|
0
|
1 /*
|
|
|
2 * MMX optimized DSP utils
|
|
|
3 * Copyright (c) 2000, 2001 Gerard Lantau.
|
|
|
4 *
|
|
|
5 * This program is free software; you can redistribute it and/or modify
|
|
|
6 * it under the terms of the GNU General Public License as published by
|
|
|
7 * the Free Software Foundation; either version 2 of the License, or
|
|
|
8 * (at your option) any later version.
|
|
|
9 *
|
|
|
10 * This program is distributed in the hope that it will be useful,
|
|
|
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
13 * GNU General Public License for more details.
|
|
|
14 *
|
|
|
15 * You should have received a copy of the GNU General Public License
|
|
|
16 * along with this program; if not, write to the Free Software
|
|
|
17 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
|
|
|
18 *
|
|
|
19 * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
|
|
|
20 */
|
|
|
21
|
|
|
22 #include "../dsputil.h"
|
|
|
23
|
|
5
|
24 int mm_flags; /* multimedia extension flags */
|
|
|
25
|
|
0
|
26 int pix_abs16x16_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h);
|
|
|
27 int pix_abs16x16_sse(UINT8 *blk1, UINT8 *blk2, int lx, int h);
|
|
|
28 int pix_abs16x16_x2_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h);
|
|
|
29 int pix_abs16x16_y2_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h);
|
|
|
30 int pix_abs16x16_xy2_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h);
|
|
|
31
|
|
|
32 /* pixel operations */
|
|
|
33 static const unsigned short mm_wone[4] __attribute__ ((aligned(8))) = { 0x1, 0x1, 0x1, 0x1 };
|
|
|
34 static const unsigned short mm_wtwo[4] __attribute__ ((aligned(8))) = { 0x2, 0x2, 0x2, 0x2 };
|
|
|
35
|
|
|
36 /***********************************/
|
|
|
37 /* 3Dnow specific */
|
|
|
38
|
|
|
39 #define DEF(x) x ## _3dnow
|
|
|
40 /* for Athlons PAVGUSB is prefered */
|
|
|
41 #define PAVGB "pavgusb"
|
|
|
42
|
|
|
43 #include "dsputil_mmx_avg.h"
|
|
|
44
|
|
|
45 #undef DEF
|
|
|
46 #undef PAVGB
|
|
|
47
|
|
|
48 /***********************************/
|
|
|
49 /* MMX2 specific */
|
|
|
50
|
|
|
51 #define DEF(x) x ## _sse
|
|
|
52
|
|
|
53 /* Introduced only in MMX2 set */
|
|
|
54 #define PAVGB "pavgb"
|
|
|
55
|
|
|
56 #include "dsputil_mmx_avg.h"
|
|
|
57
|
|
|
58 #undef DEF
|
|
|
59 #undef PAVGB
|
|
|
60
|
|
|
61 /***********************************/
|
|
|
62 /* standard MMX */
|
|
|
63
|
|
|
64 static void get_pixels_mmx(DCTELEM *block, const UINT8 *pixels, int line_size)
|
|
|
65 {
|
|
|
66 DCTELEM *p;
|
|
|
67 const UINT8 *pix;
|
|
|
68 int i;
|
|
|
69
|
|
|
70 /* read the pixels */
|
|
|
71 p = block;
|
|
|
72 pix = pixels;
|
|
|
73 __asm __volatile("pxor %%mm7, %%mm7":::"memory");
|
|
|
74 for(i=0;i<4;i++) {
|
|
|
75 __asm __volatile(
|
|
|
76 "movq %1, %%mm0\n\t"
|
|
|
77 "movq %2, %%mm1\n\t"
|
|
|
78 "movq %%mm0, %%mm2\n\t"
|
|
|
79 "movq %%mm1, %%mm3\n\t"
|
|
|
80 "punpcklbw %%mm7, %%mm0\n\t"
|
|
|
81 "punpckhbw %%mm7, %%mm2\n\t"
|
|
|
82 "punpcklbw %%mm7, %%mm1\n\t"
|
|
|
83 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
84 "movq %%mm0, %0\n\t"
|
|
|
85 "movq %%mm2, 8%0\n\t"
|
|
|
86 "movq %%mm1, 16%0\n\t"
|
|
|
87 "movq %%mm3, 24%0\n\t"
|
|
|
88 :"=m"(*p)
|
|
|
89 :"m"(*pix), "m"(*(pix+line_size))
|
|
|
90 :"memory");
|
|
|
91 pix += line_size*2;
|
|
|
92 p += 16;
|
|
|
93 }
|
|
|
94 }
|
|
|
95
|
|
|
96 static void put_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size)
|
|
|
97 {
|
|
|
98 const DCTELEM *p;
|
|
|
99 UINT8 *pix;
|
|
|
100 int i;
|
|
|
101
|
|
|
102 /* read the pixels */
|
|
|
103 p = block;
|
|
|
104 pix = pixels;
|
|
|
105 for(i=0;i<2;i++) {
|
|
|
106 __asm __volatile(
|
|
|
107 "movq %4, %%mm0\n\t"
|
|
|
108 "movq 8%4, %%mm1\n\t"
|
|
|
109 "movq 16%4, %%mm2\n\t"
|
|
|
110 "movq 24%4, %%mm3\n\t"
|
|
|
111 "movq 32%4, %%mm4\n\t"
|
|
|
112 "movq 40%4, %%mm5\n\t"
|
|
|
113 "movq 48%4, %%mm6\n\t"
|
|
|
114 "movq 56%4, %%mm7\n\t"
|
|
|
115 "packuswb %%mm1, %%mm0\n\t"
|
|
|
116 "packuswb %%mm3, %%mm2\n\t"
|
|
|
117 "packuswb %%mm5, %%mm4\n\t"
|
|
|
118 "packuswb %%mm7, %%mm6\n\t"
|
|
|
119 "movq %%mm0, %0\n\t"
|
|
|
120 "movq %%mm2, %1\n\t"
|
|
|
121 "movq %%mm4, %2\n\t"
|
|
|
122 "movq %%mm6, %3\n\t"
|
|
|
123 :"=m"(*pix), "=m"(*(pix+line_size))
|
|
|
124 ,"=m"(*(pix+line_size*2)), "=m"(*(pix+line_size*3))
