mc.c 14.5 KB
Newer Older
Laurent Aimar's avatar
Laurent Aimar committed
1 2 3
/*****************************************************************************
 * mc.c: h264 encoder library (Motion Compensation)
 *****************************************************************************
4
 * Copyright (C) 2003-2008 x264 project
Laurent Aimar's avatar
Laurent Aimar committed
5 6
 *
 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7
 *          Loren Merritt <lorenm@u.washington.edu>
Laurent Aimar's avatar
Laurent Aimar committed
8 9 10 11 12 13 14 15 16 17 18 19 20
 *
 * 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
21
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
Laurent Aimar's avatar
Laurent Aimar committed
22 23
 *****************************************************************************/

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
#endif


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;
    }
}

52
static inline void pixel_avg_wxh( uint8_t *dst, int i_dst, uint8_t *src1, int i_src1, uint8_t *src2, int i_src2, int width, int height )
53 54 55 56 57 58
{
    int x, y;
    for( y = 0; y < height; y++ )
    {
        for( x = 0; x < width; x++ )
        {
59
            dst[x] = ( src1[x] + src2[x] + 1 ) >> 1;
60
        }
61 62
        src1 += i_src1;
        src2 += i_src2;
63 64 65 66 67 68
        dst += i_dst;
    }
}

#define PIXEL_AVG_C( name, width, height ) \
static void name( uint8_t *pix1, int i_stride_pix1, \
69 70
                  uint8_t *pix2, int i_stride_pix2, \
                  uint8_t *pix3, int i_stride_pix3 ) \
71
{ \
72
    pixel_avg_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, pix3, i_stride_pix3, width, height ); \
73 74 75 76 77 78 79 80 81 82 83 84 85 86 87
}
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 */
88 89 90
#define op_scale2(x) dst[x] = x264_clip_uint8( (src1[x]*i_weight1 + src2[x]*i_weight2 + (1<<5)) >> 6 )
static inline void pixel_avg_weight_wxh( uint8_t *dst, int i_dst, uint8_t *src1, int i_src1, uint8_t *src2, int i_src2, int width, int height, int i_weight1 )
{
91 92
    int y;
    const int i_weight2 = 64 - i_weight1;
93 94
    for( y = 0; y<height; y++, dst += i_dst, src1 += i_src1, src2 += i_src2 )
    {
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
        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, \
120 121
                uint8_t *pix2, int i_stride_pix2, \
                uint8_t *pix3, int i_stride_pix3, int i_weight1 ) \
122
{ \
123
    pixel_avg_weight_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, pix3, i_stride_pix3, width, height, i_weight1 ); \
124 125 126 127 128 129 130 131 132 133 134 135 136 137 138
}

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
139 140 141 142 143 144 145 146 147 148 149 150 151
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;
    }
}

Loren Merritt's avatar
Loren Merritt committed
152 153 154
#define TAPFILTER(pix, d) ((pix)[x-2*d] + (pix)[x+3*d] - 5*((pix)[x-d] + (pix)[x+2*d]) + 20*((pix)[x] + (pix)[x+d]))
static void hpel_filter( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src,
                         int stride, int width, int height )
Laurent Aimar's avatar
Laurent Aimar committed
155
{
Loren Merritt's avatar
Loren Merritt committed
156
    int16_t *buf = x264_malloc((width+5)*sizeof(int16_t));
Laurent Aimar's avatar
Laurent Aimar committed
157
    int x, y;
Loren Merritt's avatar
Loren Merritt committed
158
    for( y=0; y<height; y++ )
Laurent Aimar's avatar
Laurent Aimar committed
159
    {
Loren Merritt's avatar
Loren Merritt committed
160
        for( x=-2; x<width+3; x++ )
Laurent Aimar's avatar
Laurent Aimar committed
161
        {
Loren Merritt's avatar
Loren Merritt committed
162 163 164
            int v = TAPFILTER(src,stride);
            dstv[x] = x264_clip_uint8((v + 16) >> 5);
            buf[x+2] = v;
Laurent Aimar's avatar
Laurent Aimar committed
165
        }
Loren Merritt's avatar
Loren Merritt committed
166 167 168 169 170 171 172 173
        for( x=0; x<width; x++ )
            dstc[x] = x264_clip_uint8((TAPFILTER(buf+2,1) + 512) >> 10);
        for( x=0; x<width; x++ )
            dsth[x] = x264_clip_uint8((TAPFILTER(src,1) + 16) >> 5);
        dsth += stride;
        dstv += stride;
        dstc += stride;
        src += stride;
Laurent Aimar's avatar
Laurent Aimar committed
174
    }
Loren Merritt's avatar
Loren Merritt committed
175
    x264_free(buf);
176 177
}

