%xmm3, %%xmm0 \n\t"
711 "paddd %%xmm3, %%xmm2 \n\t"
712
713 "mov %1, %%"FF_REG_D" \n\t"
714 "mov "FF_PTR_SIZE"(%%"FF_REG_D"), %%"FF_REG_D"; \n\t"
715 "add %3, %%"FF_REG_D" \n\t"
716
717 "movdqa (%%"FF_REG_D"), %%xmm4 \n\t"
718 "movdqa %%xmm5, %%xmm6 \n\t"
719 "punpckhwd %%xmm7, %%xmm5 \n\t"
720 "punpcklwd %%xmm7, %%xmm6 \n\t"
721 "paddd %%xmm6, %%xmm4 \n\t"
722 "movdqa 16(%%"FF_REG_D"), %%xmm6\n\t"
723 "paddd %%xmm5, %%xmm6 \n\t"
724 "paddd %%xmm3, %%xmm4 \n\t"
725 "paddd %%xmm3, %%xmm6 \n\t"
726
727 "psrad $8, %%xmm0 \n\t" /* FRAC_BITS. */
728 "psrad $8, %%xmm2 \n\t" /* FRAC_BITS. */
729 "packssdw %%xmm2, %%xmm0 \n\t"
730 "packuswb %%xmm7, %%xmm0 \n\t"
731 "movq %%xmm0, (%%"FF_REG_d") \n\t"
732
733 "psrad $8, %%xmm4 \n\t" /* FRAC_BITS. */
734 "psrad $8, %%xmm6 \n\t" /* FRAC_BITS. */
735 "packssdw %%xmm6, %%xmm4 \n\t"
736 "packuswb %%xmm7, %%xmm4 \n\t"
737 "movq %%xmm4, (%%"FF_REG_d",%%"FF_REG_c"); \n\t"
738 snow_inner_add_yblock_sse2_end_8
739 }
740
inner_add_yblock_bw_16_obmc_32_sse2(const uint8_t *obmc, const x86_reg obmc_stride, uint8_t * * block, int b_w, x86_reg b_h, int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8)741 static void inner_add_yblock_bw_16_obmc_32_sse2(const uint8_t *obmc, const x86_reg obmc_stride, uint8_t * * block, int b_w, x86_reg b_h,
742 int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8){
743 snow_inner_add_yblock_sse2_header
744 snow_inner_add_yblock_sse2_start_16("xmm1", "xmm5", "3", "0")
745 snow_inner_add_yblock_sse2_accum_16("2", "16")
746 snow_inner_add_yblock_sse2_accum_16("1", "512")
747 snow_inner_add_yblock_sse2_accum_16("0", "528")
748
749 "mov %0, %%"FF_REG_d" \n\t"
750 "psrlw $4, %%xmm1 \n\t"
751 "psrlw $4, %%xmm5 \n\t"
752 "paddw (%%"FF_REG_D"), %%xmm1 \n\t"
753 "paddw 16(%%"FF_REG_D"), %%xmm5 \n\t"
754 "paddw %%xmm3, %%xmm1 \n\t"
755 "paddw %%xmm3, %%xmm5 \n\t"
756 "psraw $4, %%xmm1 \n\t" /* FRAC_BITS. */
757 "psraw $4, %%xmm5 \n\t" /* FRAC_BITS. */
758 "packuswb %%xmm5, %%xmm1 \n\t"
759
760 "movdqu %%xmm1, (%%"FF_REG_d") \n\t"
761
762 snow_inner_add_yblock_sse2_end_16
763 }
764
765 #define snow_inner_add_yblock_mmx_header \
766 IDWTELEM * * dst_array = sb->line + src_y;\
767 x86_reg tmp;\
768 __asm__ volatile(\
769 "mov %7, %%"FF_REG_c" \n\t"\
770 "mov %6, %2 \n\t"\
771 "mov %4, %%"FF_REG_S" \n\t"\
772 "pxor %%mm7, %%mm7 \n\t" /* 0 */\
773 "pcmpeqd %%mm3, %%mm3 \n\t"\
774 "psllw $15, %%mm3 \n\t"\
775 "psrlw $12, %%mm3 \n\t" /* FRAC_BITS >> 1 */\
776 "1: \n\t"\
777 "mov %1, %%"FF_REG_D" \n\t"\
778 "mov (%%"FF_REG_D"), %%"FF_REG_D" \n\t"\
779 "add %3, %%"FF_REG_D" \n\t"
780
781 #define snow_inner_add_yblock_mmx_start(out_reg1, out_reg2, ptr_offset, s_offset, d_offset)\
782 "mov "FF_PTR_SIZE"*"ptr_offset"(%%"FF_REG_a"), %%"FF_REG_d"; \n\t"\
783 "movd "d_offset"(%%"FF_REG_d"), %%"out_reg1" \n\t"\