|
|
|
125 :"m"(*p)
|
|
|
126 :"memory");
|
|
|
127 pix += line_size*4;
|
|
|
128 p += 32;
|
|
|
129 }
|
|
|
130 }
|
|
|
131
|
|
|
132 static void add_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size)
|
|
|
133 {
|
|
|
134 const DCTELEM *p;
|
|
|
135 UINT8 *pix;
|
|
|
136 int i;
|
|
|
137
|
|
|
138 /* read the pixels */
|
|
|
139 p = block;
|
|
|
140 pix = pixels;
|
|
|
141 __asm __volatile("pxor %%mm7, %%mm7":::"memory");
|
|
|
142 for(i=0;i<4;i++) {
|
|
|
143 __asm __volatile(
|
|
|
144 "movq %2, %%mm0\n\t"
|
|
|
145 "movq 8%2, %%mm1\n\t"
|
|
|
146 "movq 16%2, %%mm2\n\t"
|
|
|
147 "movq 24%2, %%mm3\n\t"
|
|
|
148 "movq %0, %%mm4\n\t"
|
|
|
149 "movq %1, %%mm6\n\t"
|
|
|
150 "movq %%mm4, %%mm5\n\t"
|
|
|
151 "punpcklbw %%mm7, %%mm4\n\t"
|
|
|
152 "punpckhbw %%mm7, %%mm5\n\t"
|
|
|
153 "paddsw %%mm4, %%mm0\n\t"
|
|
|
154 "paddsw %%mm5, %%mm1\n\t"
|
|
|
155 "movq %%mm6, %%mm5\n\t"
|
|
|
156 "punpcklbw %%mm7, %%mm6\n\t"
|
|
|
157 "punpckhbw %%mm7, %%mm5\n\t"
|
|
|
158 "paddsw %%mm6, %%mm2\n\t"
|
|
|
159 "paddsw %%mm5, %%mm3\n\t"
|
|
|
160 "packuswb %%mm1, %%mm0\n\t"
|
|
|
161 "packuswb %%mm3, %%mm2\n\t"
|
|
|
162 "movq %%mm0, %0\n\t"
|
|
|
163 "movq %%mm2, %1\n\t"
|
|
|
164 :"=m"(*pix), "=m"(*(pix+line_size))
|
|
|
165 :"m"(*p)
|
|
|
166 :"memory");
|
|
|
167 pix += line_size*2;
|
|
|
168 p += 16;
|
|
|
169 }
|
|
|
170 }
|
|
|
171
|
|
|
172 static void put_pixels_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)
|
|
|
173 {
|
|
|
174 int dh, hh;
|
|
|
175 UINT8 *p;
|
|
|
176 const UINT8 *pix;
|
|
|
177 p = block;
|
|
|
178 pix = pixels;
|
|
|
179 hh=h>>2;
|
|
|
180 dh=h&3;
|
|
|
181 while(hh--) {
|
|
|
182 __asm __volatile(
|
|
|
183 "movq %4, %%mm0\n\t"
|
|
|
184 "movq %5, %%mm1\n\t"
|
|
|
185 "movq %6, %%mm2\n\t"
|
|
|
186 "movq %7, %%mm3\n\t"
|
|
|
187 "movq %%mm0, %0\n\t"
|
|
|
188 "movq %%mm1, %1\n\t"
|
|
|
189 "movq %%mm2, %2\n\t"
|
|
|
190 "movq %%mm3, %3\n\t"
|
|
|
191 :"=m"(*p), "=m"(*(p+line_size)), "=m"(*(p+line_size*2)), "=m"(*(p+line_size*3))
|
|
|
192 :"m"(*pix), "m"(*(pix+line_size)), "m"(*(pix+line_size*2)), "m"(*(pix+line_size*3))
|
|
|
193 :"memory");
|
|
|
194 pix = pix + line_size*4;
|
|
|
195 p = p + line_size*4;
|
|
|
196 }
|
|
|
197 while(dh--) {
|
|
|
198 __asm __volatile(
|
|
|
199 "movq %1, %%mm0\n\t"
|
|
|
200 "movq %%mm0, %0\n\t"
|
|
|
201 :"=m"(*p)
|
|
|
202 :"m"(*pix)
|
|
|
203 :"memory");
|
|
|
204 pix = pix + line_size;
|
|
|
205 p = p + line_size;
|
|
|
206 }
|
|
|
207 }
|
|
|
208
|
|
|
209 static void put_pixels_x2_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)
|
|
|
210 {
|
|
|
211 UINT8 *p;
|
|
|
212 const UINT8 *pix;
|
|
|
213 p = block;
|
|
|
214 pix = pixels;
|
|
|
215 __asm __volatile(
|
|
|
216 "pxor %%mm7, %%mm7\n\t"
|
|
|
217 "movq %0, %%mm4\n\t"
|
|
|
218 ::"m"(mm_wone[0]):"memory");
|
|
|
219 do {
|
|
|
220 __asm __volatile(
|
|
|
221 "movq %1, %%mm0\n\t"
|
|
|
222 "movq 1%1, %%mm1\n\t"
|
|
|
223 "movq %%mm0, %%mm2\n\t"
|
|
|
224 "movq %%mm1, %%mm3\n\t"
|
|
|
225 "punpcklbw %%mm7, %%mm0\n\t"
|
|
|
226 "punpcklbw %%mm7, %%mm1\n\t"
|
|
|
227 "punpckhbw %%mm7, %%mm2\n\t"
|
|
|
228 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
229 "paddusw %%mm1, %%mm0\n\t"
|
|
|
230 "paddusw %%mm3, %%mm2\n\t"
|
|
|
231 "paddusw %%mm4, %%mm0\n\t"
|
|
|
232 "paddusw %%mm4, %%mm2\n\t"
|
|
|
233 "psrlw $1, %%mm0\n\t"
|
|
|
234 "psrlw $1, %%mm2\n\t"
|
|
|
235 "packuswb %%mm2, %%mm0\n\t"
|
|
|
236 "movq %%mm0, %0\n\t"
|
|
|
237 :"=m"(*p)
|
|
|
238 :"m"(*pix)
|
|
|
239 :"memory");
|
|
|
240 pix += line_size; p += line_size;
|
|
|
241 } while (--h);
|
|
|
242 }
|
|
|
243
|
|
|
244 static void put_pixels_y2_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)
|
|
|
245 {
|
|
|
246 UINT8 *p;
|
|
|
247 const UINT8 *pix;
|
|
|
248 p = block;
|
|
|
249 pix = pixels;
|
|
|
250 __asm __volatile(
|
|
|
251 "pxor %%mm7, %%mm7\n\t"
|
|
|
252 "movq %0, %%mm4\n\t"
|
|
|
253 ::"m"(mm_wone[0]):"memory");
|
|
|
254 do {
|
|
|
255 __asm __volatile(
|
|
|
256 "movq %1, %%mm0\n\t"
|
|
|
257 "movq %2, %%mm1\n\t"
|
|
|
258 "movq %%mm0, %%mm2\n\t"
|
|
|
259 "movq %%mm1, %%mm3\n\t"
|
|
|
260 "punpcklbw %%mm7, %%mm0\n\t"
|
|
|
261 "punpcklbw %%mm7, %%mm1\n\t"
|
|
|
262 "punpckhbw %%mm7, %%mm2\n\t"
|
|
|
263 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
