mc.c 15.3 KB
Newer Older
Laurent Aimar's avatar
Laurent Aimar committed
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
/*****************************************************************************
 * mc.c: h264 encoder library (Motion Compensation)
 *****************************************************************************
 * Copyright (C) 2003 Laurent Aimar
 * $Id: mc.c,v 1.1 2004/06/03 19:27:07 fenrir Exp $
 *
 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111, USA.
 *****************************************************************************/

Loren Merritt's avatar
Loren Merritt committed
24
#include "common.h"
Laurent Aimar's avatar
Laurent Aimar committed
25

26
#ifdef HAVE_MMX
27
#include "x86/mc.h"
Laurent Aimar's avatar
Laurent Aimar committed
28
#endif
29
#ifdef ARCH_PPC
30
#include "ppc/mc.h"
Laurent Aimar's avatar
Laurent Aimar committed
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
#endif


static inline int x264_tapfilter( uint8_t *pix, int i_pix_next )
{
    return pix[-2*i_pix_next] - 5*pix[-1*i_pix_next] + 20*(pix[0] + pix[1*i_pix_next]) - 5*pix[ 2*i_pix_next] + pix[ 3*i_pix_next];
}
static inline int x264_tapfilter1( uint8_t *pix )
{
    return pix[-2] - 5*pix[-1] + 20*(pix[0] + pix[1]) - 5*pix[ 2] + pix[ 3];
}

static inline void pixel_avg( uint8_t *dst,  int i_dst_stride,
                              uint8_t *src1, int i_src1_stride,
                              uint8_t *src2, int i_src2_stride,
                              int i_width, int i_height )
{
    int x, y;
    for( y = 0; y < i_height; y++ )
    {
        for( x = 0; x < i_width; x++ )
        {
            dst[x] = ( src1[x] + src2[x] + 1 ) >> 1;
        }
        dst  += i_dst_stride;
        src1 += i_src1_stride;
        src2 += i_src2_stride;
    }
}

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 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142
static inline void pixel_avg_wxh( uint8_t *dst, int i_dst, uint8_t *src, int i_src, int width, int height )
{
    int x, y;
    for( y = 0; y < height; y++ )
    {
        for( x = 0; x < width; x++ )
        {
            dst[x] = ( dst[x] + src[x] + 1 ) >> 1;
        }
        dst += i_dst;
        src += i_src;
    }
}

#define PIXEL_AVG_C( name, width, height ) \
static void name( uint8_t *pix1, int i_stride_pix1, \
                  uint8_t *pix2, int i_stride_pix2 ) \
{ \
    pixel_avg_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height ); \
}
PIXEL_AVG_C( pixel_avg_16x16, 16, 16 )
PIXEL_AVG_C( pixel_avg_16x8,  16, 8 )
PIXEL_AVG_C( pixel_avg_8x16,  8, 16 )
PIXEL_AVG_C( pixel_avg_8x8,   8, 8 )
PIXEL_AVG_C( pixel_avg_8x4,   8, 4 )
PIXEL_AVG_C( pixel_avg_4x8,   4, 8 )
PIXEL_AVG_C( pixel_avg_4x4,   4, 4 )
PIXEL_AVG_C( pixel_avg_4x2,   4, 2 )
PIXEL_AVG_C( pixel_avg_2x4,   2, 4 )
PIXEL_AVG_C( pixel_avg_2x2,   2, 2 )


/* Implicit weighted bipred only:
 * assumes log2_denom = 5, offset = 0, weight1 + weight2 = 64 */
#define op_scale2(x) dst[x] = x264_clip_uint8( (dst[x]*i_weight1 + src[x]*i_weight2 + (1<<5)) >> 6 )
static inline void pixel_avg_weight_wxh( uint8_t *dst, int i_dst, uint8_t *src, int i_src, int width, int height, int i_weight1 ){
    int y;
    const int i_weight2 = 64 - i_weight1;
    for(y=0; y<height; y++, dst += i_dst, src += i_src){
        op_scale2(0);
        op_scale2(1);
        if(width==2) continue;
        op_scale2(2);
        op_scale2(3);
        if(width==4) continue;
        op_scale2(4);
        op_scale2(5);
        op_scale2(6);
        op_scale2(7);
        if(width==8) continue;
        op_scale2(8);
        op_scale2(9);
        op_scale2(10);
        op_scale2(11);
        op_scale2(12);
        op_scale2(13);
        op_scale2(14);
        op_scale2(15);
    }
}