Loren Merritt's avatar
Loren Merritt committed
178 179 180
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};

181 182
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
183
                     int mvx, int mvy,
Loren Merritt's avatar
Loren Merritt committed
184
                     int i_width, int i_height )
185
{
Loren Merritt's avatar
Loren Merritt committed
186 187 188
    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;
189

Loren Merritt's avatar
Loren Merritt committed
190
    if( qpel_idx & 5 ) /* qpel interpolation needed */
191
    {
Loren Merritt's avatar
Loren Merritt committed
192
        uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
193 194 195 196 197 198 199 200 201
        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 );
    }
}

202 203
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
204
                         int mvx, int mvy,
Loren Merritt's avatar
Loren Merritt committed
205
                         int i_width, int i_height )
206
{
Loren Merritt's avatar
Loren Merritt committed
207 208 209
    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;
210

Loren Merritt's avatar
Loren Merritt committed
211
    if( qpel_idx & 5 ) /* qpel interpolation needed */
212
    {
Loren Merritt's avatar
Loren Merritt committed
213
        uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
214 215 216 217 218 219 220 221 222 223 224
        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
225
/* full chroma mc (ie until 1/8 pixel)*/
226 227 228 229
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
230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258
{
    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
259 260 261 262 263 264 265 266 267
#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 )

268 269 270 271 272 273 274 275 276 277 278
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;
    }
}

Fiona Glaser's avatar
Fiona Glaser committed
279 280
static void prefetch_fenc_null( uint8_t *pix_y, int stride_y,
                                uint8_t *pix_uv, int stride_uv, int mb_x )
281 282
{}

Fiona Glaser's avatar
Fiona Glaser committed
283
static void prefetch_ref_null( uint8_t *pix, int stride, int parity )
284 285
{}

Fiona Glaser's avatar
Fiona Glaser committed
286 287 288 289 290
static void memzero_aligned( void * dst, int n )
{
    memset( dst, 0, n );
}

Loren Merritt's avatar
Loren Merritt committed
291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306
void x264_frame_init_lowres( x264_t *h, x264_frame_t *frame )
{
    uint8_t *src = frame->plane[0];
    int i_stride = frame->i_stride[0];
    int i_height = frame->i_lines[0];
    int i_width  = frame->i_width[0];
    int x, y;

    // duplicate last row and column so that their interpolation doesn't have to be special-cased
    for( y=0; y<i_height; y++ )
        src[i_width+y*i_stride] = src[i_width-1+y*i_stride];
    h->mc.memcpy_aligned( src+i_stride*i_height, src+i_stride*(i_height-1), i_width );
    h->mc.frame_init_lowres_core( src, frame->lowres[0], frame->lowres[1], frame->lowres[2], frame->lowres[3],
                                  i_stride, frame->i_stride_lowres, frame->i_width_lowres, frame->i_lines_lowres );
    x264_frame_expand_border_lowres( frame );

307 308 309 310 311
    memset( frame->i_cost_est, -1, sizeof(frame->i_cost_est) );

    for( x = 0; x < h->param.i_bframe + 2; x++ )
        for( y = 0; y < h->param.i_bframe + 2; y++ )
            frame->i_row_satds[y][x][0] = -1;
312 313 314 315

    for( y = 0; y <= !!h->param.i_bframe; y++ )
        for( x = 0; x <= h->param.i_bframe; x++ )
            frame->lowres_mvs[y][x][0][0] = 0x7FFF;
Loren Merritt's avatar
Loren Merritt committed
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
}

static void frame_init_lowres_core( uint8_t *src0, uint8_t *dst0, uint8_t *dsth, uint8_t *dstv, uint8_t *dstc,
                                    int src_stride, int dst_stride, int width, int height )
{
    int x,y;
    for( y=0; y<height; y++ )
    {
        uint8_t *src1 = src0+src_stride;
        uint8_t *src2 = src1+src_stride;
        for( x=0; x<width; x++ )
        {
            // slower than naive bilinear, but matches asm
#define FILTER(a,b,c,d) ((((a+b+1)>>1)+((c+d+1)>>1)+1)>>1)
            dst0[x] = FILTER(src0[2*x  ], src1[2*x  ], src0[2*x+1], src1[2*x+1]);
            dsth[x] = FILTER(src0[2*x+1], src1[2*x+1], src0[2*x+2], src1[2*x+2]);
            dstv[x] = FILTER(src1[2*x  ], src2[2*x  ], src1[2*x+1], src2[2*x+1]);
            dstc[x] = FILTER(src1[2*x+1], src2[2*x+1], src1[2*x+2], src2[2*x+2]);
#undef FILTER
        }
        src0 += src_stride*2;
        dst0 += dst_stride;
        dsth += dst_stride;
        dstv += dst_stride;
        dstc += dst_stride;
    }
}