264 "paddusw %%mm1, %%mm0\n\t"
|
|
|
265 "paddusw %%mm3, %%mm2\n\t"
|
|
|
266 "paddusw %%mm4, %%mm0\n\t"
|
|
|
267 "paddusw %%mm4, %%mm2\n\t"
|
|
|
268 "psrlw $1, %%mm0\n\t"
|
|
|
269 "psrlw $1, %%mm2\n\t"
|
|
|
270 "packuswb %%mm2, %%mm0\n\t"
|
|
|
271 "movq %%mm0, %0\n\t"
|
|
|
272 :"=m"(*p)
|
|
|
273 :"m"(*pix),
|
|
|
274 "m"(*(pix+line_size))
|
|
|
275 :"memory");
|
|
|
276 pix += line_size;
|
|
|
277 p += line_size;
|
|
|
278 } while (--h);
|
|
|
279 }
|
|
|
280
|
|
|
281 static void put_pixels_xy2_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)
|
|
|
282 {
|
|
|
283 UINT8 *p;
|
|
|
284 const UINT8 *pix;
|
|
|
285 p = block;
|
|
|
286 pix = pixels;
|
|
|
287 __asm __volatile(
|
|
|
288 "pxor %%mm7, %%mm7\n\t"
|
|
|
289 "movq %0, %%mm6\n\t"
|
|
|
290 ::"m"(mm_wtwo[0]):"memory");
|
|
|
291 do {
|
|
|
292 __asm __volatile(
|
|
|
293 "movq %1, %%mm0\n\t"
|
|
|
294 "movq %2, %%mm1\n\t"
|
|
|
295 "movq 1%1, %%mm4\n\t"
|
|
|
296 "movq 1%2, %%mm5\n\t"
|
|
|
297 "movq %%mm0, %%mm2\n\t"
|
|
|
298 "movq %%mm1, %%mm3\n\t"
|
|
|
299 "punpcklbw %%mm7, %%mm0\n\t"
|
|
|
300 "punpcklbw %%mm7, %%mm1\n\t"
|
|
|
301 "punpckhbw %%mm7, %%mm2\n\t"
|
|
|
302 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
303 "paddusw %%mm1, %%mm0\n\t"
|
|
|
304 "paddusw %%mm3, %%mm2\n\t"
|
|
|
305 "movq %%mm4, %%mm1\n\t"
|
|
|
306 "movq %%mm5, %%mm3\n\t"
|
|
|
307 "punpcklbw %%mm7, %%mm4\n\t"
|
|
|
308 "punpcklbw %%mm7, %%mm5\n\t"
|
|
|
309 "punpckhbw %%mm7, %%mm1\n\t"
|
|
|
310 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
311 "paddusw %%mm5, %%mm4\n\t"
|
|
|
312 "paddusw %%mm3, %%mm1\n\t"
|
|
|
313 "paddusw %%mm6, %%mm4\n\t"
|
|
|
314 "paddusw %%mm6, %%mm1\n\t"
|
|
|
315 "paddusw %%mm4, %%mm0\n\t"
|
|
|
316 "paddusw %%mm1, %%mm2\n\t"
|
|
|
317 "psrlw $2, %%mm0\n\t"
|
|
|
318 "psrlw $2, %%mm2\n\t"
|
|
|
319 "packuswb %%mm2, %%mm0\n\t"
|
|
|
320 "movq %%mm0, %0\n\t"
|
|
|
321 :"=m"(*p)
|
|
|
322 :"m"(*pix),
|
|
|
323 "m"(*(pix+line_size))
|
|
|
324 :"memory");
|
|
|
325 pix += line_size;
|
|
|
326 p += line_size;
|
|
|
327 } while(--h);
|
|
|
328 }
|
|
|
329
|
|
|
330 static void put_no_rnd_pixels_x2_mmx( UINT8 *block, const UINT8 *pixels, int line_size, int h)
|
|
|
331 {
|
|
|
332 UINT8 *p;
|
|
|
333 const UINT8 *pix;
|
|
|
334 p = block;
|
|
|
335 pix = pixels;
|
|
|
336 __asm __volatile("pxor %%mm7, %%mm7\n\t":::"memory");
|
|
|
337 do {
|
|
|
338 __asm __volatile(
|
|
|
339 "movq %1, %%mm0\n\t"
|
|
|
340 "movq 1%1, %%mm1\n\t"
|
|
|
341 "movq %%mm0, %%mm2\n\t"
|
|
|
342 "movq %%mm1, %%mm3\n\t"
|
|
|
343 "punpcklbw %%mm7, %%mm0\n\t"
|
|
|
344 "punpcklbw %%mm7, %%mm1\n\t"
|
|
|
345 "punpckhbw %%mm7, %%mm2\n\t"
|
|
|
346 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
347 "paddusw %%mm1, %%mm0\n\t"
|
|
|
348 "paddusw %%mm3, %%mm2\n\t"
|
|
|
349 "psrlw $1, %%mm0\n\t"
|
|
|
350 "psrlw $1, %%mm2\n\t"
|
|
|
351 "packuswb %%mm2, %%mm0\n\t"
|
|
|
352 "movq %%mm0, %0\n\t"
|
|
|
353 :"=m"(*p)
|
|
|
354 :"m"(*pix)
|
|
|
355 :"memory");
|
|
|
356 pix += line_size;
|
|
|
357 p += line_size;
|
|
|
358 } while (--h);
|
|
|
359 }
|
|
|
360
|
|
|
361 static void put_no_rnd_pixels_y2_mmx( UINT8 *block, const UINT8 *pixels, int line_size, int h)
|
|
|
362 {
|
|
|
363 UINT8 *p;
|
|
|
364 const UINT8 *pix;
|
|
|
365 p = block;
|
|
|
366 pix = pixels;
|
|
|
367 __asm __volatile("pxor %%mm7, %%mm7\n\t":::"memory");
|
|
|
368 do {
|
|
|
369 __asm __volatile(
|
|
|
370 "movq %1, %%mm0\n\t"
|
|
|
371 "movq %2, %%mm1\n\t"
|
|
|
372 "movq %%mm0, %%mm2\n\t"
|
|
|
373 "movq %%mm1, %%mm3\n\t"
|
|
|
374 "punpcklbw %%mm7, %%mm0\n\t"
|
|
|
375 "punpcklbw %%mm7, %%mm1\n\t"
|
|
|
376 "punpckhbw %%mm7, %%mm2\n\t"
|
|
|
377 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
378 "paddusw %%mm1, %%mm0\n\t"
|
|
|
379 "paddusw %%mm3, %%mm2\n\t"
|
|
|
380 "psrlw $1, %%mm0\n\t"
|
|
|
381 "psrlw $1, %%mm2\n\t"
|
|
|
382 "packuswb %%mm2, %%mm0\n\t"
|
|
|
383 "movq %%mm0, %0\n\t"
|
|
|
384 :"=m"(*p)
|
|
|
385 :"m"(*pix),
|
|
|
386 "m"(*(pix+line_size))
|
|
|
387 :"memory");
|
|
|
388 pix += line_size;
|
|
|
389 p += line_size;
|
|
|
390 } while(--h);
|
|
|
391 }
|
|
|
392
|
|
|
393 static void put_no_rnd_pixels_xy2_mmx( UINT8 *block, const UINT8 *pixels, int line_size, int h)
|
|
|
394 {
|
|
|
395 UINT8 *p;
|
|
|
396 const UINT8 *pix;
