motion_est_mmx.c 13 KB
Newer Older
1 2
/*
 * MMX optimized motion estimation
Fabrice Bellard's avatar
Fabrice Bellard committed
3
 * Copyright (c) 2001 Fabrice Bellard.
4
 *
Fabrice Bellard's avatar
Fabrice Bellard committed
5 6 7 8
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2 of the License, or (at your option) any later version.
9
 *
Fabrice Bellard's avatar
Fabrice Bellard committed
10
 * This library is distributed in the hope that it will be useful,
11
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
Fabrice Bellard's avatar
Fabrice Bellard committed
12 13
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
14
 *
Fabrice Bellard's avatar
Fabrice Bellard committed
15 16 17
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
18
 *
19
 * mostly by Michael Niedermayer <michaelni@gmx.at>
20 21 22
 */
#include "../dsputil.h"

23
static const __attribute__ ((aligned(8))) uint64_t round_tab[3]={
24 25 26 27
0x0000000000000000,
0x0001000100010001,
0x0002000200020002,
};
28

Michael Niedermayer's avatar
Michael Niedermayer committed
29 30
static __attribute__ ((aligned(8))) uint64_t bone= 0x0101010101010101LL;

31
static inline void sad8_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h)
32
{
33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64
    int len= -(stride<<h);
    asm volatile(
        ".balign 16			\n\t"
        "1:				\n\t"
        "movq (%1, %%eax), %%mm0	\n\t"
        "movq (%2, %%eax), %%mm2	\n\t"
        "movq (%2, %%eax), %%mm4	\n\t"
        "addl %3, %%eax			\n\t"
        "psubusb %%mm0, %%mm2		\n\t"
        "psubusb %%mm4, %%mm0		\n\t"
        "movq (%1, %%eax), %%mm1	\n\t"
        "movq (%2, %%eax), %%mm3	\n\t"
        "movq (%2, %%eax), %%mm5	\n\t"
        "psubusb %%mm1, %%mm3		\n\t"
        "psubusb %%mm5, %%mm1		\n\t"
        "por %%mm2, %%mm0		\n\t"
        "por %%mm1, %%mm3		\n\t"
        "movq %%mm0, %%mm1		\n\t"
        "movq %%mm3, %%mm2		\n\t"
        "punpcklbw %%mm7, %%mm0		\n\t"
        "punpckhbw %%mm7, %%mm1		\n\t"
        "punpcklbw %%mm7, %%mm3		\n\t"
        "punpckhbw %%mm7, %%mm2		\n\t"
        "paddw %%mm1, %%mm0		\n\t"
        "paddw %%mm3, %%mm2		\n\t"
        "paddw %%mm2, %%mm0		\n\t"
        "paddw %%mm0, %%mm6		\n\t"
        "addl %3, %%eax			\n\t"
        " js 1b				\n\t"
        : "+a" (len)
        : "r" (blk1 - len), "r" (blk2 - len), "r" (stride)
    );
65 66
}

67
static inline void sad8_mmx2(uint8_t *blk1, uint8_t *blk2, int stride, int h)
68
{
69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86
    int len= -(stride<<h);
    asm volatile(
        ".balign 16			\n\t"
        "1:				\n\t"
        "movq (%1, %%eax), %%mm0	\n\t"
        "movq (%2, %%eax), %%mm2	\n\t"
        "psadbw %%mm2, %%mm0		\n\t"
        "addl %3, %%eax			\n\t"
        "movq (%1, %%eax), %%mm1	\n\t"
        "movq (%2, %%eax), %%mm3	\n\t"
        "psadbw %%mm1, %%mm3		\n\t"
        "paddw %%mm3, %%mm0		\n\t"
        "paddw %%mm0, %%mm6		\n\t"
        "addl %3, %%eax			\n\t"
        " js 1b				\n\t"
        : "+a" (len)
        : "r" (blk1 - len), "r" (blk2 - len), "r" (stride)
    );
87 88
}