#define PIXEL_AVG_WEIGHT_C( width, height ) \
static void pixel_avg_weight_##width##x##height( \
                uint8_t *pix1, int i_stride_pix1, \
                uint8_t *pix2, int i_stride_pix2, int i_weight1 ) \
{ \
    pixel_avg_weight_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height, i_weight1 ); \
}

PIXEL_AVG_WEIGHT_C(16,16)
PIXEL_AVG_WEIGHT_C(16,8)
PIXEL_AVG_WEIGHT_C(8,16)
PIXEL_AVG_WEIGHT_C(8,8)
PIXEL_AVG_WEIGHT_C(8,4)
PIXEL_AVG_WEIGHT_C(4,8)
PIXEL_AVG_WEIGHT_C(4,4)
PIXEL_AVG_WEIGHT_C(4,2)
PIXEL_AVG_WEIGHT_C(2,4)
PIXEL_AVG_WEIGHT_C(2,2)
#undef op_scale2
#undef PIXEL_AVG_WEIGHT_C

Laurent Aimar's avatar
Laurent Aimar committed
143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164
typedef void (*pf_mc_t)(uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_width, int i_height );

static void mc_copy( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_width, int i_height )
{
    int y;

    for( y = 0; y < i_height; y++ )
    {
        memcpy( dst, src, i_width );

        src += i_src_stride;
        dst += i_dst_stride;
    }
}
static inline void mc_hh( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_width, int i_height )
{
    int x, y;

    for( y = 0; y < i_height; y++ )
    {
        for( x = 0; x < i_width; x++ )
        {
Loren Merritt's avatar
Loren Merritt committed
165
            dst[x] = x264_clip_uint8( ( x264_tapfilter1( &src[x] ) + 16 ) >> 5 );
Laurent Aimar's avatar
Laurent Aimar committed
166 167 168 169 170 171 172 173 174 175 176 177 178
        }
        src += i_src_stride;
        dst += i_dst_stride;
    }
}
static inline void mc_hv( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_width, int i_height )
{
    int x, y;

    for( y = 0; y < i_height; y++ )
    {
        for( x = 0; x < i_width; x++ )
        {
Loren Merritt's avatar
Loren Merritt committed
179
            dst[x] = x264_clip_uint8( ( x264_tapfilter( &src[x], i_src_stride ) + 16 ) >> 5 );
Laurent Aimar's avatar
Laurent Aimar 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
        }
        src += i_src_stride;
        dst += i_dst_stride;
    }
}
static inline void mc_hc( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_width, int i_height )
{
    uint8_t *out;
    uint8_t *pix;
    int x, y;

    for( x = 0; x < i_width; x++ )
    {
        int tap[6];

        pix = &src[x];
        out = &dst[x];

        tap[0] = x264_tapfilter1( &pix[-2*i_src_stride] );
        tap[1] = x264_tapfilter1( &pix[-1*i_src_stride] );
        tap[2] = x264_tapfilter1( &pix[ 0*i_src_stride] );
        tap[3] = x264_tapfilter1( &pix[ 1*i_src_stride] );
        tap[4] = x264_tapfilter1( &pix[ 2*i_src_stride] );

        for( y = 0; y < i_height; y++ )
        {
            tap[5] = x264_tapfilter1( &pix[ 3*i_src_stride] );

Loren Merritt's avatar
Loren Merritt committed
208
            *out = x264_clip_uint8( ( tap[0] - 5*tap[1] + 20 * tap[2] + 20 * tap[3] -5*tap[4] + tap[5] + 512 ) >> 10 );
Laurent Aimar's avatar
Laurent Aimar committed
209 210 211 212 213 214 215 216 217 218 219 220 221

            /* Next line */
            pix += i_src_stride;
            out += i_dst_stride;
            tap[0] = tap[1];
            tap[1] = tap[2];
            tap[2] = tap[3];
            tap[3] = tap[4];
            tap[4] = tap[5];
        }
    }
}

222 223 224 225 226 227 228 229
static void hpel_filter( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src,
                         int stride, int width, int height )
{
    mc_hh( src, stride, dsth, stride, width, height );
    mc_hv( src, stride, dstv, stride, width, height );
    mc_hc( src, stride, dstc, stride, width, height );
}

Loren Merritt's avatar
Loren Merritt committed
230 231 232
static const int hpel_ref0[16] = {0,1,1,1,0,1,1,1,2,3,3,3,0,1,1,1};
static const int hpel_ref1[16] = {0,0,0,0,2,2,3,2,2,2,3,2,2,2,3,2};