|
|
|
397 p = block;
|
|
|
398 pix = pixels;
|
|
|
399 __asm __volatile(
|
|
|
400 "pxor %%mm7, %%mm7\n\t"
|
|
|
401 "movq %0, %%mm6\n\t"
|
|
|
402 ::"m"(mm_wone[0]):"memory");
|
|
|
403 do {
|
|
|
404 __asm __volatile(
|
|
|
405 "movq %1, %%mm0\n\t"
|
|
|
406 "movq %2, %%mm1\n\t"
|
|
|
407 "movq 1%1, %%mm4\n\t"
|
|
|
408 "movq 1%2, %%mm5\n\t"
|
|
|
409 "movq %%mm0, %%mm2\n\t"
|
|
|
410 "movq %%mm1, %%mm3\n\t"
|
|
|
411 "punpcklbw %%mm7, %%mm0\n\t"
|
|
|
412 "punpcklbw %%mm7, %%mm1\n\t"
|
|
|
413 "punpckhbw %%mm7, %%mm2\n\t"
|
|
|
414 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
415 "paddusw %%mm1, %%mm0\n\t"
|
|
|
416 "paddusw %%mm3, %%mm2\n\t"
|
|
|
417 "movq %%mm4, %%mm1\n\t"
|
|
|
418 "movq %%mm5, %%mm3\n\t"
|
|
|
419 "punpcklbw %%mm7, %%mm4\n\t"
|
|
|
420 "punpcklbw %%mm7, %%mm5\n\t"
|
|
|
421 "punpckhbw %%mm7, %%mm1\n\t"
|
|
|
422 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
423 "paddusw %%mm5, %%mm4\n\t"
|
|
|
424 "paddusw %%mm3, %%mm1\n\t"
|
|
|
425 "paddusw %%mm6, %%mm4\n\t"
|
|
|
426 "paddusw %%mm6, %%mm1\n\t"
|
|
|
427 "paddusw %%mm4, %%mm0\n\t"
|
|
|
428 "paddusw %%mm1, %%mm2\n\t"
|
|
|
429 "psrlw $2, %%mm0\n\t"
|
|
|
430 "psrlw $2, %%mm2\n\t"
|
|
|
431 "packuswb %%mm2, %%mm0\n\t"
|
|
|
432 "movq %%mm0, %0\n\t"
|
|
|
433 :"=m"(*p)
|
|
|
434 :"m"(*pix),
|
|
|
435 "m"(*(pix+line_size))
|
|
|
436 :"memory");
|
|
|
437 pix += line_size;
|
|
|
438 p += line_size;
|
|
|
439 } while(--h);
|
|
|
440 }
|
|
|
441
|
|
|
442 static void avg_pixels_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)
|
|
|
443 {
|
|
|
444 UINT8 *p;
|
|
|
445 const UINT8 *pix;
|
|
|
446 p = block;
|
|
|
447 pix = pixels;
|
|
|
448 __asm __volatile(
|
|
|
449 "pxor %%mm7, %%mm7\n\t"
|
|
|
450 "movq %0, %%mm6\n\t"
|
|
|
451 ::"m"(mm_wone[0]):"memory");
|
|
|
452 do {
|
|
|
453 __asm __volatile(
|
|
|
454 "movq %0, %%mm0\n\t"
|
|
|
455 "movq %1, %%mm1\n\t"
|
|
|
456 "movq %%mm0, %%mm2\n\t"
|
|
|
457 "movq %%mm1, %%mm3\n\t"
|
|
|
458 "punpcklbw %%mm7, %%mm0\n\t"
|
|
|
459 "punpcklbw %%mm7, %%mm1\n\t"
|
|
|
460 "punpckhbw %%mm7, %%mm2\n\t"
|
|
|
461 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
462 "paddusw %%mm1, %%mm0\n\t"
|
|
|
463 "paddusw %%mm3, %%mm2\n\t"
|
|
|
464 "paddusw %%mm6, %%mm0\n\t"
|
|
|
465 "paddusw %%mm6, %%mm2\n\t"
|
|
|
466 "psrlw $1, %%mm0\n\t"
|
|
|
467 "psrlw $1, %%mm2\n\t"
|
|
|
468 "packuswb %%mm2, %%mm0\n\t"
|
|
|
469 "movq %%mm0, %0\n\t"
|
|
|
470 :"=m"(*p)
|
|
|
471 :"m"(*pix)
|
|
|
472 :"memory");
|
|
|
473 pix += line_size;
|
|
|
474 p += line_size;
|
|
|
475 }
|
|
|
476 while (--h);
|
|
|
477 }
|
|
|
478
|
|
|
479 static void avg_pixels_x2_mmx( UINT8 *block, const UINT8 *pixels, int line_size, int h)
|
|
|
480 {
|
|
|
481 UINT8 *p;
|
|
|
482 const UINT8 *pix;
|
|
|
483 p = block;
|
|
|
484 pix = pixels;
|
|
|
485 __asm __volatile(
|
|
|
486 "pxor %%mm7, %%mm7\n\t"
|
|
|
487 "movq %0, %%mm6\n\t"
|
|
|
488 ::"m"(mm_wone[0]):"memory");
|
|
|
489 do {
|
|
|
490 __asm __volatile(
|
|
|
491 "movq %1, %%mm1\n\t"
|
|
|
492 "movq %0, %%mm0\n\t"
|
|
|
493 "movq 1%1, %%mm4\n\t"
|
|
|
494 "movq %%mm0, %%mm2\n\t"
|
|
|
495 "movq %%mm1, %%mm3\n\t"
|
|
|
496 "movq %%mm4, %%mm5\n\t"
|
|
|
497 "punpcklbw %%mm7, %%mm1\n\t"
|
|
|
498 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
499 "punpcklbw %%mm7, %%mm4\n\t"
|
|
|
500 "punpckhbw %%mm7, %%mm5\n\t"
|
|
|
501 "punpcklbw %%mm7, %%mm0\n\t"
|
|
|
502 "punpckhbw %%mm7, %%mm2\n\t"
|
|
|
503 "paddusw %%mm4, %%mm1\n\t"
|
|
|
504 "paddusw %%mm5, %%mm3\n\t"
|
|
|
505 "paddusw %%mm6, %%mm1\n\t"
|
|
|
506 "paddusw %%mm6, %%mm3\n\t"
|
|
|
507 "psrlw $1, %%mm1\n\t"
|
|
|
508 "psrlw $1, %%mm3\n\t"
|
|
|
509 "paddusw %%mm6, %%mm0\n\t"
|
|
|
510 "paddusw %%mm6, %%mm2\n\t"
|
|
|
511 "paddusw %%mm1, %%mm0\n\t"
|
|
|
512 "paddusw %%mm3, %%mm2\n\t"
|
|
|
513 "psrlw $1, %%mm0\n\t"
|
|
|
514 "psrlw $1, %%mm2\n\t"
|
|
|
515 "packuswb %%mm2, %%mm0\n\t"
|
|
|
516 "movq %%mm0, %0\n\t"
|
|
|
517 :"=m"(*p)
|
|
|
518 :"m"(*pix)
|
|
|
519 :"memory");
|
|
|
520 pix += line_size;
|
|
|
521 p += line_size;
|
|
|
522 } while (--h);
|
|
|
523 }
|
|
|
524
|
|
|
525 static void avg_pixels_y2_mmx( UINT8 *block, const UINT8 *pixels, int line_size, int h)
|
|
|
526 {
|
|
|
527 UINT8 *p;
|
|
|
528 const UINT8 *pix;
|
|
|
529 p = block;
|
|
|
530 pix = pixels;
