h264dsp_mmx.c 58.2 KB
Newer Older
1 2 3
/*
 * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
 *
4 5 6
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
7 8
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
9
 * version 2.1 of the License, or (at your option) any later version.
10
 *
11
 * FFmpeg is distributed in the hope that it will be useful,
12 13 14 15 16
 * 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
17
 * License along with FFmpeg; if not, write to the Free Software
18
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 20 21 22 23 24 25 26 27 28 29 30 31 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
 */


/***********************************/
/* IDCT */

/* in/out: mma=mma+mmb, mmb=mmb-mma */
#define SUMSUB_BA( a, b ) \
    "paddw "#b", "#a" \n\t"\
    "paddw "#b", "#b" \n\t"\
    "psubw "#a", "#b" \n\t"

#define SUMSUB_BADC( a, b, c, d ) \
    "paddw "#b", "#a" \n\t"\
    "paddw "#d", "#c" \n\t"\
    "paddw "#b", "#b" \n\t"\
    "paddw "#d", "#d" \n\t"\
    "psubw "#a", "#b" \n\t"\
    "psubw "#c", "#d" \n\t"

#define SUMSUBD2_AB( a, b, t ) \
    "movq  "#b", "#t" \n\t"\
    "psraw  $1 , "#b" \n\t"\
    "paddw "#a", "#b" \n\t"\
    "psraw  $1 , "#a" \n\t"\
    "psubw "#t", "#a" \n\t"

#define IDCT4_1D( s02, s13, d02, d13, t ) \
    SUMSUB_BA  ( s02, d02 )\
    SUMSUBD2_AB( s13, d13, t )\
    SUMSUB_BADC( d13, s02, s13, d02 )

#define TRANSPOSE4(a,b,c,d,t)\
    SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\
    SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\
    SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\
    SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */

#define STORE_DIFF_4P( p, t, z ) \
    "psraw      $6,     "#p" \n\t"\
    "movd       (%0),   "#t" \n\t"\
    "punpcklbw "#z",    "#t" \n\t"\
    "paddsw    "#t",    "#p" \n\t"\
    "packuswb  "#z",    "#p" \n\t"\
    "movd      "#p",    (%0) \n\t"

lorenm's avatar
lorenm committed
65
static void ff_h264_idct_add_mmx(uint8_t *dst, int16_t *block, int stride)
66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103
{
    /* Load dct coeffs */
    asm volatile(
        "movq   (%0), %%mm0 \n\t"
        "movq  8(%0), %%mm1 \n\t"
        "movq 16(%0), %%mm2 \n\t"
        "movq 24(%0), %%mm3 \n\t"
    :: "r"(block) );

    asm volatile(
        /* mm1=s02+s13  mm2=s02-s13  mm4=d02+d13  mm0=d02-d13 */
        IDCT4_1D( %%mm2, %%mm1, %%mm0, %%mm3, %%mm4 )

        "movq      %0,    %%mm6 \n\t"
        /* in: 1,4,0,2  out: 1,2,3,0 */
        TRANSPOSE4( %%mm3, %%mm1, %%mm0, %%mm2, %%mm4 )

        "paddw     %%mm6, %%mm3 \n\t"

        /* mm2=s02+s13  mm3=s02-s13  mm4=d02+d13  mm1=d02-d13 */
        IDCT4_1D( %%mm4, %%mm2, %%mm3, %%mm0, %%mm1 )

        "pxor %%mm7, %%mm7    \n\t"
    :: "m"(ff_pw_32));

    asm volatile(
    STORE_DIFF_4P( %%mm0, %%mm1, %%mm7)
        "add %1, %0             \n\t"
    STORE_DIFF_4P( %%mm2, %%mm1, %%mm7)
        "add %1, %0             \n\t"
    STORE_DIFF_4P( %%mm3, %%mm1, %%mm7)
        "add %1, %0             \n\t"
    STORE_DIFF_4P( %%mm4, %%mm1, %%mm7)
        : "+r"(dst)
        : "r" ((long)stride)
    );
}

lorenm's avatar
lorenm committed
104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178
static inline void h264_idct8_1d(int16_t *block)
{
    asm volatile(
        "movq 112(%0), %%mm7  \n\t"
        "movq  80(%0), %%mm5  \n\t"
        "movq  48(%0), %%mm3  \n\t"
        "movq  16(%0), %%mm1  \n\t"

        "movq   %%mm7, %%mm4  \n\t"
        "movq   %%mm3, %%mm6  \n\t"
        "movq   %%mm5, %%mm0  \n\t"
        "movq   %%mm7, %%mm2  \n\t"
        "psraw  $1,    %%mm4  \n\t"
        "psraw  $1,    %%mm6  \n\t"
        "psubw  %%mm7, %%mm0  \n\t"
        "psubw  %%mm6, %%mm2  \n\t"
        "psubw  %%mm4, %%mm0  \n\t"
        "psubw  %%mm3, %%mm2  \n\t"
        "psubw  %%mm3, %%mm0  \n\t"
        "paddw  %%mm1, %%mm2  \n\t"

        "movq   %%mm5, %%mm4  \n\t"
        "movq   %%mm1, %%mm6  \n\t"
        "psraw  $1,    %%mm4  \n\t"
        "psraw  $1,    %%mm6  \n\t"
        "paddw  %%mm5, %%mm4  \n\t"
        "paddw  %%mm1, %%mm6  \n\t"
        "paddw  %%mm7, %%mm4  \n\t"
        "paddw  %%mm5, %%mm6  \n\t"
        "psubw  %%mm1, %%mm4  \n\t"
        "paddw  %%mm3, %%mm6  \n\t"

        "movq   %%mm0, %%mm1  \n\t"
        "movq   %%mm4, %%mm3  \n\t"
        "movq   %%mm2, %%mm5  \n\t"
        "movq   %%mm6, %%mm7  \n\t"
        "psraw  $2,    %%mm6  \n\t"
        "psraw  $2,    %%mm3  \n\t"
        "psraw  $2,    %%mm5  \n\t"
        "psraw  $2,    %%mm0  \n\t"
        "paddw  %%mm6, %%mm1  \n\t"
        "paddw  %%mm2, %%mm3  \n\t"
        "psubw  %%mm4, %%mm5  \n\t"
        "psubw  %%mm0, %%mm7  \n\t"

        "movq  32(%0), %%mm2  \n\t"
        "movq  96(%0), %%mm6  \n\t"
        "movq   %%mm2, %%mm4  \n\t"
        "movq   %%mm6, %%mm0  \n\t"
        "psraw  $1,    %%mm4  \n\t"
        "psraw  $1,    %%mm6  \n\t"
        "psubw  %%mm0, %%mm4  \n\t"
        "paddw  %%mm2, %%mm6  \n\t"

        "movq    (%0), %%mm2  \n\t"
        "movq  64(%0), %%mm0  \n\t"
        SUMSUB_BA( %%mm0, %%mm2 )
        SUMSUB_BA( %%mm6, %%mm0 )
        SUMSUB_BA( %%mm4, %%mm2 )
        SUMSUB_BA( %%mm7, %%mm6 )
        SUMSUB_BA( %%mm5, %%mm4 )
        SUMSUB_BA( %%mm3, %%mm2 )
        SUMSUB_BA( %%mm1, %%mm0 )
        :: "r"(block)
    );
}

static void ff_h264_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride)
{
    int i;
    int16_t __attribute__ ((aligned(8))) b2[64];

    block[0] += 32;

    for(i=0; i<2; i++){
michael's avatar
michael committed
179
        DECLARE_ALIGNED_8(uint64_t, tmp);
lorenm's avatar
lorenm committed
180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230

        h264_idct8_1d(block+4*i);

        asm volatile(
            "movq   %%mm7,    %0   \n\t"
            TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 )
            "movq   %%mm0,  8(%1)  \n\t"
            "movq   %%mm6, 24(%1)  \n\t"
            "movq   %%mm7, 40(%1)  \n\t"
            "movq   %%mm4, 56(%1)  \n\t"
            "movq    %0,    %%mm7  \n\t"
            TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 )
            "movq   %%mm7,   (%1)  \n\t"
            "movq   %%mm1, 16(%1)  \n\t"
            "movq   %%mm0, 32(%1)  \n\t"
            "movq   %%mm3, 48(%1)  \n\t"
            : "=m"(tmp)
            : "r"(b2+32*i)
            : "memory"
        );
    }

    for(i=0; i<2; i++){
        h264_idct8_1d(b2+4*i);

        asm volatile(
            "psraw     $6, %%mm7  \n\t"
            "psraw     $6, %%mm6  \n\t"
            "psraw     $6, %%mm5  \n\t"
            "psraw     $6, %%mm4  \n\t"
            "psraw     $6, %%mm3  \n\t"
            "psraw     $6, %%mm2  \n\t"
            "psraw     $6, %%mm1  \n\t"
            "psraw     $6, %%mm0  \n\t"

