motion_est_alpha.c 9.57 KB
Newer Older
1 2 3 4
/*
 * Alpha optimized DSP utils
 * Copyright (c) 2002 Falk Hueffner <falk@debian.org>
 *
5 6 7
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
8 9
 * modify it under the terms of the GNU Lesser General Public
 * 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 14 15 16 17
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * 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 "dsputil_alpha.h"
23 24
#include "asm.h"

Diego Biurrun's avatar
Diego Biurrun committed
25
void get_pixels_mvi(int16_t *restrict block,
26 27 28 29 30 31 32 33 34
                    const uint8_t *restrict pixels, int line_size)
{
    int h = 8;

    do {
        uint64_t p;

        p = ldq(pixels);
        stq(unpkbw(p),       block);
35
        stq(unpkbw(p >> 32), block + 4);
36 37 38 39 40 41

        pixels += line_size;
        block += 8;
    } while (--h);
}

Diego Biurrun's avatar
Diego Biurrun committed
42
void diff_pixels_mvi(int16_t *block, const uint8_t *s1, const uint8_t *s2,
43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88
                     int stride) {
    int h = 8;
    uint64_t mask = 0x4040;

    mask |= mask << 16;
    mask |= mask << 32;
    do {
        uint64_t x, y, c, d, a;
        uint64_t signs;

        x = ldq(s1);
        y = ldq(s2);
        c = cmpbge(x, y);
        d = x - y;
        a = zap(mask, c);       /* We use 0x4040404040404040 here...  */
        d += 4 * a;             /* ...so we can use s4addq here.      */
        signs = zap(-1, c);

        stq(unpkbw(d)       | (unpkbw(signs)       << 8), block);
        stq(unpkbw(d >> 32) | (unpkbw(signs >> 32) << 8), block + 4);

        s1 += stride;
        s2 += stride;
        block += 8;
    } while (--h);
}

static inline uint64_t avg2(uint64_t a, uint64_t b)
{
    return (a | b) - (((a ^ b) & BYTE_VEC(0xfe)) >> 1);
}

static inline uint64_t avg4(uint64_t l1, uint64_t l2, uint64_t l3, uint64_t l4)
{
    uint64_t r1 = ((l1 & ~BYTE_VEC(0x03)) >> 2)
                + ((l2 & ~BYTE_VEC(0x03)) >> 2)
                + ((l3 & ~BYTE_VEC(0x03)) >> 2)
                + ((l4 & ~BYTE_VEC(0x03)) >> 2);
    uint64_t r2 = ((  (l1 & BYTE_VEC(0x03))
                    + (l2 & BYTE_VEC(0x03))
                    + (l3 & BYTE_VEC(0x03))
                    + (l4 & BYTE_VEC(0x03))
                    + BYTE_VEC(0x02)) >> 2) & BYTE_VEC(0x03);
    return r1 + r2;
}

89
int pix_abs8x8_mvi(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120
{
    int result = 0;

    if ((size_t) pix2 & 0x7) {
        /* works only when pix2 is actually unaligned */
        do {                    /* do 8 pixel a time */
            uint64_t p1, p2;

            p1  = ldq(pix1);
            p2  = uldq(pix2);
            result += perr(p1, p2);

            pix1 += line_size;
            pix2 += line_size;
        } while (--h);
    } else {
        do {
            uint64_t p1, p2;

            p1 = ldq(pix1);
            p2 = ldq(pix2);
            result += perr(p1, p2);

            pix1 += line_size;
            pix2 += line_size;
        } while (--h);
    }

    return result;
}

121
#if 0                           /* now done in assembly */
122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161
int pix_abs16x16_mvi(uint8_t *pix1, uint8_t *pix2, int line_size)
{
    int result = 0;
    int h = 16;

    if ((size_t) pix2 & 0x7) {
        /* works only when pix2 is actually unaligned */
        do {                    /* do 16 pixel a time */
            uint64_t p1_l, p1_r, p2_l, p2_r;
            uint64_t t;

            p1_l  = ldq(pix1);
            p1_r  = ldq(pix1 + 8);
            t     = ldq_u(pix2 + 8);
            p2_l  = extql(ldq_u(pix2), pix2) | extqh(t, pix2);
            p2_r  = extql(t, pix2) | extqh(ldq_u(pix2 + 16), pix2);
            pix1 += line_size;
            pix2 += line_size;

            result += perr(p1_l, p2_l)
                    + perr(p1_r, p2_r);
        } while (--h);
    } else {
        do {
            uint64_t p1_l, p1_r, p2_l, p2_r;

            p1_l = ldq(pix1);
            p1_r = ldq(pix1 + 8);
            p2_l = ldq(pix2);
            p2_r = ldq(pix2 + 8);
            pix1 += line_size;
            pix2 += line_size;

            result += perr(p1_l, p2_l)
                    + perr(p1_r, p2_r);
        } while (--h);
    }

    return result;
}
162
#endif
163

164
int pix_abs16x16_x2_mvi(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191
{
    int result = 0;
    uint64_t disalign = (size_t) pix2 & 0x7;

    switch (disalign) {
    case 0:
        do {
            uint64_t p1_l, p1_r, p2_l, p2_r;
            uint64_t l, r;

            p1_l = ldq(pix1);
            p1_r = ldq(pix1 + 8);
            l    = ldq(pix2);
            r    = ldq(pix2 + 8);
            p2_l = avg2(l, (l >> 8) | ((uint64_t) r << 56));
            p2_r = avg2(r, (r >> 8) | ((uint64_t) pix2[16] << 56));
            pix1 += line_size;
            pix2 += line_size;