|
|
|
531 __asm __volatile(
|
|
|
532 "pxor %%mm7, %%mm7\n\t"
|
|
|
533 "movq %0, %%mm6\n\t"
|
|
|
534 ::"m"(mm_wone[0]):"memory");
|
|
|
535 do {
|
|
|
536 __asm __volatile(
|
|
|
537 "movq %1, %%mm1\n\t"
|
|
|
538 "movq %0, %%mm0\n\t"
|
|
|
539 "movq %2, %%mm4\n\t"
|
|
|
540 "movq %%mm0, %%mm2\n\t"
|
|
|
541 "movq %%mm1, %%mm3\n\t"
|
|
|
542 "movq %%mm4, %%mm5\n\t"
|
|
|
543 "punpcklbw %%mm7, %%mm1\n\t"
|
|
|
544 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
545 "punpcklbw %%mm7, %%mm4\n\t"
|
|
|
546 "punpckhbw %%mm7, %%mm5\n\t"
|
|
|
547 "punpcklbw %%mm7, %%mm0\n\t"
|
|
|
548 "punpckhbw %%mm7, %%mm2\n\t"
|
|
|
549 "paddusw %%mm4, %%mm1\n\t"
|
|
|
550 "paddusw %%mm5, %%mm3\n\t"
|
|
|
551 "paddusw %%mm6, %%mm1\n\t"
|
|
|
552 "paddusw %%mm6, %%mm3\n\t"
|
|
|
553 "psrlw $1, %%mm1\n\t"
|
|
|
554 "psrlw $1, %%mm3\n\t"
|
|
|
555 "paddusw %%mm6, %%mm0\n\t"
|
|
|
556 "paddusw %%mm6, %%mm2\n\t"
|
|
|
557 "paddusw %%mm1, %%mm0\n\t"
|
|
|
558 "paddusw %%mm3, %%mm2\n\t"
|
|
|
559 "psrlw $1, %%mm0\n\t"
|
|
|
560 "psrlw $1, %%mm2\n\t"
|
|
|
561 "packuswb %%mm2, %%mm0\n\t"
|
|
|
562 "movq %%mm0, %0\n\t"
|
|
|
563 :"=m"(*p)
|
|
|
564 :"m"(*pix), "m"(*(pix+line_size))
|
|
|
565 :"memory");
|
|
|
566 pix += line_size;
|
|
|
567 p += line_size ;
|
|
|
568 } while(--h);
|
|
|
569 }
|
|
|
570
|
|
|
571 static void avg_pixels_xy2_mmx( UINT8 *block, const UINT8 *pixels, int line_size, int h)
|
|
|
572 {
|
|
|
573 UINT8 *p;
|
|
|
574 const UINT8 *pix;
|
|
|
575 p = block;
|
|
|
576 pix = pixels;
|
|
|
577 __asm __volatile(
|
|
|
578 "pxor %%mm7, %%mm7\n\t"
|
|
|
579 "movq %0, %%mm6\n\t"
|
|
|
580 ::"m"(mm_wtwo[0]):"memory");
|
|
|
581 do {
|
|
|
582 __asm __volatile(
|
|
|
583 "movq %1, %%mm0\n\t"
|
|
|
584 "movq %2, %%mm1\n\t"
|
|
|
585 "movq 1%1, %%mm4\n\t"
|
|
|
586 "movq 1%2, %%mm5\n\t"
|
|
|
587 "movq %%mm0, %%mm2\n\t"
|
|
|
588 "movq %%mm1, %%mm3\n\t"
|
|
|
589 "punpcklbw %%mm7, %%mm0\n\t"
|
|
|
590 "punpcklbw %%mm7, %%mm1\n\t"
|
|
|
591 "punpckhbw %%mm7, %%mm2\n\t"
|
|
|
592 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
593 "paddusw %%mm1, %%mm0\n\t"
|
|
|
594 "paddusw %%mm3, %%mm2\n\t"
|
|
|
595 "movq %%mm4, %%mm1\n\t"
|
|
|
596 "movq %%mm5, %%mm3\n\t"
|
|
|
597 "punpcklbw %%mm7, %%mm4\n\t"
|
|
|
598 "punpcklbw %%mm7, %%mm5\n\t"
|
|
|
599 "punpckhbw %%mm7, %%mm1\n\t"
|
|
|
600 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
601 "paddusw %%mm5, %%mm4\n\t"
|
|
|
602 "paddusw %%mm3, %%mm1\n\t"
|
|
|
603 "paddusw %%mm6, %%mm4\n\t"
|
|
|
604 "paddusw %%mm6, %%mm1\n\t"
|
|
|
605 "paddusw %%mm4, %%mm0\n\t"
|
|
|
606 "paddusw %%mm1, %%mm2\n\t"
|
|
|
607 "movq %3, %%mm5\n\t"
|
|
|
608 "psrlw $2, %%mm0\n\t"
|
|
|
609 "movq %0, %%mm1\n\t"
|
|
|
610 "psrlw $2, %%mm2\n\t"
|
|
|
611 "movq %%mm1, %%mm3\n\t"
|
|
|
612 "punpcklbw %%mm7, %%mm1\n\t"
|
|
|
613 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
614 "paddusw %%mm1, %%mm0\n\t"
|
|
|
615 "paddusw %%mm3, %%mm2\n\t"
|
|
|
616 "paddusw %%mm5, %%mm0\n\t"
|
|
|
617 "paddusw %%mm5, %%mm2\n\t"
|
|
|
618 "psrlw $1, %%mm0\n\t"
|
|
|
619 "psrlw $1, %%mm2\n\t"
|
|
|
620 "packuswb %%mm2, %%mm0\n\t"
|
|
|
621 "movq %%mm0, %0\n\t"
|
|
|
622 :"=m"(*p)
|
|
|
623 :"m"(*pix),
|
|
|
624 "m"(*(pix+line_size)), "m"(mm_wone[0])
|
|
|
625 :"memory");
|
|
|
626 pix += line_size;
|
|
|
627 p += line_size ;
|
|
|
628 } while(--h);
|
|
|
629 }
|
|
|
630
|
|
|
631 static void avg_no_rnd_pixels_mmx( UINT8 *block, const UINT8 *pixels, int line_size, int h)
|
|
|
632 {
|
|
|
633 UINT8 *p;
|
|
|
634 const UINT8 *pix;
|
|
|
635 p = block;
|
|
|
636 pix = pixels;
|
|
|
637 __asm __volatile("pxor %%mm7, %%mm7\n\t":::"memory");
|
|
|
638 do {
|
|
|
639 __asm __volatile(
|
|
|
640 "movq %1, %%mm0\n\t"
|
|
|
641 "movq %0, %%mm1\n\t"
|
|
|
642 "movq %%mm0, %%mm2\n\t"
|
|
|
643 "movq %%mm1, %%mm3\n\t"
|
|
|
644 "punpcklbw %%mm7, %%mm0\n\t"
|
|
|
645 "punpcklbw %%mm7, %%mm1\n\t"
|
|
|
646 "punpckhbw %%mm7, %%mm2\n\t"
|
|
|
647 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
648 "paddusw %%mm1, %%mm0\n\t"
|
|
|
649 "paddusw %%mm3, %%mm2\n\t"
|
|
|
650 "psrlw $1, %%mm0\n\t"
|
|
|
651 "psrlw $1, %%mm2\n\t"
|
|
|
652 "packuswb %%mm2, %%mm0\n\t"
|
|
|
653 "movq %%mm0, %0\n\t"
|
|
|
654 :"=m"(*p)
|
|
|
655 :"m"(*pix)
|
|
|
656 :"memory");
|
|
|
657 pix += line_size;