784 "movd "d_offset"+4(%%"FF_REG_d"), %%"out_reg2" \n\t"\
785 "punpcklbw %%mm7, %%"out_reg1" \n\t"\
786 "punpcklbw %%mm7, %%"out_reg2" \n\t"\
787 "movd "s_offset"(%%"FF_REG_S"), %%mm0 \n\t"\
788 "movd "s_offset"+4(%%"FF_REG_S"), %%mm4 \n\t"\
789 "punpcklbw %%mm7, %%mm0 \n\t"\
790 "punpcklbw %%mm7, %%mm4 \n\t"\
791 "pmullw %%mm0, %%"out_reg1" \n\t"\
792 "pmullw %%mm4, %%"out_reg2" \n\t"
793
794 #define snow_inner_add_yblock_mmx_accum(ptr_offset, s_offset, d_offset) \
795 snow_inner_add_yblock_mmx_start("mm2", "mm6", ptr_offset, s_offset, d_offset)\
796 "paddusw %%mm2, %%mm1 \n\t"\
797 "paddusw %%mm6, %%mm5 \n\t"
798
799 #define snow_inner_add_yblock_mmx_mix(read_offset, write_offset)\
800 "mov %0, %%"FF_REG_d" \n\t"\
801 "psrlw $4, %%mm1 \n\t"\
802 "psrlw $4, %%mm5 \n\t"\
803 "paddw "read_offset"(%%"FF_REG_D"), %%mm1 \n\t"\
804 "paddw "read_offset"+8(%%"FF_REG_D"), %%mm5 \n\t"\
805 "paddw %%mm3, %%mm1 \n\t"\
806 "paddw %%mm3, %%mm5 \n\t"\
807 "psraw $4, %%mm1 \n\t"\
808 "psraw $4, %%mm5 \n\t"\
809 "packuswb %%mm5, %%mm1 \n\t"\
810 "movq %%mm1, "write_offset"(%%"FF_REG_d") \n\t"
811
812 #define snow_inner_add_yblock_mmx_end(s_step)\
813 "add $"s_step", %%"FF_REG_S" \n\t"\
814 "add %%"FF_REG_c", "FF_PTR_SIZE"*3(%%"FF_REG_a"); \n\t"\
815 "add %%"FF_REG_c", "FF_PTR_SIZE"*2(%%"FF_REG_a"); \n\t"\
816 "add %%"FF_REG_c", "FF_PTR_SIZE"*1(%%"FF_REG_a"); \n\t"\
817 "add %%"FF_REG_c", (%%"FF_REG_a") \n\t"\
818 "add"FF_OPSIZE " $"FF_PTR_SIZE"*1, %1 \n\t"\
819 "add %%"FF_REG_c", %0 \n\t"\
820 "dec %2 \n\t"\
821 "jnz 1b \n\t"\
822 :"+m"(dst8),"+m"(dst_array),"=&r"(tmp)\
823 :\
824 "rm"((x86_reg)(src_x<<1)),"m"(obmc),"a"(block),"m"(b_h),"m"(src_stride):\
825 "%"FF_REG_c"","%"FF_REG_S"","%"FF_REG_D"","%"FF_REG_d"");
826
inner_add_yblock_bw_8_obmc_16_mmx(const uint8_t *obmc, const x86_reg obmc_stride, uint8_t * * block, int b_w, x86_reg b_h, int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8)827 static void inner_add_yblock_bw_8_obmc_16_mmx(const uint8_t *obmc, const x86_reg obmc_stride, uint8_t * * block, int b_w, x86_reg b_h,
828 int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8){
829 snow_inner_add_yblock_mmx_header
830 snow_inner_add_yblock_mmx_start("mm1", "mm5", "3", "0", "0")
831 snow_inner_add_yblock_mmx_accum("2", "8", "0")
832 snow_inner_add_yblock_mmx_accum("1", "128", "0")
833 snow_inner_add_yblock_mmx_accum("0", "136", "0")
834 snow_inner_add_yblock_mmx_mix("0", "0")
835 snow_inner_add_yblock_mmx_end("16")
836 }
837
inner_add_yblock_bw_16_obmc_32_mmx(const uint8_t *obmc, const x86_reg obmc_stride, uint8_t * * block, int b_w, x86_reg b_h, int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8)838 static void inner_add_yblock_bw_16_obmc_32_mmx(const uint8_t *obmc, const x86_reg obmc_stride, uint8_t * * block, int b_w, x86_reg b_h,