            result += perr(p1_l, p2_l)
                    + perr(p1_r, p2_r);
        } while (--h);
        break;
    case 7:
        /* |.......l|lllllllr|rrrrrrr*|
           This case is special because disalign1 would be 8, which
           gets treated as 0 by extqh.  At least it is a bit faster
192
           that way :)  */
193 194 195 196 197 198 199 200 201 202 203 204 205
        do {
            uint64_t p1_l, p1_r, p2_l, p2_r;
            uint64_t l, m, r;

            p1_l = ldq(pix1);
            p1_r = ldq(pix1 + 8);
            l     = ldq_u(pix2);
            m     = ldq_u(pix2 + 8);
            r     = ldq_u(pix2 + 16);
            p2_l  = avg2(extql(l, disalign) | extqh(m, disalign), m);
            p2_r  = avg2(extql(m, disalign) | extqh(r, disalign), r);
            pix1 += line_size;
            pix2 += line_size;
206

207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236
            result += perr(p1_l, p2_l)
                    + perr(p1_r, p2_r);
        } while (--h);
        break;
    default:
        do {
            uint64_t disalign1 = disalign + 1;
            uint64_t p1_l, p1_r, p2_l, p2_r;
            uint64_t l, m, r;

            p1_l  = ldq(pix1);
            p1_r  = ldq(pix1 + 8);
            l     = ldq_u(pix2);
            m     = ldq_u(pix2 + 8);
            r     = ldq_u(pix2 + 16);
            p2_l  = avg2(extql(l, disalign) | extqh(m, disalign),
                         extql(l, disalign1) | extqh(m, disalign1));
            p2_r  = avg2(extql(m, disalign) | extqh(r, disalign),
                         extql(m, disalign1) | extqh(r, disalign1));
            pix1 += line_size;
            pix2 += line_size;

            result += perr(p1_l, p2_l)
                    + perr(p1_r, p2_r);
        } while (--h);
        break;
    }
    return result;
}

237
int pix_abs16x16_y2_mvi(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289
{
    int result = 0;

    if ((size_t) pix2 & 0x7) {
        uint64_t t, p2_l, p2_r;
        t     = ldq_u(pix2 + 8);
        p2_l  = extql(ldq_u(pix2), pix2) | extqh(t, pix2);
        p2_r  = extql(t, pix2) | extqh(ldq_u(pix2 + 16), pix2);

        do {
            uint64_t p1_l, p1_r, np2_l, np2_r;
            uint64_t t;

            p1_l  = ldq(pix1);
            p1_r  = ldq(pix1 + 8);
            pix2 += line_size;
            t     = ldq_u(pix2 + 8);
            np2_l = extql(ldq_u(pix2), pix2) | extqh(t, pix2);
            np2_r = extql(t, pix2) | extqh(ldq_u(pix2 + 16), pix2);

            result += perr(p1_l, avg2(p2_l, np2_l))
                    + perr(p1_r, avg2(p2_r, np2_r));

            pix1 += line_size;
            p2_l  = np2_l;
            p2_r  = np2_r;

        } while (--h);
    } else {
        uint64_t p2_l, p2_r;
        p2_l = ldq(pix2);
        p2_r = ldq(pix2 + 8);
        do {
            uint64_t p1_l, p1_r, np2_l, np2_r;

            p1_l = ldq(pix1);
            p1_r = ldq(pix1 + 8);
            pix2 += line_size;
            np2_l = ldq(pix2);
            np2_r = ldq(pix2 + 8);

            result += perr(p1_l, avg2(p2_l, np2_l))
                    + perr(p1_r, avg2(p2_r, np2_r));

            pix1 += line_size;
            p2_l  = np2_l;
            p2_r  = np2_r;
        } while (--h);
    }
    return result;
}

290
int pix_abs16x16_xy2_mvi(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
291 292
{
    int result = 0;
293

294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345
    uint64_t p1_l, p1_r;
    uint64_t p2_l, p2_r, p2_x;

    p1_l = ldq(pix1);
    p1_r = ldq(pix1 + 8);

    if ((size_t) pix2 & 0x7) { /* could be optimized a lot */
        p2_l = uldq(pix2);
        p2_r = uldq(pix2 + 8);
        p2_x = (uint64_t) pix2[16] << 56;
    } else {
        p2_l = ldq(pix2);
        p2_r = ldq(pix2 + 8);
        p2_x = ldq(pix2 + 16) << 56;
    }

    do {
        uint64_t np1_l, np1_r;
        uint64_t np2_l, np2_r, np2_x;

        pix1 += line_size;
        pix2 += line_size;

        np1_l = ldq(pix1);
        np1_r = ldq(pix1 + 8);

        if ((size_t) pix2 & 0x7) { /* could be optimized a lot */
            np2_l = uldq(pix2);
            np2_r = uldq(pix2 + 8);
            np2_x = (uint64_t) pix2[16] << 56;
        } else {
            np2_l = ldq(pix2);
            np2_r = ldq(pix2 + 8);
            np2_x = ldq(pix2 + 16) << 56;
        }

        result += perr(p1_l,
                       avg4( p2_l, ( p2_l >> 8) | ((uint64_t)  p2_r << 56),
                            np2_l, (np2_l >> 8) | ((uint64_t) np2_r << 56)))
                + perr(p1_r,
                       avg4( p2_r, ( p2_r >> 8) | ((uint64_t)  p2_x),
                            np2_r, (np2_r >> 8) | ((uint64_t) np2_x)));

        p1_l = np1_l;
        p1_r = np1_r;
        p2_l = np2_l;
        p2_r = np2_r;
        p2_x = np2_x;
    } while (--h);

    return result;
}