            "movq   %%mm7,    (%0)  \n\t"
            "movq   %%mm5,  16(%0)  \n\t"
            "movq   %%mm3,  32(%0)  \n\t"
            "movq   %%mm1,  48(%0)  \n\t"
            "movq   %%mm0,  64(%0)  \n\t"
            "movq   %%mm2,  80(%0)  \n\t"
            "movq   %%mm4,  96(%0)  \n\t"
            "movq   %%mm6, 112(%0)  \n\t"
            :: "r"(b2+4*i)
            : "memory"
        );
    }

    add_pixels_clamped_mmx(b2, dst, stride);
}

lorenm's avatar
lorenm committed
231
static void ff_h264_idct_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266
{
    int dc = (block[0] + 32) >> 6;
    asm volatile(
        "movd          %0, %%mm0 \n\t"
        "pshufw $0, %%mm0, %%mm0 \n\t"
        "pxor       %%mm1, %%mm1 \n\t"
        "psubw      %%mm0, %%mm1 \n\t"
        "packuswb   %%mm0, %%mm0 \n\t"
        "packuswb   %%mm1, %%mm1 \n\t"
        ::"r"(dc)
    );
    asm volatile(
        "movd          %0, %%mm2 \n\t"
        "movd          %1, %%mm3 \n\t"
        "movd          %2, %%mm4 \n\t"
        "movd          %3, %%mm5 \n\t"
        "paddusb    %%mm0, %%mm2 \n\t"
        "paddusb    %%mm0, %%mm3 \n\t"
        "paddusb    %%mm0, %%mm4 \n\t"
        "paddusb    %%mm0, %%mm5 \n\t"
        "psubusb    %%mm1, %%mm2 \n\t"
        "psubusb    %%mm1, %%mm3 \n\t"
        "psubusb    %%mm1, %%mm4 \n\t"
        "psubusb    %%mm1, %%mm5 \n\t"
        "movd       %%mm2, %0    \n\t"
        "movd       %%mm3, %1    \n\t"
        "movd       %%mm4, %2    \n\t"
        "movd       %%mm5, %3    \n\t"
        :"+m"(*(uint32_t*)(dst+0*stride)),
         "+m"(*(uint32_t*)(dst+1*stride)),
         "+m"(*(uint32_t*)(dst+2*stride)),
         "+m"(*(uint32_t*)(dst+3*stride))
    );
}

lorenm's avatar
lorenm committed
267
static void ff_h264_idct8_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305
{
    int dc = (block[0] + 32) >> 6;
    int y;
    asm volatile(
        "movd          %0, %%mm0 \n\t"
        "pshufw $0, %%mm0, %%mm0 \n\t"
        "pxor       %%mm1, %%mm1 \n\t"
        "psubw      %%mm0, %%mm1 \n\t"
        "packuswb   %%mm0, %%mm0 \n\t"
        "packuswb   %%mm1, %%mm1 \n\t"
        ::"r"(dc)
    );
    for(y=2; y--; dst += 4*stride){
    asm volatile(
        "movq          %0, %%mm2 \n\t"
        "movq          %1, %%mm3 \n\t"
        "movq          %2, %%mm4 \n\t"
        "movq          %3, %%mm5 \n\t"
        "paddusb    %%mm0, %%mm2 \n\t"
        "paddusb    %%mm0, %%mm3 \n\t"
        "paddusb    %%mm0, %%mm4 \n\t"
        "paddusb    %%mm0, %%mm5 \n\t"
        "psubusb    %%mm1, %%mm2 \n\t"
        "psubusb    %%mm1, %%mm3 \n\t"
        "psubusb    %%mm1, %%mm4 \n\t"
        "psubusb    %%mm1, %%mm5 \n\t"
        "movq       %%mm2, %0    \n\t"
        "movq       %%mm3, %1    \n\t"
        "movq       %%mm4, %2    \n\t"
        "movq       %%mm5, %3    \n\t"
        :"+m"(*(uint64_t*)(dst+0*stride)),
         "+m"(*(uint64_t*)(dst+1*stride)),
         "+m"(*(uint64_t*)(dst+2*stride)),
         "+m"(*(uint64_t*)(dst+3*stride))
    );
    }
}

306 307 308 309 310 311 312 313 314 315 316 317 318 319

/***********************************/
/* deblocking */

// out: o = |x-y|>a
// clobbers: t
#define DIFF_GT_MMX(x,y,a,o,t)\
    "movq     "#y", "#t"  \n\t"\
    "movq     "#x", "#o"  \n\t"\
    "psubusb  "#x", "#t"  \n\t"\
    "psubusb  "#y", "#o"  \n\t"\
    "por      "#t", "#o"  \n\t"\
    "psubusb  "#a", "#o"  \n\t"

michael's avatar
michael committed
320 321 322 323 324 325 326 327 328 329 330
// out: o = |x-y|>a
// clobbers: t
#define DIFF_GT2_MMX(x,y,a,o,t)\
    "movq     "#y", "#t"  \n\t"\
    "movq     "#x", "#o"  \n\t"\
    "psubusb  "#x", "#t"  \n\t"\
    "psubusb  "#y", "#o"  \n\t"\
    "psubusb  "#a", "#t"  \n\t"\
    "psubusb  "#a", "#o"  \n\t"\
    "pcmpeqb  "#t", "#o"  \n\t"\

331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350
// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1
// out: mm5=beta-1, mm7=mask
// clobbers: mm4,mm6
#define H264_DEBLOCK_MASK(alpha1, beta1) \
    "pshufw $0, "#alpha1", %%mm4 \n\t"\
    "pshufw $0, "#beta1 ", %%mm5 \n\t"\
    "packuswb  %%mm4, %%mm4      \n\t"\
    "packuswb  %%mm5, %%mm5      \n\t"\
    DIFF_GT_MMX(%%mm1, %%mm2, %%mm4, %%mm7, %%mm6) /* |p0-q0| > alpha-1 */\
    DIFF_GT_MMX(%%mm0, %%mm1, %%mm5, %%mm4, %%mm6) /* |p1-p0| > beta-1 */\
    "por       %%mm4, %%mm7      \n\t"\
    DIFF_GT_MMX(%%mm3, %%mm2, %%mm5, %%mm4, %%mm6) /* |q1-q0| > beta-1 */\
    "por       %%mm4, %%mm7      \n\t"\
    "pxor      %%mm6, %%mm6      \n\t"\
    "pcmpeqb   %%mm6, %%mm7      \n\t"

// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask)
// out: mm1=p0' mm2=q0'
// clobbers: mm0,3-6
#define H264_DEBLOCK_P0_Q0(pb_01, pb_3f)\
michael's avatar
michael committed
351 352 353 354 355 356 357 358 359 360
        "movq    %%mm1              , %%mm5 \n\t"\
        "pxor    %%mm2              , %%mm5 \n\t" /* p0^q0*/\
        "pand    "#pb_01"           , %%mm5 \n\t" /* (p0^q0)&1*/\
        "pcmpeqb %%mm4              , %%mm4 \n\t"\
        "pxor    %%mm4              , %%mm3 \n\t"\
        "pavgb   %%mm0              , %%mm3 \n\t" /* (p1 - q1 + 256)>>1*/\
        "pavgb   "MANGLE(ff_pb_3)"  , %%mm3 \n\t" /*(((p1 - q1 + 256)>>1)+4)>>1 = 64+2+(p1-q1)>>2*/\
        "pxor    %%mm1              , %%mm4 \n\t"\
        "pavgb   %%mm2              , %%mm4 \n\t" /* (q0 - p0 + 256)>>1*/\
        "pavgb   %%mm5              , %%mm3 \n\t"\
michael's avatar
michael committed
361
        "paddusb %%mm4              , %%mm3 \n\t" /* d+128+33*/\
michael's avatar
michael committed
362 363
        "movq    "MANGLE(ff_pb_A1)" , %%mm6 \n\t"\
        "psubusb %%mm3              , %%mm6 \n\t"\
michael's avatar
michael committed
364 365
        "psubusb "MANGLE(ff_pb_A1)" , %%mm3 \n\t"\
        "pminub  %%mm7              , %%mm6 \n\t"\
michael's avatar
michael committed
366
        "pminub  %%mm7              , %%mm3 \n\t"\
michael's avatar
michael committed
367
        "psubusb %%mm6              , %%mm1 \n\t"\
michael's avatar
michael committed
368 369 370
        "psubusb %%mm3              , %%mm2 \n\t"\
        "paddusb %%mm3              , %%mm1 \n\t"\
        "paddusb %%mm6              , %%mm2 \n\t"
371 372

// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask) %8=mm_bone
373
// out: (q1addr) = av_clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )
374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390
// clobbers: q2, tmp, tc0
#define H264_DEBLOCK_Q1(p1, q2, q2addr, q1addr, tc0, tmp)\
        "movq     %%mm1,  "#tmp"   \n\t"\
        "pavgb    %%mm2,  "#tmp"   \n\t"\
        "pavgb    "#tmp", "#q2"    \n\t" /* avg(p2,avg(p0,q0)) */\
        "pxor   "q2addr", "#tmp"   \n\t"\
        "pand     %8,     "#tmp"   \n\t" /* (p2^avg(p0,q0))&1 */\
        "psubusb  "#tmp", "#q2"    \n\t" /* (p2+((p0+q0+1)>>1))>>1 */\
        "movq     "#p1",  "#tmp"   \n\t"\
        "psubusb  "#tc0", "#tmp"   \n\t"\
        "paddusb  "#p1",  "#tc0"   \n\t"\
        "pmaxub   "#tmp", "#q2"    \n\t"\
        "pminub   "#tc0", "#q2"    \n\t"\
        "movq     "#q2",  "q1addr" \n\t"

static inline void h264_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
{
michael's avatar
michael committed
391
    DECLARE_ALIGNED_8(uint64_t, tmp0[2]);
392 393 394 395 396 397 398

    asm volatile(
        "movq    (%1,%3), %%mm0    \n\t" //p1
        "movq    (%1,%3,2), %%mm1  \n\t" //p0
        "movq    (%2),    %%mm2    \n\t" //q0
        "movq    (%2,%3), %%mm3    \n\t" //q1
        H264_DEBLOCK_MASK(%6, %7)
399 400 401 402 403 404 405 406 407 408

        "movd      %5,    %%mm4    \n\t"
        "punpcklbw %%mm4, %%mm4    \n\t"
        "punpcklwd %%mm4, %%mm4    \n\t"
        "pcmpeqb   %%mm3, %%mm3    \n\t"
        "movq      %%mm4, %%mm6    \n\t"
        "pcmpgtb   %%mm3, %%mm4    \n\t"
        "movq      %%mm6, 8+%0     \n\t"
        "pand      %%mm4, %%mm7    \n\t"
        "movq      %%mm7, %0       \n\t"
409 410 411

        /* filter p1 */
        "movq     (%1),   %%mm3    \n\t" //p2
michael's avatar
michael committed
412
        DIFF_GT2_MMX(%%mm1, %%mm3, %%mm5, %%mm6, %%mm4) // |p2-p0|>beta-1
413
        "pand     %%mm7,  %%mm6    \n\t" // mask & |p2-p0|<beta
michael's avatar
michael committed
414 415
        "pand     8+%0,   %%mm7    \n\t" // mask & tc0
        "movq     %%mm7,  %%mm4    \n\t"
416
        "psubb    %%mm6,  %%mm7    \n\t"
417 418 419 420 421
        "pand     %%mm4,  %%mm6    \n\t" // mask & |p2-p0|<beta & tc0
        H264_DEBLOCK_Q1(%%mm0, %%mm3, "(%1)", "(%1,%3)", %%mm6, %%mm4)

        /* filter q1 */
        "movq    (%2,%3,2), %%mm4  \n\t" //q2
michael's avatar
michael committed
422
        DIFF_GT2_MMX(%%mm2, %%mm4, %%mm5, %%mm6, %%mm3) // |q2-q0|>beta-1
423
        "pand     %0,     %%mm6    \n\t"
michael's avatar
michael committed
424
        "movq     8+%0,   %%mm5    \n\t" // can be merged with the and below but is slower then
425
        "pand     %%mm6,  %%mm5    \n\t"
426
        "psubb    %%mm6,  %%mm7    \n\t"
427 428 429 430
        "movq    (%2,%3), %%mm3    \n\t"
        H264_DEBLOCK_Q1(%%mm3, %%mm4, "(%2,%3,2)", "(%2,%3)", %%mm5, %%mm6)

        /* filter p0, q0 */
431
        H264_DEBLOCK_P0_Q0(%8, unused)
432 433 434
        "movq      %%mm1, (%1,%3,2) \n\t"
        "movq      %%mm2, (%2)      \n\t"

435
        : "=m"(*tmp0)
436
        : "r"(pix-3*stride), "r"(pix), "r"((long)stride),
437 438
          "m"(*tmp0/*unused*/), "m"(*(uint32_t*)tc0), "m"(alpha1), "m"(beta1),
          "m"(mm_bone)
439 440 441 442 443 444 445 446 447 448 449 450 451 452
    );
}

static void h264_v_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
{
    if((tc0[0] & tc0[1]) >= 0)
        h264_loop_filter_luma_mmx2(pix, stride, alpha-1, beta-1, tc0);
    if((tc0[2] & tc0[3]) >= 0)
        h264_loop_filter_luma_mmx2(pix+8, stride, alpha-1, beta-1, tc0+2);
}
static void h264_h_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
{
    //FIXME: could cut some load/stores by merging transpose with filter
    // also, it only needs to transpose 6x8
michael's avatar
michael committed
453
    DECLARE_ALIGNED_8(uint8_t, trans[8*8]);
454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496
    int i;
    for(i=0; i<2; i++, pix+=8*stride, tc0+=2) {
        if((tc0[0] & tc0[1]) < 0)
            continue;
        transpose4x4(trans,       pix-4,          8, stride);
        transpose4x4(trans  +4*8, pix,            8, stride);
        transpose4x4(trans+4,     pix-4+4*stride, 8, stride);
        transpose4x4(trans+4+4*8, pix  +4*stride, 8, stride);
        h264_loop_filter_luma_mmx2(trans+4*8, 8, alpha-1, beta-1, tc0);
        transpose4x4(pix-2,          trans  +2*8, stride, 8);
        transpose4x4(pix-2+4*stride, trans+4+2*8, stride, 8);
    }
}

static inline void h264_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
{
    asm volatile(
        "movq    (%0),    %%mm0     \n\t" //p1
        "movq    (%0,%2), %%mm1     \n\t" //p0
        "movq    (%1),    %%mm2     \n\t" //q0
        "movq    (%1,%2), %%mm3     \n\t" //q1
        H264_DEBLOCK_MASK(%4, %5)
        "movd      %3,    %%mm6     \n\t"
        "punpcklbw %%mm6, %%mm6     \n\t"
        "pand      %%mm6, %%mm7     \n\t" // mm7 = tc&mask
        H264_DEBLOCK_P0_Q0(%6, %7)
        "movq      %%mm1, (%0,%2)   \n\t"
        "movq      %%mm2, (%1)      \n\t"

        :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
           "r"(*(uint32_t*)tc0),
           "m"(alpha1), "m"(beta1), "m"(mm_bone), "m"(ff_pb_3F)
    );
}

static void h264_v_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
{
    h264_loop_filter_chroma_mmx2(pix, stride, alpha-1, beta-1, tc0);
}

static void h264_h_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
{
    //FIXME: could cut some load/stores by merging transpose with filter
michael's avatar
michael committed
497
    DECLARE_ALIGNED_8(uint8_t, trans[8*4]);
498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546
    transpose4x4(trans, pix-2, 8, stride);
    transpose4x4(trans+4, pix-2+4*stride, 8, stride);
    h264_loop_filter_chroma_mmx2(trans+2*8, 8, alpha-1, beta-1, tc0);
    transpose4x4(pix-2, trans, stride, 8);
    transpose4x4(pix-2+4*stride, trans+4, stride, 8);
}

// p0 = (p0 + q1 + 2*p1 + 2) >> 2
#define H264_FILTER_CHROMA4(p0, p1, q1, one) \
    "movq    "#p0", %%mm4  \n\t"\
    "pxor    "#q1", %%mm4  \n\t"\
    "pand   "#one", %%mm4  \n\t" /* mm4 = (p0^q1)&1 */\
    "pavgb   "#q1", "#p0"  \n\t"\
    "psubusb %%mm4, "#p0"  \n\t"\
    "pavgb   "#p1", "#p0"  \n\t" /* dst = avg(p1, avg(p0,q1) - ((p0^q1)&1)) */\

static inline void h264_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha1, int beta1)
{
    asm volatile(
        "movq    (%0),    %%mm0     \n\t"
        "movq    (%0,%2), %%mm1     \n\t"
        "movq    (%1),    %%mm2     \n\t"
        "movq    (%1,%2), %%mm3     \n\t"
        H264_DEBLOCK_MASK(%3, %4)
        "movq    %%mm1,   %%mm5     \n\t"
        "movq    %%mm2,   %%mm6     \n\t"
        H264_FILTER_CHROMA4(%%mm1, %%mm0, %%mm3, %5) //p0'
        H264_FILTER_CHROMA4(%%mm2, %%mm3, %%mm0, %5) //q0'
        "psubb   %%mm5,   %%mm1     \n\t"
        "psubb   %%mm6,   %%mm2     \n\t"
        "pand    %%mm7,   %%mm1     \n\t"
        "pand    %%mm7,   %%mm2     \n\t"
        "paddb   %%mm5,   %%mm1     \n\t"
        "paddb   %%mm6,   %%mm2     \n\t"
        "movq    %%mm1,   (%0,%2)   \n\t"
        "movq    %%mm2,   (%1)      \n\t"
        :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
           "m"(alpha1), "m"(beta1), "m"(mm_bone)
    );
}

static void h264_v_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
{
    h264_loop_filter_chroma_intra_mmx2(pix, stride, alpha-1, beta-1);
}

static void h264_h_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
{
    //FIXME: could cut some load/stores by merging transpose with filter
michael's avatar
michael committed
547
    DECLARE_ALIGNED_8(uint8_t, trans[8*4]);
548 549 550 551 552 553 554
    transpose4x4(trans, pix-2, 8, stride);
    transpose4x4(trans+4, pix-2+4*stride, 8, stride);
    h264_loop_filter_chroma_intra_mmx2(trans+2*8, 8, alpha-1, beta-1);
    transpose4x4(pix-2, trans, stride, 8);
    transpose4x4(pix-2+4*stride, trans+4, stride, 8);
}