233 234
static void mc_luma( uint8_t *dst,    int i_dst_stride,
                     uint8_t *src[4], int i_src_stride,
Loren Merritt's avatar
Loren Merritt committed
235
                     int mvx, int mvy,
Loren Merritt's avatar
Loren Merritt committed
236
                     int i_width, int i_height )
237
{
Loren Merritt's avatar
Loren Merritt committed
238 239 240
    int qpel_idx = ((mvy&3)<<2) + (mvx&3);
    int offset = (mvy>>2)*i_src_stride + (mvx>>2);
    uint8_t *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
241

Loren Merritt's avatar
Loren Merritt committed
242
    if( qpel_idx & 5 ) /* qpel interpolation needed */
243
    {
Loren Merritt's avatar
Loren Merritt committed
244
        uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
245 246 247 248 249 250 251 252 253
        pixel_avg( dst, i_dst_stride, src1, i_src_stride,
                   src2, i_src_stride, i_width, i_height );
    }
    else
    {
        mc_copy( src1, i_src_stride, dst, i_dst_stride, i_width, i_height );
    }
}

254 255
static uint8_t *get_ref( uint8_t *dst,   int *i_dst_stride,
                         uint8_t *src[4], int i_src_stride,
Loren Merritt's avatar
Loren Merritt committed
256
                         int mvx, int mvy,
Loren Merritt's avatar
Loren Merritt committed
257
                         int i_width, int i_height )
258
{
Loren Merritt's avatar
Loren Merritt committed
259 260 261
    int qpel_idx = ((mvy&3)<<2) + (mvx&3);
    int offset = (mvy>>2)*i_src_stride + (mvx>>2);
    uint8_t *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
262

Loren Merritt's avatar
Loren Merritt committed
263
    if( qpel_idx & 5 ) /* qpel interpolation needed */
264
    {
Loren Merritt's avatar
Loren Merritt committed
265
        uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
266 267 268 269 270 271 272 273 274 275 276
        pixel_avg( dst, *i_dst_stride, src1, i_src_stride,
                   src2, i_src_stride, i_width, i_height );
        return dst;
    }
    else
    {
        *i_dst_stride = i_src_stride;
        return src1;
    }
}

Laurent Aimar's avatar
Laurent Aimar committed
277
/* full chroma mc (ie until 1/8 pixel)*/
278 279 280 281
static void mc_chroma( uint8_t *dst, int i_dst_stride,
                       uint8_t *src, int i_src_stride,
                       int mvx, int mvy,
                       int i_width, int i_height )
Laurent Aimar's avatar
Laurent Aimar committed
282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310
{
    uint8_t *srcp;
    int x, y;

    const int d8x = mvx&0x07;
    const int d8y = mvy&0x07;

    const int cA = (8-d8x)*(8-d8y);
    const int cB = d8x    *(8-d8y);
    const int cC = (8-d8x)*d8y;
    const int cD = d8x    *d8y;

    src  += (mvy >> 3) * i_src_stride + (mvx >> 3);
    srcp = &src[i_src_stride];

    for( y = 0; y < i_height; y++ )
    {
        for( x = 0; x < i_width; x++ )
        {
            dst[x] = ( cA*src[x]  + cB*src[x+1] +
                       cC*srcp[x] + cD*srcp[x+1] + 32 ) >> 6;
        }
        dst  += i_dst_stride;

        src   = srcp;
        srcp += i_src_stride;
    }
}

Loren Merritt's avatar
Loren Merritt committed
311 312 313 314 315 316 317 318 319
#define MC_COPY(W) \
static void mc_copy_w##W( uint8_t *dst, int i_dst, uint8_t *src, int i_src, int i_height ) \
{ \
    mc_copy( src, i_src, dst, i_dst, W, i_height ); \
}
MC_COPY( 16 )
MC_COPY( 8 )
MC_COPY( 4 )

320 321 322 323 324 325 326 327 328 329 330
static void plane_copy( uint8_t *dst, int i_dst,
                        uint8_t *src, int i_src, int w, int h)
{
    while( h-- )
    {
        memcpy( dst, src, w );
        dst += i_dst;
        src += i_src;
    }
}