839 int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8){
840 snow_inner_add_yblock_mmx_header
841 snow_inner_add_yblock_mmx_start("mm1", "mm5", "3", "0", "0")
842 snow_inner_add_yblock_mmx_accum("2", "16", "0")
843 snow_inner_add_yblock_mmx_accum("1", "512", "0")
844 snow_inner_add_yblock_mmx_accum("0", "528", "0")
845 snow_inner_add_yblock_mmx_mix("0", "0")
846
847 snow_inner_add_yblock_mmx_start("mm1", "mm5", "3", "8", "8")
848 snow_inner_add_yblock_mmx_accum("2", "24", "8")
849 snow_inner_add_yblock_mmx_accum("1", "520", "8")
850 snow_inner_add_yblock_mmx_accum("0", "536", "8")
851 snow_inner_add_yblock_mmx_mix("16", "8")
852 snow_inner_add_yblock_mmx_end("32")
853 }
854
ff_snow_inner_add_yblock_sse2(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8)855 static void ff_snow_inner_add_yblock_sse2(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
856 int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){
857
858 if (b_w == 16)
859 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);
860 else if (b_w == 8 && obmc_stride == 16) {
861 if (!(b_h & 1))
862 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);
863 else
864 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);
865 } else
866 ff_snow_inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
867 }
868
ff_snow_inner_add_yblock_mmx(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8)869 static void ff_snow_inner_add_yblock_mmx(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
870 int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){
871 if (b_w == 16)
872 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);
873 else if (b_w == 8 && obmc_stride == 16)
874 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);
875 else
876 ff_snow_inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
877 }
878 #endif /* HAVE_6REGS */
879
880 #endif /* HAVE_INLINE_ASM */
881
ff_dwt_init_x86(SnowDWTContext *c)882 av_cold void ff_dwt_init_x86(SnowDWTContext *c)
883 {
884 #if HAVE_INLINE_ASM
885 int mm_flags = av_get_cpu_flags();
886
887 if (mm_flags & AV_CPU_FLAG_MMX) {
888 if(mm_flags & AV_CPU_FLAG_SSE2 & 0){
889 c->horizontal_compose97i = ff_snow_horizontal_compose97i_sse2;
890 #if HAVE_7REGS
891 c->vertical_compose97i = ff_snow_vertical_compose97i_sse2;
892 #endif
893 #if HAVE_6REGS
894 c->inner_add_yblock = ff_snow_inner_add_yblock_sse2;
895 #endif
896 }
897 else{
898 if (mm_flags & AV_CPU_FLAG_MMXEXT) {
899 c->horizontal_compose97i = ff_snow_horizontal_compose97i_mmx;
900 #if HAVE_7REGS
901 c->vertical_compose97i = ff_snow_vertical_compose97i_mmx;
902 #endif
903 }
904 #if HAVE_6REGS
905 c->inner_add_yblock = ff_snow_inner_add_yblock_mmx;
906 #endif
907 }
908 }
909 #endif /* HAVE_INLINE_ASM */
910 }
911