555 556 557 558 559 560 561 562 563 564 565 566 567 568 569
static void h264_loop_filter_strength_mmx2( int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],
                                            int bidir, int edges, int step, int mask_mv0, int mask_mv1 ) {
    int dir;
    asm volatile(
        "pxor %%mm7, %%mm7 \n\t"
        "movq %0, %%mm6 \n\t"
        "movq %1, %%mm5 \n\t"
        "movq %2, %%mm4 \n\t"
        ::"m"(ff_pb_1), "m"(ff_pb_3), "m"(ff_pb_7)
    );
    // could do a special case for dir==0 && edges==1, but it only reduces the
    // average filter time by 1.2%
    for( dir=1; dir>=0; dir-- ) {
        const int d_idx = dir ? -8 : -1;
        const int mask_mv = dir ? mask_mv1 : mask_mv0;
michael's avatar
michael committed
570
        DECLARE_ALIGNED_8(const uint64_t, mask_dir) = dir ? 0 : 0xffffffffffffffffULL;
571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649
        int b_idx, edge, l;
        for( b_idx=12, edge=0; edge<edges; edge+=step, b_idx+=8*step ) {
            asm volatile(
                "pand %0, %%mm0 \n\t"
                ::"m"(mask_dir)
            );
            if(!(mask_mv & edge)) {
                asm volatile("pxor %%mm0, %%mm0 \n\t":);
                for( l = bidir; l >= 0; l-- ) {
                    asm volatile(
                        "movd %0, %%mm1 \n\t"
                        "punpckldq %1, %%mm1 \n\t"
                        "movq %%mm1, %%mm2 \n\t"
                        "psrlw $7, %%mm2 \n\t"
                        "pand %%mm6, %%mm2 \n\t"
                        "por %%mm2, %%mm1 \n\t" // ref_cache with -2 mapped to -1
                        "punpckldq %%mm1, %%mm2 \n\t"
                        "pcmpeqb %%mm2, %%mm1 \n\t"
                        "paddb %%mm6, %%mm1 \n\t"
                        "punpckhbw %%mm7, %%mm1 \n\t" // ref[b] != ref[bn]
                        "por %%mm1, %%mm0 \n\t"

                        "movq %2, %%mm1 \n\t"
                        "movq %3, %%mm2 \n\t"
                        "psubw %4, %%mm1 \n\t"
                        "psubw %5, %%mm2 \n\t"
                        "packsswb %%mm2, %%mm1 \n\t"
                        "paddb %%mm5, %%mm1 \n\t"
                        "pminub %%mm4, %%mm1 \n\t"
                        "pcmpeqb %%mm4, %%mm1 \n\t" // abs(mv[b] - mv[bn]) >= limit
                        "por %%mm1, %%mm0 \n\t"
                        ::"m"(ref[l][b_idx]),
                          "m"(ref[l][b_idx+d_idx]),
                          "m"(mv[l][b_idx][0]),
                          "m"(mv[l][b_idx+2][0]),
                          "m"(mv[l][b_idx+d_idx][0]),
                          "m"(mv[l][b_idx+d_idx+2][0])
                    );
                }
            }
            asm volatile(
                "movd %0, %%mm1 \n\t"
                "por  %1, %%mm1 \n\t"
                "punpcklbw %%mm7, %%mm1 \n\t"
                "pcmpgtw %%mm7, %%mm1 \n\t" // nnz[b] || nnz[bn]
                ::"m"(nnz[b_idx]),
                  "m"(nnz[b_idx+d_idx])
            );
            asm volatile(
                "pcmpeqw %%mm7, %%mm0 \n\t"
                "pcmpeqw %%mm7, %%mm0 \n\t"
                "psrlw $15, %%mm0 \n\t" // nonzero -> 1
                "psrlw $14, %%mm1 \n\t"
                "movq %%mm0, %%mm2 \n\t"
                "por %%mm1, %%mm2 \n\t"
                "psrlw $1, %%mm1 \n\t"
                "pandn %%mm2, %%mm1 \n\t"
                "movq %%mm1, %0 \n\t"
                :"=m"(*bS[dir][edge])
                ::"memory"
            );
        }
        edges = 4;
        step = 1;
    }
    asm volatile(
        "movq   (%0), %%mm0 \n\t"
        "movq  8(%0), %%mm1 \n\t"
        "movq 16(%0), %%mm2 \n\t"
        "movq 24(%0), %%mm3 \n\t"
        TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4)
        "movq %%mm0,   (%0) \n\t"
        "movq %%mm3,  8(%0) \n\t"
        "movq %%mm4, 16(%0) \n\t"
        "movq %%mm2, 24(%0) \n\t"
        ::"r"(bS[0])
        :"memory"
    );
}
650 651 652 653 654

/***********************************/
/* motion compensation */

#define QPEL_H264V(A,B,C,D,E,F,OP)\
655 656 657 658 659 660 661 662 663 664 665 666 667 668
        "movd (%0), "#F"            \n\t"\
        "movq "#C", %%mm6           \n\t"\
        "paddw "#D", %%mm6          \n\t"\
        "psllw $2, %%mm6            \n\t"\
        "psubw "#B", %%mm6          \n\t"\
        "psubw "#E", %%mm6          \n\t"\
        "pmullw %4, %%mm6           \n\t"\
        "add %2, %0                 \n\t"\
        "punpcklbw %%mm7, "#F"      \n\t"\
        "paddw %5, "#A"             \n\t"\
        "paddw "#F", "#A"           \n\t"\
        "paddw "#A", %%mm6          \n\t"\
        "psraw $5, %%mm6            \n\t"\
        "packuswb %%mm6, %%mm6      \n\t"\
669
        OP(%%mm6, (%1), A, d)\
670
        "add %3, %1                 \n\t"
671 672

#define QPEL_H264HV(A,B,C,D,E,F,OF)\
673 674 675 676 677 678 679 680 681 682 683 684
        "movd (%0), "#F"            \n\t"\
        "movq "#C", %%mm6           \n\t"\
        "paddw "#D", %%mm6          \n\t"\
        "psllw $2, %%mm6            \n\t"\
        "psubw "#B", %%mm6          \n\t"\
        "psubw "#E", %%mm6          \n\t"\
        "pmullw %3, %%mm6           \n\t"\
        "add %2, %0                 \n\t"\
        "punpcklbw %%mm7, "#F"      \n\t"\
        "paddw "#F", "#A"           \n\t"\
        "paddw "#A", %%mm6          \n\t"\
        "movq %%mm6, "#OF"(%1)      \n\t"
685