331 332 333 334 335 336 337
void prefetch_fenc_null( uint8_t *pix_y, int stride_y,
                         uint8_t *pix_uv, int stride_uv, int mb_x )
{}

void prefetch_ref_null( uint8_t *pix, int stride, int parity )
{}

338
void x264_mc_init( int cpu, x264_mc_functions_t *pf )
Laurent Aimar's avatar
Laurent Aimar committed
339
{
340 341
    pf->mc_luma   = mc_luma;
    pf->get_ref   = get_ref;
342
    pf->mc_chroma = mc_chroma;
Laurent Aimar's avatar
Laurent Aimar committed
343

344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365
    pf->avg[PIXEL_16x16]= pixel_avg_16x16;
    pf->avg[PIXEL_16x8] = pixel_avg_16x8;
    pf->avg[PIXEL_8x16] = pixel_avg_8x16;
    pf->avg[PIXEL_8x8]  = pixel_avg_8x8;
    pf->avg[PIXEL_8x4]  = pixel_avg_8x4;
    pf->avg[PIXEL_4x8]  = pixel_avg_4x8;
    pf->avg[PIXEL_4x4]  = pixel_avg_4x4;
    pf->avg[PIXEL_4x2]  = pixel_avg_4x2;
    pf->avg[PIXEL_2x4]  = pixel_avg_2x4;
    pf->avg[PIXEL_2x2]  = pixel_avg_2x2;
    
    pf->avg_weight[PIXEL_16x16]= pixel_avg_weight_16x16;
    pf->avg_weight[PIXEL_16x8] = pixel_avg_weight_16x8;
    pf->avg_weight[PIXEL_8x16] = pixel_avg_weight_8x16;
    pf->avg_weight[PIXEL_8x8]  = pixel_avg_weight_8x8;
    pf->avg_weight[PIXEL_8x4]  = pixel_avg_weight_8x4;
    pf->avg_weight[PIXEL_4x8]  = pixel_avg_weight_4x8;
    pf->avg_weight[PIXEL_4x4]  = pixel_avg_weight_4x4;
    pf->avg_weight[PIXEL_4x2]  = pixel_avg_weight_4x2;
    pf->avg_weight[PIXEL_2x4]  = pixel_avg_weight_2x4;
    pf->avg_weight[PIXEL_2x2]  = pixel_avg_weight_2x2;

Loren Merritt's avatar
Loren Merritt committed
366 367 368 369
    pf->copy[PIXEL_16x16] = mc_copy_w16;
    pf->copy[PIXEL_8x8]   = mc_copy_w8;
    pf->copy[PIXEL_4x4]   = mc_copy_w4;

370
    pf->plane_copy = plane_copy;
371
    pf->hpel_filter = hpel_filter;
372

373 374
    pf->prefetch_fenc = prefetch_fenc_null;
    pf->prefetch_ref  = prefetch_ref_null;
Fiona Glaser's avatar
Fiona Glaser committed
375
    pf->memcpy_aligned = memcpy;
376

377
#ifdef HAVE_MMX
Loren Merritt's avatar
Loren Merritt committed
378
    x264_mc_init_mmx( cpu, pf );
379
#endif
380
#ifdef ARCH_PPC
Laurent Aimar's avatar
Laurent Aimar committed
381 382 383 384 385
    if( cpu&X264_CPU_ALTIVEC )
        x264_mc_altivec_init( pf );
#endif
}

386
void x264_frame_filter( x264_t *h, x264_frame_t *frame, int mb_y, int b_end )
387
{
388
    const int b_interlaced = h->sh.b_mbaff;
389
    const int stride = frame->i_stride[0] << b_interlaced;
390
    const int width = frame->i_width[0];
391
    int start = (mb_y*16 >> b_interlaced) - 8; // buffer = 4 for deblock + 3 for 6tap, rounded to 8
Loren Merritt's avatar
Loren Merritt committed
392
    int height = ((b_end ? frame->i_lines[0] : mb_y*16) >> b_interlaced) + 8;
393
    int offs = start*stride - 8; // buffer = 3 for 6tap, aligned to 8 for simd
394 395
    int x, y;

Loren Merritt's avatar
Loren Merritt committed
396 397
    if( mb_y & b_interlaced )
        return;
398

399
    for( y=0; y<=b_interlaced; y++, offs+=frame->i_stride[0] )
400
    {
401
        h->mc.hpel_filter(
402 403 404 405
            frame->filtered[1] + offs,
            frame->filtered[2] + offs,
            frame->filtered[3] + offs,
            frame->plane[0] + offs,
406
            stride, width + 16, height - start );
407
    }
408 409