89
static inline void sad8_2_mmx2(uint8_t *blk1a, uint8_t *blk1b, uint8_t *blk2, int stride, int h)
90
{
91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112
    int len= -(stride<<h);
    asm volatile(
        ".balign 16			\n\t"
        "1:				\n\t"
        "movq (%1, %%eax), %%mm0	\n\t"
        "movq (%2, %%eax), %%mm2	\n\t"
        "pavgb %%mm2, %%mm0		\n\t"
        "movq (%3, %%eax), %%mm2	\n\t"
        "psadbw %%mm2, %%mm0		\n\t"
        "addl %4, %%eax			\n\t"
        "movq (%1, %%eax), %%mm1	\n\t"
        "movq (%2, %%eax), %%mm3	\n\t"
        "pavgb %%mm1, %%mm3		\n\t"
        "movq (%3, %%eax), %%mm1	\n\t"
        "psadbw %%mm1, %%mm3		\n\t"
        "paddw %%mm3, %%mm0		\n\t"
        "paddw %%mm0, %%mm6		\n\t"
        "addl %4, %%eax			\n\t"
        " js 1b				\n\t"
        : "+a" (len)
        : "r" (blk1a - len), "r" (blk1b -len), "r" (blk2 - len), "r" (stride)
    );
113 114
}

115
static inline void sad8_4_mmx2(uint8_t *blk1, uint8_t *blk2, int stride, int h)
116 117 118 119
{ //FIXME reuse src
    int len= -(stride<<h);
    asm volatile(
        ".balign 16			\n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
120
        "movq "MANGLE(bone)", %%mm5	\n\t"
121
        "1:				\n\t"
122 123 124 125 126 127
        "movq (%1, %%eax), %%mm0	\n\t"
        "movq (%2, %%eax), %%mm2	\n\t"
        "movq 1(%1, %%eax), %%mm1	\n\t"
        "movq 1(%2, %%eax), %%mm3	\n\t"
        "pavgb %%mm2, %%mm0		\n\t"
        "pavgb %%mm1, %%mm3		\n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
128
        "psubusb %%mm5, %%mm3		\n\t"
129 130 131 132 133 134 135 136 137 138
        "pavgb %%mm3, %%mm0		\n\t"
        "movq (%3, %%eax), %%mm2	\n\t"
        "psadbw %%mm2, %%mm0		\n\t"
        "addl %4, %%eax			\n\t"
        "movq (%1, %%eax), %%mm1	\n\t"
        "movq (%2, %%eax), %%mm3	\n\t"
        "movq 1(%1, %%eax), %%mm2	\n\t"
        "movq 1(%2, %%eax), %%mm4	\n\t"
        "pavgb %%mm3, %%mm1		\n\t"
        "pavgb %%mm4, %%mm2		\n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
139
        "psubusb %%mm5, %%mm2		\n\t"
140 141 142 143 144 145 146 147 148 149
        "pavgb %%mm1, %%mm2		\n\t"
        "movq (%3, %%eax), %%mm1	\n\t"
        "psadbw %%mm1, %%mm2		\n\t"
        "paddw %%mm2, %%mm0		\n\t"
        "paddw %%mm0, %%mm6		\n\t"
        "addl %4, %%eax			\n\t"
        " js 1b				\n\t"
        : "+a" (len)
        : "r" (blk1 - len), "r" (blk1 - len + stride), "r" (blk2 - len), "r" (stride)
    );
150 151
}

152
static inline void sad8_2_mmx(uint8_t *blk1a, uint8_t *blk1b, uint8_t *blk2, int stride, int h)
153
{
154 155 156 157 158 159 160 161 162 163 164 165 166 167
    int len= -(stride<<h);
    asm volatile(
        ".balign 16			\n\t"
        "1:				\n\t"
        "movq (%1, %%eax), %%mm0	\n\t"
        "movq (%2, %%eax), %%mm1	\n\t"
        "movq (%1, %%eax), %%mm2	\n\t"
        "movq (%2, %%eax), %%mm3	\n\t"
        "punpcklbw %%mm7, %%mm0		\n\t"
        "punpcklbw %%mm7, %%mm1		\n\t"
        "punpckhbw %%mm7, %%mm2		\n\t"
        "punpckhbw %%mm7, %%mm3		\n\t"
        "paddw %%mm0, %%mm1		\n\t"
        "paddw %%mm2, %%mm3		\n\t"
168
        "movq (%3, %%eax), %%mm4	\n\t"
169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187
        "movq (%3, %%eax), %%mm2	\n\t"
        "paddw %%mm5, %%mm1		\n\t"
        "paddw %%mm5, %%mm3		\n\t"
        "psrlw $1, %%mm1		\n\t"
        "psrlw $1, %%mm3		\n\t"
        "packuswb %%mm3, %%mm1		\n\t"
        "psubusb %%mm1, %%mm4		\n\t"
        "psubusb %%mm2, %%mm1		\n\t"
        "por %%mm4, %%mm1		\n\t"
        "movq %%mm1, %%mm0		\n\t"
        "punpcklbw %%mm7, %%mm0		\n\t"
        "punpckhbw %%mm7, %%mm1		\n\t"
        "paddw %%mm1, %%mm0		\n\t"
        "paddw %%mm0, %%mm6		\n\t"
        "addl %4, %%eax			\n\t"
        " js 1b				\n\t"
        : "+a" (len)
        : "r" (blk1a - len), "r" (blk1b -len), "r" (blk2 - len), "r" (stride)
    );
188 189
}