|
|
|
658 p += line_size ;
|
|
|
659 } while (--h);
|
|
|
660 }
|
|
|
661
|
|
|
662 static void avg_no_rnd_pixels_x2_mmx( UINT8 *block, const UINT8 *pixels, int line_size, int h)
|
|
|
663 {
|
|
|
664 UINT8 *p;
|
|
|
665 const UINT8 *pix;
|
|
|
666 p = block;
|
|
|
667 pix = pixels;
|
|
|
668 __asm __volatile(
|
|
|
669 "pxor %%mm7, %%mm7\n\t":::"memory");
|
|
|
670 do {
|
|
|
671 __asm __volatile(
|
|
|
672 "movq %1, %%mm0\n\t"
|
|
|
673 "movq 1%1, %%mm1\n\t"
|
|
|
674 "movq %0, %%mm4\n\t"
|
|
|
675 "movq %%mm0, %%mm2\n\t"
|
|
|
676 "movq %%mm1, %%mm3\n\t"
|
|
|
677 "movq %%mm4, %%mm5\n\t"
|
|
|
678 "punpcklbw %%mm7, %%mm0\n\t"
|
|
|
679 "punpcklbw %%mm7, %%mm1\n\t"
|
|
|
680 "punpckhbw %%mm7, %%mm2\n\t"
|
|
|
681 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
682 "punpcklbw %%mm7, %%mm4\n\t"
|
|
|
683 "punpckhbw %%mm7, %%mm5\n\t"
|
|
|
684 "paddusw %%mm1, %%mm0\n\t"
|
|
|
685 "paddusw %%mm3, %%mm2\n\t"
|
|
|
686 "psrlw $1, %%mm0\n\t"
|
|
|
687 "psrlw $1, %%mm2\n\t"
|
|
|
688 "paddusw %%mm4, %%mm0\n\t"
|
|
|
689 "paddusw %%mm5, %%mm2\n\t"
|
|
|
690 "psrlw $1, %%mm0\n\t"
|
|
|
691 "psrlw $1, %%mm2\n\t"
|
|
|
692 "packuswb %%mm2, %%mm0\n\t"
|
|
|
693 "movq %%mm0, %0\n\t"
|
|
|
694 :"=m"(*p)
|
|
|
695 :"m"(*pix)
|
|
|
696 :"memory");
|
|
|
697 pix += line_size;
|
|
|
698 p += line_size;
|
|
|
699 } while (--h);
|
|
|
700 }
|
|
|
701
|
|
|
702 static void avg_no_rnd_pixels_y2_mmx( UINT8 *block, const UINT8 *pixels, int line_size, int h)
|
|
|
703 {
|
|
|
704 UINT8 *p;
|
|
|
705 const UINT8 *pix;
|
|
|
706 p = block;
|
|
|
707 pix = pixels;
|
|
|
708 __asm __volatile(
|
|
|
709 "pxor %%mm7, %%mm7\n\t":::"memory");
|
|
|
710 do {
|
|
|
711 __asm __volatile(
|
|
|
712 "movq %1, %%mm0\n\t"
|
|
|
713 "movq %2, %%mm1\n\t"
|
|
|
714 "movq %0, %%mm4\n\t"
|
|
|
715 "movq %%mm0, %%mm2\n\t"
|
|
|
716 "movq %%mm1, %%mm3\n\t"
|
|
|
717 "movq %%mm4, %%mm5\n\t"
|
|
|
718 "punpcklbw %%mm7, %%mm0\n\t"
|
|
|
719 "punpcklbw %%mm7, %%mm1\n\t"
|
|
|
720 "punpckhbw %%mm7, %%mm2\n\t"
|
|
|
721 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
722 "punpcklbw %%mm7, %%mm4\n\t"
|
|
|
723 "punpckhbw %%mm7, %%mm5\n\t"
|
|
|
724 "paddusw %%mm1, %%mm0\n\t"
|
|
|
725 "paddusw %%mm3, %%mm2\n\t"
|
|
|
726 "psrlw $1, %%mm0\n\t"
|
|
|
727 "psrlw $1, %%mm2\n\t"
|
|
|
728 "paddusw %%mm4, %%mm0\n\t"
|
|
|
729 "paddusw %%mm5, %%mm2\n\t"
|
|
|
730 "psrlw $1, %%mm0\n\t"
|
|
|
731 "psrlw $1, %%mm2\n\t"
|
|
|
732 "packuswb %%mm2, %%mm0\n\t"
|
|
|
733 "movq %%mm0, %0\n\t"
|
|
|
734 :"=m"(*p)
|
|
|
735 :"m"(*pix), "m"(*(pix+line_size))
|
|
|
736 :"memory");
|
|
|
737 pix += line_size;
|
|
|
738 p += line_size ;
|
|
|
739 } while(--h);
|
|
|
740 }
|
|
|
741
|
|
|
742 static void avg_no_rnd_pixels_xy2_mmx( UINT8 *block, const UINT8 *pixels, int line_size, int h)
|
|
|
743 {
|
|
|
744 UINT8 *p;
|
|
|
745 const UINT8 *pix;
|
|
|
746 p = block;
|
|
|
747 pix = pixels;
|
|
|
748 __asm __volatile(
|
|
|
749 "pxor %%mm7, %%mm7\n\t"
|
|
|
750 "movq %0, %%mm6\n\t"
|
|
|
751 ::"m"(mm_wone[0]):"memory");
|
|
|
752 do {
|
|
|
753 __asm __volatile(
|
|
|
754 "movq %1, %%mm0\n\t"
|
|
|
755 "movq %2, %%mm1\n\t"
|
|
|
756 "movq 1%1, %%mm4\n\t"
|
|
|
757 "movq 1%2, %%mm5\n\t"
|
|
|
758 "movq %%mm0, %%mm2\n\t"
|
|
|
759 "movq %%mm1, %%mm3\n\t"
|
|
|
760 "punpcklbw %%mm7, %%mm0\n\t"
|
|
|
761 "punpcklbw %%mm7, %%mm1\n\t"
|
|
|
762 "punpckhbw %%mm7, %%mm2\n\t"
|
|
|
763 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
764 "paddusw %%mm1, %%mm0\n\t"
|
|
|
765 "paddusw %%mm3, %%mm2\n\t"
|
|
|
766 "movq %%mm4, %%mm1\n\t"
|
|
|
767 "movq %%mm5, %%mm3\n\t"
|
|
|
768 "punpcklbw %%mm7, %%mm4\n\t"
|
|
|
769 "punpcklbw %%mm7, %%mm5\n\t"
|
|
|
770 "punpckhbw %%mm7, %%mm1\n\t"
|
|
|
771 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
772 "paddusw %%mm5, %%mm4\n\t"
|
|
|
773 "paddusw %%mm3, %%mm1\n\t"
|
|
|
774 "paddusw %%mm6, %%mm4\n\t"
|
|
|
775 "paddusw %%mm6, %%mm1\n\t"
|
|
|
776 "paddusw %%mm4, %%mm0\n\t"
|
|
|
777 "paddusw %%mm1, %%mm2\n\t"
|
|
|
778 "movq %0, %%mm1\n\t"
|
|
|
779 "psrlw $2, %%mm0\n\t"
|
|
|
780 "movq %%mm1, %%mm3\n\t"
|
|
|
781 "psrlw $2, %%mm2\n\t"
|
|
|
782 "punpcklbw %%mm7, %%mm1\n\t"
|
|
|
783 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
784 "paddusw %%mm1, %%mm0\n\t"
|
|
|
785 "paddusw %%mm3, %%mm2\n\t"
|
|