686
#define QPEL_H264(OPNAME, OP, MMX)\
687
static av_noinline void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
688 689 690
    int h=4;\
\
    asm volatile(\
691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716
        "pxor %%mm7, %%mm7          \n\t"\
        "movq %5, %%mm4             \n\t"\
        "movq %6, %%mm5             \n\t"\
        "1:                         \n\t"\
        "movd  -1(%0), %%mm1        \n\t"\
        "movd    (%0), %%mm2        \n\t"\
        "movd   1(%0), %%mm3        \n\t"\
        "movd   2(%0), %%mm0        \n\t"\
        "punpcklbw %%mm7, %%mm1     \n\t"\
        "punpcklbw %%mm7, %%mm2     \n\t"\
        "punpcklbw %%mm7, %%mm3     \n\t"\
        "punpcklbw %%mm7, %%mm0     \n\t"\
        "paddw %%mm0, %%mm1         \n\t"\
        "paddw %%mm3, %%mm2         \n\t"\
        "movd  -2(%0), %%mm0        \n\t"\
        "movd   3(%0), %%mm3        \n\t"\
        "punpcklbw %%mm7, %%mm0     \n\t"\
        "punpcklbw %%mm7, %%mm3     \n\t"\
        "paddw %%mm3, %%mm0         \n\t"\
        "psllw $2, %%mm2            \n\t"\
        "psubw %%mm1, %%mm2         \n\t"\
        "pmullw %%mm4, %%mm2        \n\t"\
        "paddw %%mm5, %%mm0         \n\t"\
        "paddw %%mm2, %%mm0         \n\t"\
        "psraw $5, %%mm0            \n\t"\
        "packuswb %%mm0, %%mm0      \n\t"\
717
        OP(%%mm0, (%1),%%mm6, d)\
718 719 720 721
        "add %3, %0                 \n\t"\
        "add %4, %1                 \n\t"\
        "decl %2                    \n\t"\
        " jnz 1b                    \n\t"\
722 723 724 725 726
        : "+a"(src), "+c"(dst), "+m"(h)\
        : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
        : "memory"\
    );\
}\
727
static av_noinline void OPNAME ## h264_qpel4_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
lorenm's avatar
lorenm committed
728 729 730
    int h=4;\
    asm volatile(\
        "pxor %%mm7, %%mm7          \n\t"\
lorenm's avatar
lorenm committed
731 732 733 734 735 736
        "movq %0, %%mm4             \n\t"\
        "movq %1, %%mm5             \n\t"\
        :: "m"(ff_pw_5), "m"(ff_pw_16)\
    );\
    do{\
    asm volatile(\
lorenm's avatar
lorenm committed
737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761
        "movd  -1(%0), %%mm1        \n\t"\
        "movd    (%0), %%mm2        \n\t"\
        "movd   1(%0), %%mm3        \n\t"\
        "movd   2(%0), %%mm0        \n\t"\
        "punpcklbw %%mm7, %%mm1     \n\t"\
        "punpcklbw %%mm7, %%mm2     \n\t"\
        "punpcklbw %%mm7, %%mm3     \n\t"\
        "punpcklbw %%mm7, %%mm0     \n\t"\
        "paddw %%mm0, %%mm1         \n\t"\
        "paddw %%mm3, %%mm2         \n\t"\
        "movd  -2(%0), %%mm0        \n\t"\
        "movd   3(%0), %%mm3        \n\t"\
        "punpcklbw %%mm7, %%mm0     \n\t"\
        "punpcklbw %%mm7, %%mm3     \n\t"\
        "paddw %%mm3, %%mm0         \n\t"\
        "psllw $2, %%mm2            \n\t"\
        "psubw %%mm1, %%mm2         \n\t"\
        "pmullw %%mm4, %%mm2        \n\t"\
        "paddw %%mm5, %%mm0         \n\t"\
        "paddw %%mm2, %%mm0         \n\t"\
        "movd   (%2), %%mm3         \n\t"\
        "psraw $5, %%mm0            \n\t"\
        "packuswb %%mm0, %%mm0      \n\t"\
        PAVGB" %%mm3, %%mm0         \n\t"\
        OP(%%mm0, (%1),%%mm6, d)\
lorenm's avatar
lorenm committed
762 763 764 765 766
        "add %4, %0                 \n\t"\
        "add %4, %1                 \n\t"\
        "add %3, %2                 \n\t"\
        : "+a"(src), "+c"(dst), "+d"(src2)\
        : "D"((long)src2Stride), "S"((long)dstStride)\
lorenm's avatar
lorenm committed
767 768
        : "memory"\
    );\
lorenm's avatar
lorenm committed
769
    }while(--h);\
lorenm's avatar
lorenm committed
770
}\
771
static av_noinline void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
772 773
    src -= 2*srcStride;\
    asm volatile(\
774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789
        "pxor %%mm7, %%mm7          \n\t"\
        "movd (%0), %%mm0           \n\t"\
        "add %2, %0                 \n\t"\
        "movd (%0), %%mm1           \n\t"\
        "add %2, %0                 \n\t"\
        "movd (%0), %%mm2           \n\t"\
        "add %2, %0                 \n\t"\
        "movd (%0), %%mm3           \n\t"\
        "add %2, %0                 \n\t"\
        "movd (%0), %%mm4           \n\t"\
        "add %2, %0                 \n\t"\
        "punpcklbw %%mm7, %%mm0     \n\t"\
        "punpcklbw %%mm7, %%mm1     \n\t"\
        "punpcklbw %%mm7, %%mm2     \n\t"\
        "punpcklbw %%mm7, %%mm3     \n\t"\
        "punpcklbw %%mm7, %%mm4     \n\t"\
790 791 792 793 794 795 796 797 798 799
        QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
        QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
        QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
        QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
         \
        : "+a"(src), "+c"(dst)\
        : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
        : "memory"\
    );\
}\
800
static av_noinline void OPNAME ## h264_qpel4_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
801 802 803 804 805
    int h=4;\
    int w=3;\
    src -= 2*srcStride+2;\
    while(w--){\
        asm volatile(\
806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821
            "pxor %%mm7, %%mm7      \n\t"\
            "movd (%0), %%mm0       \n\t"\
            "add %2, %0             \n\t"\
            "movd (%0), %%mm1       \n\t"\
            "add %2, %0             \n\t"\
            "movd (%0), %%mm2       \n\t"\
            "add %2, %0             \n\t"\
            "movd (%0), %%mm3       \n\t"\
            "add %2, %0             \n\t"\
            "movd (%0), %%mm4       \n\t"\
            "add %2, %0             \n\t"\
            "punpcklbw %%mm7, %%mm0 \n\t"\
            "punpcklbw %%mm7, %%mm1 \n\t"\
            "punpcklbw %%mm7, %%mm2 \n\t"\
            "punpcklbw %%mm7, %%mm3 \n\t"\
            "punpcklbw %%mm7, %%mm4 \n\t"\
822 823 824 825 826 827 828 829 830 831 832 833 834 835
            QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*3)\
            QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*3)\
            QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*3)\
            QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*3)\
             \
            : "+a"(src)\
            : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\
            : "memory"\
        );\
        tmp += 4;\
        src += 4 - 9*srcStride;\
    }\
    tmp -= 3*4;\
    asm volatile(\
836 837 838 839 840 841 842 843 844 845 846 847
        "movq %4, %%mm6             \n\t"\
        "1:                         \n\t"\
        "movq     (%0), %%mm0       \n\t"\
        "paddw  10(%0), %%mm0       \n\t"\
        "movq    2(%0), %%mm1       \n\t"\
        "paddw   8(%0), %%mm1       \n\t"\
        "movq    4(%0), %%mm2       \n\t"\
        "paddw   6(%0), %%mm2       \n\t"\
        "psubw %%mm1, %%mm0         \n\t"/*a-b   (abccba)*/\
        "psraw $2, %%mm0            \n\t"/*(a-b)/4 */\
        "psubw %%mm1, %%mm0         \n\t"/*(a-b)/4-b */\
        "paddsw %%mm2, %%mm0        \n\t"\
848
        "psraw $2, %%mm0            \n\t"/*((a-b)/4-b+c)/4 */\
849
        "paddw %%mm6, %%mm2         \n\t"\
850
        "paddw %%mm2, %%mm0         \n\t"/*(a-5*b+20*c)/16 +32 */\
851 852
        "psraw $6, %%mm0            \n\t"\
        "packuswb %%mm0, %%mm0      \n\t"\
853
        OP(%%mm0, (%1),%%mm7, d)\
854 855 856 857
        "add $24, %0                \n\t"\
        "add %3, %1                 \n\t"\
        "decl %2                    \n\t"\
        " jnz 1b                    \n\t"\
858 859 860 861 862 863
        : "+a"(tmp), "+c"(dst), "+m"(h)\
        : "S"((long)dstStride), "m"(ff_pw_32)\
        : "memory"\
    );\
}\
\
864
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
865 866
    int h=8;\
    asm volatile(\
867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909
        "pxor %%mm7, %%mm7          \n\t"\
        "movq %5, %%mm6             \n\t"\
        "1:                         \n\t"\
        "movq    (%0), %%mm0        \n\t"\
        "movq   1(%0), %%mm2        \n\t"\
        "movq %%mm0, %%mm1          \n\t"\
        "movq %%mm2, %%mm3          \n\t"\
        "punpcklbw %%mm7, %%mm0     \n\t"\
        "punpckhbw %%mm7, %%mm1     \n\t"\
        "punpcklbw %%mm7, %%mm2     \n\t"\
        "punpckhbw %%mm7, %%mm3     \n\t"\
        "paddw %%mm2, %%mm0         \n\t"\
        "paddw %%mm3, %%mm1         \n\t"\
        "psllw $2, %%mm0            \n\t"\
        "psllw $2, %%mm1            \n\t"\
        "movq   -1(%0), %%mm2       \n\t"\
        "movq    2(%0), %%mm4       \n\t"\
        "movq %%mm2, %%mm3          \n\t"\
        "movq %%mm4, %%mm5          \n\t"\
        "punpcklbw %%mm7, %%mm2     \n\t"\
        "punpckhbw %%mm7, %%mm3     \n\t"\
        "punpcklbw %%mm7, %%mm4     \n\t"\
        "punpckhbw %%mm7, %%mm5     \n\t"\
        "paddw %%mm4, %%mm2         \n\t"\
        "paddw %%mm3, %%mm5         \n\t"\
        "psubw %%mm2, %%mm0         \n\t"\
        "psubw %%mm5, %%mm1         \n\t"\
        "pmullw %%mm6, %%mm0        \n\t"\
        "pmullw %%mm6, %%mm1        \n\t"\
        "movd   -2(%0), %%mm2       \n\t"\
        "movd    7(%0), %%mm5       \n\t"\
        "punpcklbw %%mm7, %%mm2     \n\t"\
        "punpcklbw %%mm7, %%mm5     \n\t"\
        "paddw %%mm3, %%mm2         \n\t"\
        "paddw %%mm5, %%mm4         \n\t"\
        "movq %6, %%mm5             \n\t"\
        "paddw %%mm5, %%mm2         \n\t"\
        "paddw %%mm5, %%mm4         \n\t"\
        "paddw %%mm2, %%mm0         \n\t"\
        "paddw %%mm4, %%mm1         \n\t"\
        "psraw $5, %%mm0            \n\t"\
        "psraw $5, %%mm1            \n\t"\
        "packuswb %%mm1, %%mm0      \n\t"\
910
        OP(%%mm0, (%1),%%mm5, q)\
911 912 913 914
        "add %3, %0                 \n\t"\
        "add %4, %1                 \n\t"\
        "decl %2                    \n\t"\
        " jnz 1b                    \n\t"\
915 916 917 918 919 920
        : "+a"(src), "+c"(dst), "+m"(h)\
        : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
        : "memory"\
    );\
}\
\
921
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
lorenm's avatar
lorenm committed
922 923 924
    int h=8;\
    asm volatile(\
        "pxor %%mm7, %%mm7          \n\t"\
lorenm's avatar
lorenm committed
925 926 927 928 929
        "movq %0, %%mm6             \n\t"\
        :: "m"(ff_pw_5)\
    );\
    do{\
    asm volatile(\
lorenm's avatar
lorenm committed
930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961
        "movq    (%0), %%mm0        \n\t"\
        "movq   1(%0), %%mm2        \n\t"\
        "movq %%mm0, %%mm1          \n\t"\
        "movq %%mm2, %%mm3          \n\t"\
        "punpcklbw %%mm7, %%mm0     \n\t"\
        "punpckhbw %%mm7, %%mm1     \n\t"\
        "punpcklbw %%mm7, %%mm2     \n\t"\
        "punpckhbw %%mm7, %%mm3     \n\t"\
        "paddw %%mm2, %%mm0         \n\t"\
        "paddw %%mm3, %%mm1         \n\t"\
        "psllw $2, %%mm0            \n\t"\
        "psllw $2, %%mm1            \n\t"\
        "movq   -1(%0), %%mm2       \n\t"\
        "movq    2(%0), %%mm4       \n\t"\
        "movq %%mm2, %%mm3          \n\t"\
        "movq %%mm4, %%mm5          \n\t"\
        "punpcklbw %%mm7, %%mm2     \n\t"\
        "punpckhbw %%mm7, %%mm3     \n\t"\
        "punpcklbw %%mm7, %%mm4     \n\t"\
        "punpckhbw %%mm7, %%mm5     \n\t"\
        "paddw %%mm4, %%mm2         \n\t"\
        "paddw %%mm3, %%mm5         \n\t"\
        "psubw %%mm2, %%mm0         \n\t"\
        "psubw %%mm5, %%mm1         \n\t"\
        "pmullw %%mm6, %%mm0        \n\t"\
        "pmullw %%mm6, %%mm1        \n\t"\
        "movd   -2(%0), %%mm2       \n\t"\
        "movd    7(%0), %%mm5       \n\t"\
        "punpcklbw %%mm7, %%mm2     \n\t"\
        "punpcklbw %%mm7, %%mm5     \n\t"\
        "paddw %%mm3, %%mm2         \n\t"\
        "paddw %%mm5, %%mm4         \n\t"\
lorenm's avatar
lorenm committed
962
        "movq %5, %%mm5             \n\t"\
lorenm's avatar
lorenm committed
963 964 965 966 967 968 969 970 971 972