    /* generate integral image:
Loren Merritt's avatar
Loren Merritt committed
410 411
     * frame->integral contains 2 planes. in the upper plane, each element is
     * the sum of an 8x8 pixel region with top-left corner on that point.
Loren Merritt's avatar
Loren Merritt committed
412
     * in the lower plane, 4x4 sums (needed only with --partitions p4x4). */
413

Loren Merritt's avatar
Loren Merritt committed
414
    if( frame->integral )
415
    {
Loren Merritt's avatar
Loren Merritt committed
416 417
        if( start < 0 )
        {
418 419
            memset( frame->integral - PADV * stride - PADH, 0, stride * sizeof(uint16_t) );
            start = -PADV;
Loren Merritt's avatar
Loren Merritt committed
420 421
        }
        if( b_end )
422
            height += PADV-8;
Loren Merritt's avatar
Loren Merritt committed
423
        for( y = start; y < height; y++ )
424
        {
425 426
            uint8_t  *ref  = frame->plane[0] + y * stride - PADH;
            uint16_t *line = frame->integral + (y+1) * stride - PADH + 1;
427
            uint16_t v = line[0] = 0;
Loren Merritt's avatar
Loren Merritt committed
428
            for( x = 1; x < stride-1; x++ )
429
                line[x] = v += ref[x] + line[x-stride] - line[x-stride-1];
Loren Merritt's avatar
Loren Merritt committed
430
            line -= 8*stride;
431
            if( y >= 9-PADV )
Loren Merritt's avatar
Loren Merritt committed
432
            {
433
                uint16_t *sum4 = line + stride * (frame->i_lines[0] + PADV*2);
Loren Merritt's avatar
Loren Merritt committed
434 435 436 437 438
                for( x = 1; x < stride-8; x++, line++, sum4++ )
                {
                    sum4[0] =  line[4+4*stride] - line[4] - line[4*stride] + line[0];
                    line[0] += line[8+8*stride] - line[8] - line[8*stride];
                }
Loren Merritt's avatar
Loren Merritt committed
439 440
            }
        }
441
    }
442
}
443

444
void x264_frame_init_lowres( x264_t *h, x264_frame_t *frame )
445 446 447 448
{
    // FIXME: tapfilter?
    const int i_stride = frame->i_stride[0];
    const int i_stride2 = frame->i_stride_lowres;
449
    const int i_width2 = frame->i_width_lowres;
450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477
    int x, y, i;
    for( y = 0; y < frame->i_lines_lowres - 1; y++ )
    {
        uint8_t *src0 = &frame->plane[0][2*y*i_stride];
        uint8_t *src1 = src0+i_stride;
        uint8_t *src2 = src1+i_stride;
        uint8_t *dst0 = &frame->lowres[0][y*i_stride2];
        uint8_t *dsth = &frame->lowres[1][y*i_stride2];
        uint8_t *dstv = &frame->lowres[2][y*i_stride2];
        uint8_t *dstc = &frame->lowres[3][y*i_stride2];
        for( x = 0; x < i_width2 - 1; x++ )
        {
            dst0[x] = (src0[2*x  ] + src0[2*x+1] + src1[2*x  ] + src1[2*x+1] + 2) >> 2;
            dsth[x] = (src0[2*x+1] + src0[2*x+2] + src1[2*x+1] + src1[2*x+2] + 2) >> 2;
            dstv[x] = (src1[2*x  ] + src1[2*x+1] + src2[2*x  ] + src2[2*x+1] + 2) >> 2;
            dstc[x] = (src1[2*x+1] + src1[2*x+2] + src2[2*x+1] + src2[2*x+2] + 2) >> 2;
        }
        dst0[x] = (src0[2*x  ] + src0[2*x+1] + src1[2*x  ] + src1[2*x+1] + 2) >> 2;
        dstv[x] = (src1[2*x  ] + src1[2*x+1] + src2[2*x  ] + src2[2*x+1] + 2) >> 2;
        dsth[x] = (src0[2*x+1] + src1[2*x+1] + 1) >> 1;
        dstc[x] = (src1[2*x+1] + src2[2*x+1] + 1) >> 1;
    }
    for( i = 0; i < 4; i++ )
        memcpy( &frame->lowres[i][y*i_stride2], &frame->lowres[i][(y-1)*i_stride2], i_width2 );

    for( y = 0; y < 16; y++ )
        for( x = 0; x < 16; x++ )
            frame->i_cost_est[x][y] = -1;
478 479

    x264_frame_expand_border_lowres( frame );
480 481
}