|
786 "psrlw $1, %%mm0\n\t"
|
|
|
787 "psrlw $1, %%mm2\n\t"
|
|
|
788 "packuswb %%mm2, %%mm0\n\t"
|
|
|
789 "movq %%mm0, %0\n\t"
|
|
|
790 :"=m"(*p)
|
|
|
791 :"m"(*pix),
|
|
|
792 "m"(*(pix+line_size))
|
|
|
793 :"memory");
|
|
|
794 pix += line_size;
|
|
|
795 p += line_size;
|
|
|
796 } while(--h);
|
|
|
797 }
|
|
|
798
|
|
|
799 static void sub_pixels_mmx( DCTELEM *block, const UINT8 *pixels, int line_size, int h)
|
|
|
800 {
|
|
|
801 DCTELEM *p;
|
|
|
802 const UINT8 *pix;
|
|
|
803 p = block;
|
|
|
804 pix = pixels;
|
|
|
805 __asm __volatile("pxor %%mm7, %%mm7":::"memory");
|
|
|
806 do {
|
|
|
807 __asm __volatile(
|
|
|
808 "movq %0, %%mm0\n\t"
|
|
|
809 "movq %1, %%mm2\n\t"
|
|
|
810 "movq 8%0, %%mm1\n\t"
|
|
|
811 "movq %%mm2, %%mm3\n\t"
|
|
|
812 "punpcklbw %%mm7, %%mm2\n\t"
|
|
|
813 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
814 "psubsw %%mm2, %%mm0\n\t"
|
|
|
815 "psubsw %%mm3, %%mm1\n\t"
|
|
|
816 "movq %%mm0, %0\n\t"
|
|
|
817 "movq %%mm1, 8%0\n\t"
|
|
|
818 :"=m"(*p)
|
|
|
819 :"m"(*pix)
|
|
|
820 :"memory");
|
|
|
821 pix += line_size;
|
|
|
822 p += 8;
|
|
|
823 } while (--h);
|
|
|
824 }
|
|
|
825
|
|
|
826 static void sub_pixels_x2_mmx( DCTELEM *block, const UINT8 *pixels, int line_size, int h)
|
|
|
827 {
|
|
|
828 DCTELEM *p;
|
|
|
829 const UINT8 *pix;
|
|
|
830 p = block;
|
|
|
831 pix = pixels;
|
|
|
832 __asm __volatile(
|
|
|
833 "pxor %%mm7, %%mm7\n\t"
|
|
|
834 "movq %0, %%mm6"
|
|
|
835 ::"m"(mm_wone[0]):"memory");
|
|
|
836 do {
|
|
|
837 __asm __volatile(
|
|
|
838 "movq %0, %%mm0\n\t"
|
|
|
839 "movq %1, %%mm2\n\t"
|
|
|
840 "movq 8%0, %%mm1\n\t"
|
|
|
841 "movq 1%1, %%mm4\n\t"
|
|
|
842 "movq %%mm2, %%mm3\n\t"
|
|
|
843 "movq %%mm4, %%mm5\n\t"
|
|
|
844 "punpcklbw %%mm7, %%mm2\n\t"
|
|
|
845 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
846 "punpcklbw %%mm7, %%mm4\n\t"
|
|
|
847 "punpckhbw %%mm7, %%mm5\n\t"
|
|
|
848 "paddusw %%mm4, %%mm2\n\t"
|
|
|
849 "paddusw %%mm5, %%mm3\n\t"
|
|
|
850 "paddusw %%mm6, %%mm2\n\t"
|
|
|
851 "paddusw %%mm6, %%mm3\n\t"
|
|
|
852 "psrlw $1, %%mm2\n\t"
|
|
|
853 "psrlw $1, %%mm3\n\t"
|
|
|
854 "psubsw %%mm2, %%mm0\n\t"
|
|
|
855 "psubsw %%mm3, %%mm1\n\t"
|
|
|
856 "movq %%mm0, %0\n\t"
|
|
|
857 "movq %%mm1, 8%0\n\t"
|
|
|
858 :"=m"(*p)
|
|
|
859 :"m"(*pix)
|
|
|
860 :"memory");
|
|
|
861 pix += line_size;
|
|
|
862 p += 8;
|
|
|
863 } while (--h);
|
|
|
864 }
|
|
|
865
|
|
|
866 static void sub_pixels_y2_mmx( DCTELEM *block, const UINT8 *pixels, int line_size, int h)
|
|
|
867 {
|
|
|
868 DCTELEM *p;
|
|
|
869 const UINT8 *pix;
|
|
|
870 p = block;
|
|
|
871 pix = pixels;
|
|
|
872 __asm __volatile(
|
|
|
873 "pxor %%mm7, %%mm7\n\t"
|
|
|
874 "movq %0, %%mm6"
|
|
|
875 ::"m"(mm_wone[0]):"memory");
|
|
|
876 do {
|
|
|
877 __asm __volatile(
|
|
|
878 "movq %0, %%mm0\n\t"
|
|
|
879 "movq %1, %%mm2\n\t"
|
|
|
880 "movq 8%0, %%mm1\n\t"
|
|
|
881 "movq %2, %%mm4\n\t"
|
|
|
882 "movq %%mm2, %%mm3\n\t"
|
|
|
883 "movq %%mm4, %%mm5\n\t"
|
|
|
884 "punpcklbw %%mm7, %%mm2\n\t"
|
|
|
885 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
886 "punpcklbw %%mm7, %%mm4\n\t"
|
|
|
887 "punpckhbw %%mm7, %%mm5\n\t"
|
|
|
888 "paddusw %%mm4, %%mm2\n\t"
|
|
|
889 "paddusw %%mm5, %%mm3\n\t"
|
|
|
890 "paddusw %%mm6, %%mm2\n\t"
|
|
|
891 "paddusw %%mm6, %%mm3\n\t"
|
|
|
892 "psrlw $1, %%mm2\n\t"
|
|
|
893 "psrlw $1, %%mm3\n\t"
|
|
|
894 "psubsw %%mm2, %%mm0\n\t"
|
|
|
895 "psubsw %%mm3, %%mm1\n\t"
|
|
|
896 "movq %%mm0, %0\n\t"
|
|
|
897 "movq %%mm1, 8%0\n\t"
|
|
|
898 :"=m"(*p)
|
|
|
899 :"m"(*pix), "m"(*(pix+line_size))
|
|
|
900 :"memory");
|
|
|
901 pix += line_size;
|
|
|
902 p += 8;
|
|
|
903 } while (--h);
|
|
|
904 }
|
|
|
905
|
|
|
906 static void sub_pixels_xy2_mmx( DCTELEM *block, const UINT8 *pixels, int line_size, int h)
|
|
|
907 {
|
|
|
908 DCTELEM *p;
|
|
|
909 const UINT8 *pix;
|
|
|
910 p = block;
|
|
|
911 pix = pixels;
|
|
|
912 __asm __volatile(
|
|
|
913 "pxor %%mm7, %%mm7\n\t"
|
|
|
914 "movq %0, %%mm6\n\t"
|
|
|
915 ::"m"(mm_wtwo[0]):"memory");
|
|
|
916 do {
|
|
|
917 __asm __volatile(
|
|
|
918 "movq %1, %%mm0\n\t"
|
|
|
919 "movq %2, %%mm1\n\t"
|
|
|
920 "movq 1%1, %%mm4\n\t"
|
|
|
921 "movq 1%2, %%mm5\n\t"
|
|
|
922 "movq %%mm0, %%mm2\n\t"
|
|
|
923 "movq %%mm1, %%mm3\n\t"
|
|
|