190
static inline void sad8_4_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h)
191
{
192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217
    int len= -(stride<<h);
    asm volatile(
        ".balign 16			\n\t"
        "1:				\n\t"
        "movq (%1, %%eax), %%mm0	\n\t"
        "movq (%2, %%eax), %%mm1	\n\t"
        "movq %%mm0, %%mm4		\n\t"
        "movq %%mm1, %%mm2		\n\t"
        "punpcklbw %%mm7, %%mm0		\n\t"
        "punpcklbw %%mm7, %%mm1		\n\t"
        "punpckhbw %%mm7, %%mm4		\n\t"
        "punpckhbw %%mm7, %%mm2		\n\t"
        "paddw %%mm1, %%mm0		\n\t"
        "paddw %%mm2, %%mm4		\n\t"
        "movq 1(%1, %%eax), %%mm2	\n\t"
        "movq 1(%2, %%eax), %%mm3	\n\t"
        "movq %%mm2, %%mm1		\n\t"
        "punpcklbw %%mm7, %%mm2		\n\t"
        "punpckhbw %%mm7, %%mm1		\n\t"
        "paddw %%mm0, %%mm2		\n\t"
        "paddw %%mm4, %%mm1		\n\t"
        "movq %%mm3, %%mm4		\n\t"
        "punpcklbw %%mm7, %%mm3		\n\t"
        "punpckhbw %%mm7, %%mm4		\n\t"
        "paddw %%mm3, %%mm2		\n\t"
        "paddw %%mm4, %%mm1		\n\t"
218 219
        "movq (%3, %%eax), %%mm3	\n\t"
        "movq (%3, %%eax), %%mm4	\n\t"
220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237
        "paddw %%mm5, %%mm2		\n\t"
        "paddw %%mm5, %%mm1		\n\t"
        "psrlw $2, %%mm2		\n\t"
        "psrlw $2, %%mm1		\n\t"
        "packuswb %%mm1, %%mm2		\n\t"
        "psubusb %%mm2, %%mm3		\n\t"
        "psubusb %%mm4, %%mm2		\n\t"
        "por %%mm3, %%mm2		\n\t"
        "movq %%mm2, %%mm0		\n\t"
        "punpcklbw %%mm7, %%mm0		\n\t"
        "punpckhbw %%mm7, %%mm2		\n\t"
        "paddw %%mm2, %%mm0		\n\t"
        "paddw %%mm0, %%mm6		\n\t"
        "addl %4, %%eax			\n\t"
        " js 1b				\n\t"
        : "+a" (len)
        : "r" (blk1 - len), "r" (blk1 -len + stride), "r" (blk2 - len), "r" (stride)
    );
238 239
}

240
static inline int sum_mmx(void)
241
{
242 243 244 245 246 247 248 249 250 251 252 253
    int ret;
    asm volatile(
        "movq %%mm6, %%mm0		\n\t"
        "psrlq $32, %%mm6		\n\t"
        "paddw %%mm0, %%mm6		\n\t"
        "movq %%mm6, %%mm0		\n\t"
        "psrlq $16, %%mm6		\n\t"
        "paddw %%mm0, %%mm6		\n\t"
        "movd %%mm6, %0			\n\t"
        : "=r" (ret)
    );
    return ret&0xFFFF;
254 255
}