        "paddw %%mm5, %%mm2         \n\t"\
        "paddw %%mm5, %%mm4         \n\t"\
        "paddw %%mm2, %%mm0         \n\t"\
        "paddw %%mm4, %%mm1         \n\t"\
        "psraw $5, %%mm0            \n\t"\
        "psraw $5, %%mm1            \n\t"\
        "movq (%2), %%mm4           \n\t"\
        "packuswb %%mm1, %%mm0      \n\t"\
        PAVGB" %%mm4, %%mm0         \n\t"\
        OP(%%mm0, (%1),%%mm5, q)\
lorenm's avatar
lorenm committed
973 974 975 976
        "add %4, %0                 \n\t"\
        "add %4, %1                 \n\t"\
        "add %3, %2                 \n\t"\
        : "+a"(src), "+c"(dst), "+d"(src2)\
lorenm's avatar
lorenm committed
977
        : "D"((long)src2Stride), "S"((long)dstStride),\
lorenm's avatar
lorenm committed
978
          "m"(ff_pw_16)\
lorenm's avatar
lorenm committed
979 980
        : "memory"\
    );\
lorenm's avatar
lorenm committed
981
    }while(--h);\
lorenm's avatar
lorenm committed
982 983
}\
\
984
static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
985
    int w= 2;\
986 987
    src -= 2*srcStride;\
    \
988
    while(w--){\
989
      asm volatile(\
990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005
        "pxor %%mm7, %%mm7          \n\t"\
        "movd (%0), %%mm0           \n\t"\
        "add %2, %0                 \n\t"\
        "movd (%0), %%mm1           \n\t"\
        "add %2, %0                 \n\t"\
        "movd (%0), %%mm2           \n\t"\
        "add %2, %0                 \n\t"\
        "movd (%0), %%mm3           \n\t"\
        "add %2, %0                 \n\t"\
        "movd (%0), %%mm4           \n\t"\
        "add %2, %0                 \n\t"\
        "punpcklbw %%mm7, %%mm0     \n\t"\
        "punpcklbw %%mm7, %%mm1     \n\t"\
        "punpcklbw %%mm7, %%mm2     \n\t"\
        "punpcklbw %%mm7, %%mm3     \n\t"\
        "punpcklbw %%mm7, %%mm4     \n\t"\
1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018
        QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
        QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
        QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
        QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
        QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
        QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
        QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
        QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
         \
        : "+a"(src), "+c"(dst)\
        : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
        : "memory"\
     );\
1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036
     if(h==16){\
        asm volatile(\
            QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
            QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
            QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
            QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
            QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
            QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
            QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
            QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
            \
           : "+a"(src), "+c"(dst)\
           : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
           : "memory"\
        );\
     }\
     src += 4-(h+5)*srcStride;\
     dst += 4-h*dstStride;\
1037 1038
   }\
}\
1039
static av_noinline void OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\
1040 1041
    int h = size;\
    int w = (size+8)>>2;\
1042 1043 1044
    src -= 2*srcStride+2;\
    while(w--){\
        asm volatile(\
1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060
            "pxor %%mm7, %%mm7      \n\t"\
            "movd (%0), %%mm0       \n\t"\
            "add %2, %0             \n\t"\
            "movd (%0), %%mm1       \n\t"\
            "add %2, %0             \n\t"\
            "movd (%0), %%mm2       \n\t"\
            "add %2, %0             \n\t"\
            "movd (%0), %%mm3       \n\t"\
            "add %2, %0             \n\t"\
            "movd (%0), %%mm4       \n\t"\
            "add %2, %0             \n\t"\
            "punpcklbw %%mm7, %%mm0 \n\t"\
            "punpcklbw %%mm7, %%mm1 \n\t"\
            "punpcklbw %%mm7, %%mm2 \n\t"\
            "punpcklbw %%mm7, %%mm3 \n\t"\
            "punpcklbw %%mm7, %%mm4 \n\t"\
1061 1062 1063 1064 1065 1066 1067 1068
            QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*48)\
            QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*48)\
            QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*48)\
            QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*48)\
            QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*48)\
            QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*48)\
            QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*48)\
            QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*48)\
1069 1070 1071 1072
            : "+a"(src)\
            : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\
            : "memory"\
        );\
1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087
        if(size==16){\
            asm volatile(\
                QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1,  8*48)\
                QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2,  9*48)\
                QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 10*48)\
                QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 11*48)\
                QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 12*48)\
                QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 13*48)\
                QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 14*48)\
                QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 15*48)\
                : "+a"(src)\
                : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\
                : "memory"\
            );\
        }\
1088
        tmp += 4;\
1089
        src += 4 - (size+5)*srcStride;\
1090
    }\
1091 1092 1093 1094
    tmp -= size+8;\
    w = size>>4;\
    do{\
    h = size;\
1095
    asm volatile(\
1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126
        "movq %4, %%mm6             \n\t"\
        "1:                         \n\t"\
        "movq     (%0), %%mm0       \n\t"\
        "movq    8(%0), %%mm3       \n\t"\
        "movq    2(%0), %%mm1       \n\t"\
        "movq   10(%0), %%mm4       \n\t"\
        "paddw   %%mm4, %%mm0       \n\t"\
        "paddw   %%mm3, %%mm1       \n\t"\
        "paddw  18(%0), %%mm3       \n\t"\
        "paddw  16(%0), %%mm4       \n\t"\
        "movq    4(%0), %%mm2       \n\t"\
        "movq   12(%0), %%mm5       \n\t"\
        "paddw   6(%0), %%mm2       \n\t"\
        "paddw  14(%0), %%mm5       \n\t"\
        "psubw %%mm1, %%mm0         \n\t"\
        "psubw %%mm4, %%mm3         \n\t"\
        "psraw $2, %%mm0            \n\t"\
        "psraw $2, %%mm3            \n\t"\
        "psubw %%mm1, %%mm0         \n\t"\
        "psubw %%mm4, %%mm3         \n\t"\
        "paddsw %%mm2, %%mm0        \n\t"\
        "paddsw %%mm5, %%mm3        \n\t"\
        "psraw $2, %%mm0            \n\t"\
        "psraw $2, %%mm3            \n\t"\
        "paddw %%mm6, %%mm2         \n\t"\
        "paddw %%mm6, %%mm5         \n\t"\
        "paddw %%mm2, %%mm0         \n\t"\
        "paddw %%mm5, %%mm3         \n\t"\
        "psraw $6, %%mm0            \n\t"\
        "psraw $6, %%mm3            \n\t"\
        "packuswb %%mm3, %%mm0      \n\t"\
1127
        OP(%%mm0, (%1),%%mm7, q)\
1128
        "add $48, %0                \n\t"\
1129 1130 1131
        "add %3, %1                 \n\t"\
        "decl %2                    \n\t"\
        " jnz 1b                    \n\t"\
1132 1133 1134 1135
        : "+a"(tmp), "+c"(dst), "+m"(h)\
        : "S"((long)dstStride), "m"(ff_pw_32)\
        : "memory"\
    );\
1136 1137 1138
    tmp += 8 - size*24;\
    dst += 8 - size*dstStride;\
    }while(w--);\
1139
}\
1140 1141 1142 1143
\
static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
}\
1144
static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1145 1146
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
1147 1148
}\
\
1149
static av_noinline void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1150 1151 1152 1153 1154 1155 1156 1157
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
    src += 8*srcStride;\
    dst += 8*dstStride;\
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
}\
\
1158
static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
lorenm's avatar
lorenm committed
1159 1160 1161 1162 1163 1164 1165 1166 1167
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
    src += 8*dstStride;\
    dst += 8*dstStride;\
    src2 += 8*src2Stride;\
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
}\
\
1168 1169 1170 1171
static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
    OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 8);\
}\
\
1172
static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1173
    OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 16);\
1174
}\
1175
\
1176
static av_noinline void OPNAME ## pixels4_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1177 1178 1179 1180 1181 1182 1183
{\
    asm volatile(\
        "movq       %5,  %%mm6          \n\t"\
        "movq      (%1), %%mm0          \n\t"\
        "movq    24(%1), %%mm1          \n\t"\
        "paddw    %%mm6, %%mm0          \n\t"\
        "paddw    %%mm6, %%mm1          \n\t"\
lorenm's avatar
lorenm committed
1184 1185
        "psraw      $5,  %%mm0          \n\t"\
        "psraw      $5,  %%mm1          \n\t"\
lorenm's avatar
lorenm committed
1186 1187
        "packuswb %%mm0, %%mm0          \n\t"\
        "packuswb %%mm1, %%mm1          \n\t"\
1188 1189 1190 1191 1192 1193 1194 1195 1196 1197
        PAVGB"     (%0), %%mm0          \n\t"\
        PAVGB"  (%0,%3), %%mm1          \n\t"\
        OP(%%mm0, (%2),    %%mm4, d)\
        OP(%%mm1, (%2,%4), %%mm5, d)\
        "lea  (%0,%3,2), %0             \n\t"\
        "lea  (%2,%4,2), %2             \n\t"\
        "movq    48(%1), %%mm0          \n\t"\
        "movq    72(%1), %%mm1          \n\t"\
        "paddw    %%mm6, %%mm0          \n\t"\
        "paddw    %%mm6, %%mm1          \n\t"\
lorenm's avatar
lorenm committed
1198 1199
        "psraw      $5,  %%mm0          \n\t"\
        "psraw      $5,  %%mm1          \n\t"\
lorenm's avatar
lorenm committed
1200 1201
        "packuswb %%mm0, %%mm0          \n\t"\
        "packuswb %%mm1, %%mm1          \n\t"\
1202 1203 1204 1205 1206
        PAVGB"     (%0), %%mm0          \n\t"\
        PAVGB"  (%0,%3), %%mm1          \n\t"\
        OP(%%mm0, (%2),    %%mm4, d)\
        OP(%%mm1, (%2,%4), %%mm5, d)\
        :"+a"(src8), "+c"(src16), "+d"(dst)\
lorenm's avatar
lorenm committed
1207
        :"S"((long)src8Stride), "D"((long)dstStride), "m"(ff_pw_16)\
1208 1209
        :"memory");\
}\
1210
static av_noinline void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221
{\
    asm volatile(\
        "movq       %0,  %%mm6          \n\t"\
        ::"m"(ff_pw_16)\
        );\
    while(h--){\
    asm volatile(\
        "movq      (%1), %%mm0          \n\t"\
        "movq     8(%1), %%mm1          \n\t"\
        "paddw    %%mm6, %%mm0          \n\t"\
        "paddw    %%mm6, %%mm1          \n\t"\
lorenm's avatar
lorenm committed
1222 1223
        "psraw      $5,  %%mm0          \n\t"\
        "psraw      $5,  %%mm1          \n\t"\
1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239
        "packuswb %%mm1, %%mm0          \n\t"\
        PAVGB"     (%0), %%mm0          \n\t"\
        OP(%%mm0, (%2), %%mm5, q)\
        ::"a"(src8), "c"(src16), "d"(dst)\
        :"memory");\
        src8 += src8Stride;\
        src16 += 24;\
        dst += dstStride;\
    }\
}\
static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
{\
    OPNAME ## pixels8_l2_shift5_ ## MMX(dst  , src16  , src8  , dstStride, src8Stride, h);\
    OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
}\