344
void x264_mc_init( int cpu, x264_mc_functions_t *pf )
Laurent Aimar's avatar
Laurent Aimar committed
345
{
346 347
    pf->mc_luma   = mc_luma;
    pf->get_ref   = get_ref;
348
    pf->mc_chroma = mc_chroma;
Laurent Aimar's avatar
Laurent Aimar committed
349

350 351 352 353 354 355 356 357 358 359
    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;
Loren Merritt's avatar
Loren Merritt committed
360

361 362 363 364 365 366 367 368 369 370 371
    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
372 373 374 375
    pf->copy[PIXEL_16x16] = mc_copy_w16;
    pf->copy[PIXEL_8x8]   = mc_copy_w8;
    pf->copy[PIXEL_4x4]   = mc_copy_w4;

376
    pf->plane_copy = plane_copy;
377
    pf->hpel_filter = hpel_filter;
378

379 380
    pf->prefetch_fenc = prefetch_fenc_null;
    pf->prefetch_ref  = prefetch_ref_null;
Fiona Glaser's avatar
Fiona Glaser committed
381
    pf->memcpy_aligned = memcpy;
Fiona Glaser's avatar
Fiona Glaser committed
382
    pf->memzero_aligned = memzero_aligned;
Loren Merritt's avatar
Loren Merritt committed
383
    pf->frame_init_lowres_core = frame_init_lowres_core;
384

385
#ifdef HAVE_MMX
Loren Merritt's avatar
Loren Merritt committed
386
    x264_mc_init_mmx( cpu, pf );
387
#endif
388
#ifdef ARCH_PPC
Laurent Aimar's avatar
Laurent Aimar committed
389 390 391 392 393
    if( cpu&X264_CPU_ALTIVEC )
        x264_mc_altivec_init( pf );
#endif
}

394
void x264_frame_filter( x264_t *h, x264_frame_t *frame, int mb_y, int b_end )
395
{
396
    const int b_interlaced = h->sh.b_mbaff;
397
    const int stride = frame->i_stride[0] << b_interlaced;
398
    const int width = frame->i_width[0];
399
    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
400
    int height = ((b_end ? frame->i_lines[0] : mb_y*16) >> b_interlaced) + 8;
401
    int offs = start*stride - 8; // buffer = 3 for 6tap, aligned to 8 for simd
402 403
    int x, y;

Loren Merritt's avatar
Loren Merritt committed
404 405
    if( mb_y & b_interlaced )
        return;
406

407
    for( y=0; y<=b_interlaced; y++, offs+=frame->i_stride[0] )
408
    {
409
        h->mc.hpel_filter(
410 411 412 413
            frame->filtered[1] + offs,
            frame->filtered[2] + offs,
            frame->filtered[3] + offs,
            frame->plane[0] + offs,
414
            stride, width + 16, height - start );
415
    }
416 417

    /* generate integral image:
Loren Merritt's avatar
Loren Merritt committed
418 419
     * 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
420
     * in the lower plane, 4x4 sums (needed only with --partitions p4x4). */
421

Loren Merritt's avatar
Loren Merritt committed
422
    if( frame->integral )
423
    {
Loren Merritt's avatar
Loren Merritt committed
424 425
        if( start < 0 )
        {
426 427
            memset( frame->integral - PADV * stride - PADH, 0, stride * sizeof(uint16_t) );
            start = -PADV;
Loren Merritt's avatar
Loren Merritt committed
428 429
        }
        if( b_end )
430
            height += PADV-8;
Loren Merritt's avatar
Loren Merritt committed
431
        for( y = start; y < height; y++ )
432
        {
433 434
            uint8_t  *ref  = frame->plane[0] + y * stride - PADH;
            uint16_t *line = frame->integral + (y+1) * stride - PADH + 1;
435
            uint16_t v = line[0] = 0;
Loren Merritt's avatar
Loren Merritt committed
436
            for( x = 1; x < stride-1; x++ )
437
                line[x] = v += ref[x] + line[x-stride] - line[x-stride-1];
Loren Merritt's avatar
Loren Merritt committed
438
            line -= 8*stride;
439
            if( y >= 9-PADV )
Loren Merritt's avatar
Loren Merritt committed
440
            {
441
                uint16_t *sum4 = line + stride * (frame->i_lines[0] + PADV*2);
Loren Merritt's avatar
Loren Merritt committed
442 443 444 445 446
                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
447 448
            }
        }
449
    }
450
}