924 "punpcklbw %%mm7, %%mm0\n\t"
|
|
|
925 "punpcklbw %%mm7, %%mm1\n\t"
|
|
|
926 "punpckhbw %%mm7, %%mm2\n\t"
|
|
|
927 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
928 "paddusw %%mm1, %%mm0\n\t"
|
|
|
929 "paddusw %%mm3, %%mm2\n\t"
|
|
|
930 "movq %%mm4, %%mm1\n\t"
|
|
|
931 "movq %%mm5, %%mm3\n\t"
|
|
|
932 "punpcklbw %%mm7, %%mm4\n\t"
|
|
|
933 "punpcklbw %%mm7, %%mm5\n\t"
|
|
|
934 "punpckhbw %%mm7, %%mm1\n\t"
|
|
|
935 "punpckhbw %%mm7, %%mm3\n\t"
|
|
|
936 "paddusw %%mm5, %%mm4\n\t"
|
|
|
937 "paddusw %%mm3, %%mm1\n\t"
|
|
|
938 "paddusw %%mm6, %%mm4\n\t"
|
|
|
939 "paddusw %%mm6, %%mm1\n\t"
|
|
|
940 "paddusw %%mm4, %%mm0\n\t"
|
|
|
941 "paddusw %%mm1, %%mm2\n\t"
|
|
|
942 "movq %0, %%mm1\n\t"
|
|
|
943 "movq 8%0, %%mm3\n\t"
|
|
|
944 "psrlw $2, %%mm0\n\t"
|
|
|
945 "psrlw $2, %%mm2\n\t"
|
|
|
946 "psubsw %%mm0, %%mm1\n\t"
|
|
|
947 "psubsw %%mm2, %%mm3\n\t"
|
|
|
948 "movq %%mm1, %0\n\t"
|
|
|
949 "movq %%mm3, 8%0\n\t"
|
|
|
950 :"=m"(*p)
|
|
|
951 :"m"(*pix),
|
|
|
952 "m"(*(pix+line_size))
|
|
|
953 :"memory");
|
|
|
954 pix += line_size;
|
|
|
955 p += 8 ;
|
|
|
956 } while(--h);
|
|
|
957 }
|
|
|
958
|
|
|
959 void dsputil_init_mmx(void)
|
|
|
960 {
|
|
|
961 mm_flags = mm_support();
|
|
|
962 #if 0
|
|
|
963 printf("CPU flags:");
|
|
|
964 if (mm_flags & MM_MMX)
|
|
|
965 printf(" mmx");
|
|
|
966 if (mm_flags & MM_MMXEXT)
|
|
|
967 printf(" mmxext");
|
|
|
968 if (mm_flags & MM_3DNOW)
|
|
|
969 printf(" 3dnow");
|
|
|
970 if (mm_flags & MM_SSE)
|
|
|
971 printf(" sse");
|
|
|
972 if (mm_flags & MM_SSE2)
|
|
|
973 printf(" sse2");
|
|
|
974 printf("\n");
|
|
|
975 #endif
|
|
|
976
|
|
|
977 if (mm_flags & MM_MMX) {
|
|
|
978 get_pixels = get_pixels_mmx;
|
|
|
979 put_pixels_clamped = put_pixels_clamped_mmx;
|
|
|
980 add_pixels_clamped = add_pixels_clamped_mmx;
|
|
|
981
|
|
|
982 pix_abs16x16 = pix_abs16x16_mmx;
|
|
|
983 pix_abs16x16_x2 = pix_abs16x16_x2_mmx;
|
|
|
984 pix_abs16x16_y2 = pix_abs16x16_y2_mmx;
|
|
|
985 pix_abs16x16_xy2 = pix_abs16x16_xy2_mmx;
|
|
|
986 av_fdct = fdct_mmx;
|
|
|
987
|
|
|
988 put_pixels_tab[0] = put_pixels_mmx;
|
|
|
989 put_pixels_tab[1] = put_pixels_x2_mmx;
|
|
|
990 put_pixels_tab[2] = put_pixels_y2_mmx;
|
|
|
991 put_pixels_tab[3] = put_pixels_xy2_mmx;
|
|
|
992
|
|
|
993 put_no_rnd_pixels_tab[0] = put_pixels_mmx;
|
|
|
994 put_no_rnd_pixels_tab[1] = put_no_rnd_pixels_x2_mmx;
|
|
|
995 put_no_rnd_pixels_tab[2] = put_no_rnd_pixels_y2_mmx;
|
|
|
996 put_no_rnd_pixels_tab[3] = put_no_rnd_pixels_xy2_mmx;
|
|
|
997
|
|
|
998 avg_pixels_tab[0] = avg_pixels_mmx;
|
|
|
999 avg_pixels_tab[1] = avg_pixels_x2_mmx;
|
|
|
1000 avg_pixels_tab[2] = avg_pixels_y2_mmx;
|
|
|
1001 avg_pixels_tab[3] = avg_pixels_xy2_mmx;
|
|
|
1002
|
|
|
1003 avg_no_rnd_pixels_tab[0] = avg_no_rnd_pixels_mmx;
|
|
|
1004 avg_no_rnd_pixels_tab[1] = avg_no_rnd_pixels_x2_mmx;
|
|
|
1005 avg_no_rnd_pixels_tab[2] = avg_no_rnd_pixels_y2_mmx;
|
|
|
1006 avg_no_rnd_pixels_tab[3] = avg_no_rnd_pixels_xy2_mmx;
|
|
|
1007
|
|
|
1008 sub_pixels_tab[0] = sub_pixels_mmx;
|
|
|
1009 sub_pixels_tab[1] = sub_pixels_x2_mmx;
|
|
|
1010 sub_pixels_tab[2] = sub_pixels_y2_mmx;
|
|
|
1011 sub_pixels_tab[3] = sub_pixels_xy2_mmx;
|
|
|
1012
|
|
|
1013 if (mm_flags & MM_MMXEXT) {
|
|
|
1014 pix_abs16x16 = pix_abs16x16_sse;
|
|
|
1015 }
|
|
|
1016
|
|
|
1017 if (mm_flags & MM_SSE) {
|
|
|
1018 put_pixels_tab[1] = put_pixels_x2_sse;
|
|
|
1019 put_pixels_tab[2] = put_pixels_y2_sse;
|
|
|
1020
|
|
|
1021 avg_pixels_tab[0] = avg_pixels_sse;
|
|
|
1022 avg_pixels_tab[1] = avg_pixels_x2_sse;
|
|
|
1023 avg_pixels_tab[2] = avg_pixels_y2_sse;
|
|
|
1024 avg_pixels_tab[3] = avg_pixels_xy2_sse;
|
|
|
1025
|
|
|
1026 sub_pixels_tab[1] = sub_pixels_x2_sse;
|
|
|
1027 sub_pixels_tab[2] = sub_pixels_y2_sse;
|
|
|
1028 } else if (mm_flags & MM_3DNOW) {
|
|
|
1029 put_pixels_tab[1] = put_pixels_x2_3dnow;
|
|
|
1030 put_pixels_tab[2] = put_pixels_y2_3dnow;
|
|
|
1031
|
|
|
1032 avg_pixels_tab[0] = avg_pixels_3dnow;
|
|
|
1033 avg_pixels_tab[1] = avg_pixels_x2_3dnow;
|
|
|
1034 avg_pixels_tab[2] = avg_pixels_y2_3dnow;
|
|
|
1035 avg_pixels_tab[3] = avg_pixels_xy2_3dnow;
|
|
|
1036
|
|
|
1037 sub_pixels_tab[1] = sub_pixels_x2_3dnow;
|
|
|
1038 sub_pixels_tab[2] = sub_pixels_y2_3dnow;
|
|
|
1039 }
|
|
|
1040 }
|
|
|
1041 }
|