1240 1241 1242 1243 1244 1245 1246

#define H264_MC(OPNAME, SIZE, MMX) \
static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## pixels ## SIZE ## _mmx(dst, src, stride, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
lorenm's avatar
lorenm committed
1247
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
1248 1249 1250 1251 1252 1253 1254
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
lorenm's avatar
lorenm committed
1255
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[SIZE*SIZE/8];\
    uint8_t * const half= (uint8_t*)temp;\
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\
    OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[SIZE*SIZE/8];\
    uint8_t * const half= (uint8_t*)temp;\
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\
    OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, half, stride, stride, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
lorenm's avatar
lorenm committed
1277 1278
    uint64_t temp[SIZE*SIZE/8];\
    uint8_t * const halfV= (uint8_t*)temp;\
1279
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
lorenm's avatar
lorenm committed
1280
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfV, stride, SIZE);\
1281 1282 1283
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
lorenm's avatar
lorenm committed
1284 1285
    uint64_t temp[SIZE*SIZE/8];\
    uint8_t * const halfV= (uint8_t*)temp;\
1286
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
lorenm's avatar
lorenm committed
1287
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfV, stride, SIZE);\
1288 1289 1290
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
lorenm's avatar
lorenm committed
1291 1292
    uint64_t temp[SIZE*SIZE/8];\
    uint8_t * const halfV= (uint8_t*)temp;\
1293
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
lorenm's avatar
lorenm committed
1294
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfV, stride, SIZE);\
1295 1296 1297
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
lorenm's avatar
lorenm committed
1298 1299
    uint64_t temp[SIZE*SIZE/8];\
    uint8_t * const halfV= (uint8_t*)temp;\
1300
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
lorenm's avatar
lorenm committed
1301
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfV, stride, SIZE);\
1302 1303 1304
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1305
    uint64_t temp[SIZE*(SIZE<8?12:24)/4];\
1306 1307 1308 1309 1310
    int16_t * const tmp= (int16_t*)temp;\
    OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, tmp, src, stride, SIZE, stride);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
lorenm's avatar
lorenm committed
1311 1312 1313
    uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\
    uint8_t * const halfHV= (uint8_t*)temp;\
    int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE/2;\
1314
    assert((int)temp & 7 == 0);\
1315
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
lorenm's avatar
lorenm committed
1316
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
1317 1318 1319
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
lorenm's avatar
lorenm committed
1320 1321 1322
    uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\
    uint8_t * const halfHV= (uint8_t*)temp;\
    int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE/2;\
1323
    assert((int)temp & 7 == 0);\
1324
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
lorenm's avatar
lorenm committed
1325
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
1326 1327 1328
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1329 1330 1331
    uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\
    int16_t * const halfV= ((int16_t*)temp) + SIZE*SIZE/2;\
    uint8_t * const halfHV= ((uint8_t*)temp);\
1332
    assert((int)temp & 7 == 0);\
1333 1334
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
    OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
1335 1336 1337
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1338 1339 1340
    uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\
    int16_t * const halfV= ((int16_t*)temp) + SIZE*SIZE/2;\
    uint8_t * const halfHV= ((uint8_t*)temp);\
1341
    assert((int)temp & 7 == 0);\
1342 1343
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
    OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
1344 1345 1346 1347
}\