256
static inline int sum_mmx2(void)
257
{
258 259 260 261 262 263
    int ret;
    asm volatile(
        "movd %%mm6, %0			\n\t"
        : "=r" (ret)
    );
    return ret;
264 265
}

266

267
#define PIX_SAD(suf)\
268
static int pix_abs8x8_ ## suf(uint8_t *blk2, uint8_t *blk1, int stride)\
269 270 271 272 273 274 275 276
{\
    asm volatile("pxor %%mm7, %%mm7		\n\t"\
                 "pxor %%mm6, %%mm6		\n\t":);\
\
    sad8_ ## suf(blk1, blk2, stride, 3);\
\
    return sum_ ## suf();\
}\
277
static int sad8x8_ ## suf(void *s, uint8_t *blk2, uint8_t *blk1, int stride)\
Michael Niedermayer's avatar
Michael Niedermayer committed
278 279 280 281 282 283 284 285
{\
    asm volatile("pxor %%mm7, %%mm7		\n\t"\
                 "pxor %%mm6, %%mm6		\n\t":);\
\
    sad8_ ## suf(blk1, blk2, stride, 3);\
\
    return sum_ ## suf();\
}\
286
\
287
static int pix_abs8x8_x2_ ## suf(uint8_t *blk2, uint8_t *blk1, int stride)\
288 289 290 291 292 293 294
{\
    asm volatile("pxor %%mm7, %%mm7		\n\t"\
                 "pxor %%mm6, %%mm6		\n\t"\
                 "movq %0, %%mm5		\n\t"\
                 :: "m"(round_tab[1]) \
                 );\
\
295
    sad8_2_ ## suf(blk1, blk1+1, blk2, stride, 3);\
296 297 298 299
\
    return sum_ ## suf();\
}\
\
300
static int pix_abs8x8_y2_ ## suf(uint8_t *blk2, uint8_t *blk1, int stride)\
301 302 303 304 305 306 307 308 309 310 311 312
{\
    asm volatile("pxor %%mm7, %%mm7		\n\t"\
                 "pxor %%mm6, %%mm6		\n\t"\
                 "movq %0, %%mm5		\n\t"\
                 :: "m"(round_tab[1]) \
                 );\
\
    sad8_2_ ## suf(blk1, blk1+stride, blk2, stride, 3);\
\
    return sum_ ## suf();\
}\
\
313
static int pix_abs8x8_xy2_ ## suf(uint8_t *blk2, uint8_t *blk1, int stride)\
314 315 316 317 318 319 320 321 322 323 324 325
{\
    asm volatile("pxor %%mm7, %%mm7		\n\t"\
                 "pxor %%mm6, %%mm6		\n\t"\
                 "movq %0, %%mm5		\n\t"\
                 :: "m"(round_tab[2]) \
                 );\
\
    sad8_4_ ## suf(blk1, blk2, stride, 3);\
\
    return sum_ ## suf();\
}\
\
326
static int pix_abs16x16_ ## suf(uint8_t *blk2, uint8_t *blk1, int stride)\
327 328 329 330 331 332 333 334 335
{\
    asm volatile("pxor %%mm7, %%mm7		\n\t"\
                 "pxor %%mm6, %%mm6		\n\t":);\
\
    sad8_ ## suf(blk1  , blk2  , stride, 4);\
    sad8_ ## suf(blk1+8, blk2+8, stride, 4);\
\
    return sum_ ## suf();\
}\
336
static int sad16x16_ ## suf(void *s, uint8_t *blk2, uint8_t *blk1, int stride)\
Michael Niedermayer's avatar
Michael Niedermayer committed
337 338 339 340 341 342 343 344 345
{\
    asm volatile("pxor %%mm7, %%mm7		\n\t"\
                 "pxor %%mm6, %%mm6		\n\t":);\
\
    sad8_ ## suf(blk1  , blk2  , stride, 4);\
    sad8_ ## suf(blk1+8, blk2+8, stride, 4);\
\
    return sum_ ## suf();\
}\
346
static int pix_abs16x16_x2_ ## suf(uint8_t *blk2, uint8_t *blk1, int stride)\
347 348 349 350 351 352 353 354 355 356 357 358
{\
    asm volatile("pxor %%mm7, %%mm7		\n\t"\
                 "pxor %%mm6, %%mm6		\n\t"\
                 "movq %0, %%mm5		\n\t"\
                 :: "m"(round_tab[1]) \
                 );\
\
    sad8_2_ ## suf(blk1  , blk1+1, blk2  , stride, 4);\
    sad8_2_ ## suf(blk1+8, blk1+9, blk2+8, stride, 4);\
\
    return sum_ ## suf();\
}\
359
static int pix_abs16x16_y2_ ## suf(uint8_t *blk2, uint8_t *blk1, int stride)\
360 361 362 363 364 365 366 367 368 369 370 371
{\
    asm volatile("pxor %%mm7, %%mm7		\n\t"\
                 "pxor %%mm6, %%mm6		\n\t"\
                 "movq %0, %%mm5		\n\t"\
                 :: "m"(round_tab[1]) \
                 );\
\
    sad8_2_ ## suf(blk1  , blk1+stride,  blk2  , stride, 4);\
    sad8_2_ ## suf(blk1+8, blk1+stride+8,blk2+8, stride, 4);\
\
    return sum_ ## suf();\
}\
372
static int pix_abs16x16_xy2_ ## suf(uint8_t *blk2, uint8_t *blk1, int stride)\
373 374 375 376 377 378 379 380 381 382 383 384
{\
    asm volatile("pxor %%mm7, %%mm7		\n\t"\
                 "pxor %%mm6, %%mm6		\n\t"\
                 "movq %0, %%mm5		\n\t"\
                 :: "m"(round_tab[2]) \
                 );\
\
    sad8_4_ ## suf(blk1  , blk2  , stride, 4);\
    sad8_4_ ## suf(blk1+8, blk2+8, stride, 4);\
\
    return sum_ ## suf();\
}\
385