#define AVG_3DNOW_OP(a,b,temp, size) \
1348 1349 1350
"mov" #size " " #b ", " #temp "   \n\t"\
"pavgusb " #temp ", " #a "        \n\t"\
"mov" #size " " #a ", " #b "      \n\t"
1351
#define AVG_MMX2_OP(a,b,temp, size) \
1352 1353 1354
"mov" #size " " #b ", " #temp "   \n\t"\
"pavgb " #temp ", " #a "          \n\t"\
"mov" #size " " #a ", " #b "      \n\t"
1355

1356
#define PAVGB "pavgusb"
1357 1358
QPEL_H264(put_,       PUT_OP, 3dnow)
QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
1359 1360
#undef PAVGB
#define PAVGB "pavgb"
1361 1362
QPEL_H264(put_,       PUT_OP, mmx2)
QPEL_H264(avg_,  AVG_MMX2_OP, mmx2)
1363
#undef PAVGB
1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379

H264_MC(put_, 4, 3dnow)
H264_MC(put_, 8, 3dnow)
H264_MC(put_, 16,3dnow)
H264_MC(avg_, 4, 3dnow)
H264_MC(avg_, 8, 3dnow)
H264_MC(avg_, 16,3dnow)
H264_MC(put_, 4, mmx2)
H264_MC(put_, 8, mmx2)
H264_MC(put_, 16,mmx2)
H264_MC(avg_, 4, mmx2)
H264_MC(avg_, 8, mmx2)
H264_MC(avg_, 16,mmx2)


#define H264_CHROMA_OP(S,D)
1380
#define H264_CHROMA_OP4(S,D,T)
1381
#define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_mmx
1382
#define H264_CHROMA_MC4_TMPL put_h264_chroma_mc4_mmx
1383
#define H264_CHROMA_MC2_TMPL put_h264_chroma_mc2_mmx2
1384
#define H264_CHROMA_MC8_MV0 put_pixels8_mmx
1385 1386
#include "dsputil_h264_template_mmx.c"
#undef H264_CHROMA_OP
1387
#undef H264_CHROMA_OP4
1388
#undef H264_CHROMA_MC8_TMPL
1389
#undef H264_CHROMA_MC4_TMPL
1390
#undef H264_CHROMA_MC2_TMPL
1391
#undef H264_CHROMA_MC8_MV0
1392 1393

#define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t"
1394 1395
#define H264_CHROMA_OP4(S,D,T) "movd  " #S ", " #T " \n\t"\
                               "pavgb " #T ", " #D " \n\t"
1396
#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_mmx2
1397
#define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_mmx2
1398
#define H264_CHROMA_MC2_TMPL avg_h264_chroma_mc2_mmx2
1399
#define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2
1400 1401
#include "dsputil_h264_template_mmx.c"
#undef H264_CHROMA_OP
1402
#undef H264_CHROMA_OP4
1403
#undef H264_CHROMA_MC8_TMPL
1404
#undef H264_CHROMA_MC4_TMPL
1405
#undef H264_CHROMA_MC2_TMPL
1406
#undef H264_CHROMA_MC8_MV0
1407 1408

#define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"
1409 1410
#define H264_CHROMA_OP4(S,D,T) "movd " #S ", " #T " \n\t"\
                               "pavgusb " #T ", " #D " \n\t"
1411
#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_3dnow
1412 1413
#define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_3dnow
#define H264_CHROMA_MC8_MV0 avg_pixels8_3dnow
1414 1415
#include "dsputil_h264_template_mmx.c"
#undef H264_CHROMA_OP
1416
#undef H264_CHROMA_OP4
1417
#undef H264_CHROMA_MC8_TMPL
1418 1419
#undef H264_CHROMA_MC4_TMPL
#undef H264_CHROMA_MC8_MV0
1420

1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446
/***********************************/
/* weighted prediction */

static inline void ff_h264_weight_WxH_mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset, int w, int h)
{
    int x, y;
    offset <<= log2_denom;
    offset += (1 << log2_denom) >> 1;
    asm volatile(
        "movd    %0, %%mm4        \n\t"
        "movd    %1, %%mm5        \n\t"
        "movd    %2, %%mm6        \n\t"
        "pshufw  $0, %%mm4, %%mm4 \n\t"
        "pshufw  $0, %%mm5, %%mm5 \n\t"
        "pxor    %%mm7, %%mm7     \n\t"
        :: "g"(weight), "g"(offset), "g"(log2_denom)
    );
    for(y=0; y<h; y+=2){
        for(x=0; x<w; x+=4){
            asm volatile(
                "movd      %0,    %%mm0 \n\t"
                "movd      %1,    %%mm1 \n\t"
                "punpcklbw %%mm7, %%mm0 \n\t"
                "punpcklbw %%mm7, %%mm1 \n\t"
                "pmullw    %%mm4, %%mm0 \n\t"
                "pmullw    %%mm4, %%mm1 \n\t"
1447 1448
                "paddsw    %%mm5, %%mm0 \n\t"
                "paddsw    %%mm5, %%mm1 \n\t"
1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462
                "psraw     %%mm6, %%mm0 \n\t"
                "psraw     %%mm6, %%mm1 \n\t"
                "packuswb  %%mm7, %%mm0 \n\t"
                "packuswb  %%mm7, %%mm1 \n\t"
                "movd      %%mm0, %0    \n\t"
                "movd      %%mm1, %1    \n\t"
                : "+m"(*(uint32_t*)(dst+x)),
                  "+m"(*(uint32_t*)(dst+x+stride))
            );
        }
        dst += 2*stride;
    }
}

lorenm's avatar
lorenm committed
1463
static inline void ff_h264_biweight_WxH_mmx2(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset, int w, int h)
1464 1465
{
    int x, y;
lorenm's avatar
lorenm committed
1466
    offset = ((offset + 1) | 1) << log2_denom;
1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486
    asm volatile(
        "movd    %0, %%mm3        \n\t"
        "movd    %1, %%mm4        \n\t"
        "movd    %2, %%mm5        \n\t"
        "movd    %3, %%mm6        \n\t"
        "pshufw  $0, %%mm3, %%mm3 \n\t"
        "pshufw  $0, %%mm4, %%mm4 \n\t"
        "pshufw  $0, %%mm5, %%mm5 \n\t"
        "pxor    %%mm7, %%mm7     \n\t"
        :: "g"(weightd), "g"(weights), "g"(offset), "g"(log2_denom+1)
    );
    for(y=0; y<h; y++){
        for(x=0; x<w; x+=4){
            asm volatile(
                "movd      %0,    %%mm0 \n\t"
                "movd      %1,    %%mm1 \n\t"
                "punpcklbw %%mm7, %%mm0 \n\t"
                "punpcklbw %%mm7, %%mm1 \n\t"
                "pmullw    %%mm3, %%mm0 \n\t"
                "pmullw    %%mm4, %%mm1 \n\t"
1487 1488
                "paddsw    %%mm1, %%mm0 \n\t"
                "paddsw    %%mm5, %%mm0 \n\t"
1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501
                "psraw     %%mm6, %%mm0 \n\t"
                "packuswb  %%mm0, %%mm0 \n\t"
                "movd      %%mm0, %0    \n\t"
                : "+m"(*(uint32_t*)(dst+x))
                :  "m"(*(uint32_t*)(src+x))
            );
        }
        src += stride;
        dst += stride;
    }
}

#define H264_WEIGHT(W,H) \
lorenm's avatar
lorenm committed
1502 1503
static void ff_h264_biweight_ ## W ## x ## H ## _mmx2(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset){ \
    ff_h264_biweight_WxH_mmx2(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \
1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517
} \
static void ff_h264_weight_ ## W ## x ## H ## _mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset){ \
    ff_h264_weight_WxH_mmx2(dst, stride, log2_denom, weight, offset, W, H); \
}

H264_WEIGHT(16,16)
H264_WEIGHT(16, 8)
H264_WEIGHT( 8,16)
H264_WEIGHT( 8, 8)
H264_WEIGHT( 8, 4)
H264_WEIGHT( 4, 8)
H264_WEIGHT( 4, 4)
H264_WEIGHT( 4, 2)