386 387
PIX_SAD(mmx)
PIX_SAD(mmx2)
388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429

void dsputil_init_pix_mmx(DSPContext* c, unsigned mask)
{
    if (mm_flags & MM_MMX) {
        c->pix_abs16x16     = pix_abs16x16_mmx;
        c->pix_abs16x16_x2  = pix_abs16x16_x2_mmx;
        c->pix_abs16x16_y2  = pix_abs16x16_y2_mmx;
        c->pix_abs16x16_xy2 = pix_abs16x16_xy2_mmx;
        c->pix_abs8x8     = pix_abs8x8_mmx;
        c->pix_abs8x8_x2  = pix_abs8x8_x2_mmx;
        c->pix_abs8x8_y2  = pix_abs8x8_y2_mmx;
        c->pix_abs8x8_xy2 = pix_abs8x8_xy2_mmx;

	c->sad[0]= sad16x16_mmx;
        c->sad[1]= sad8x8_mmx;
    }
    if (mm_flags & MM_MMXEXT) {
	c->pix_abs16x16     = pix_abs16x16_mmx2;
	c->pix_abs16x16_x2  = pix_abs16x16_x2_mmx2;
	c->pix_abs16x16_y2  = pix_abs16x16_y2_mmx2;
	c->pix_abs16x16_xy2 = pix_abs16x16_xy2_mmx2;
	c->pix_abs8x8     = pix_abs8x8_mmx2;
	c->pix_abs8x8_x2  = pix_abs8x8_x2_mmx2;
	c->pix_abs8x8_y2  = pix_abs8x8_y2_mmx2;
	c->pix_abs8x8_xy2 = pix_abs8x8_xy2_mmx2;

	c->sad[0]= sad16x16_mmx2;
	c->sad[1]= sad8x8_mmx2;
    }
}

void dsputil_set_bit_exact_pix_mmx(DSPContext* c, unsigned mask)
{
    if (mm_flags & MM_MMXEXT) {
	c->pix_abs16x16_x2  = pix_abs16x16_x2_mmx;
	c->pix_abs16x16_y2  = pix_abs16x16_y2_mmx;
	c->pix_abs16x16_xy2 = pix_abs16x16_xy2_mmx;
	c->pix_abs8x8_x2  = pix_abs8x8_x2_mmx;
	c->pix_abs8x8_y2  = pix_abs8x8_y2_mmx;
	c->pix_abs8x8_xy2 = pix_abs8x8_xy2_mmx;
    }
}