mc.c 17.9 KB
Newer Older
1 2 3
/*****************************************************************************
 * mc.c: h264 encoder library (Motion Compensation)
 *****************************************************************************
4
 * Copyright (C) 2003-2008 x264 project
5 6
 *
 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7
 *          Loren Merritt <lorenm@u.washington.edu>
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.
22 23
 *****************************************************************************/

24
#include "common.h"
25

Steven Walters's avatar
Steven Walters committed
26
#if HAVE_MMX
27
#include "x86/mc.h"
28
#endif
Steven Walters's avatar
Steven Walters committed
29
#if ARCH_PPC
30
#include "ppc/mc.h"
31
#endif
Steven Walters's avatar
Steven Walters committed
32
#if ARCH_ARM
33 34
#include "arm/mc.h"
#endif
35 36


37 38 39
static inline void pixel_avg( pixel *dst,  int i_dst_stride,
                              pixel *src1, int i_src1_stride,
                              pixel *src2, int i_src2_stride,
40 41
                              int i_width, int i_height )
{
42
    for( int y = 0; y < i_height; y++ )
43
    {
44
        for( int x = 0; x < i_width; x++ )
45 46 47 48 49 50 51
            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( pixel *dst, int i_dst, pixel *src1, int i_src1, pixel *src2, int i_src2, int width, int height )
53
{
54
    for( int y = 0; y < height; y++ )
55
    {
56
        for( int x = 0; x < width; x++ )
57 58 59
            dst[x] = ( src1[x] + src2[x] + 1 ) >> 1;
        src1 += i_src1;
        src2 += i_src2;
60 61 62 63 64 65
        dst += i_dst;
    }
}

/* Implicit weighted bipred only:
 * assumes log2_denom = 5, offset = 0, weight1 + weight2 = 64 */
66 67
#define op_scale2(x) dst[x] = x264_clip_pixel( (src1[x]*i_weight1 + src2[x]*i_weight2 + (1<<5)) >> 6 )
static inline void pixel_avg_weight_wxh( pixel *dst, int i_dst, pixel *src1, int i_src1, pixel *src2, int i_src2, int width, int height, int i_weight1 )
68
{
69
    const int i_weight2 = 64 - i_weight1;
70
    for( int y = 0; y<height; y++, dst += i_dst, src1 += i_src1, src2 += i_src2 )
71
    {
72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92
        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);
    }
}
Fiona Glaser's avatar
Fiona Glaser committed
93
#undef op_scale2
94

Fiona Glaser's avatar
Fiona Glaser committed
95
#define PIXEL_AVG_C( name, width, height ) \
96 97 98
static void name( pixel *pix1, int i_stride_pix1, \
                  pixel *pix2, int i_stride_pix2, \
                  pixel *pix3, int i_stride_pix3, int weight ) \
99
{ \
Anton Mitrofanov's avatar
Anton Mitrofanov committed
100
    if( weight == 32 ) \
Fiona Glaser's avatar
Fiona Glaser committed
101
        pixel_avg_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, pix3, i_stride_pix3, width, height ); \
Anton Mitrofanov's avatar
Anton Mitrofanov committed
102
    else \
Fiona Glaser's avatar
Fiona Glaser committed
103
        pixel_avg_weight_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, pix3, i_stride_pix3, width, height, weight ); \
104
}
Fiona Glaser's avatar
Fiona Glaser committed
105 106 107 108 109 110 111 112 113 114
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 )
115

116 117 118 119
static void x264_weight_cache( x264_t *h, x264_weight_t *w )
{
    w->weightfn = h->mc.weight;
}
120 121 122
#define opscale(x) dst[x] = x264_clip_pixel( ((src[x] * scale + (1<<(denom - 1))) >> denom) + offset )
#define opscale_noden(x) dst[x] = x264_clip_pixel( src[x] * scale + offset )
static void mc_weight( pixel *dst, int i_dst_stride, pixel *src, int i_src_stride, const x264_weight_t *weight, int i_width, int i_height )
123
{
124 125 126 127
    int offset = weight->i_offset << (BIT_DEPTH-8);
    int scale = weight->i_scale;
    int denom = weight->i_denom;
    if( denom >= 1 )
128
    {
129 130
        for( int y = 0; y < i_height; y++, dst += i_dst_stride, src += i_src_stride )
            for( int x = 0; x < i_width; x++ )
131 132 133 134
                opscale( x );
    }
    else
    {
135 136
        for( int y = 0; y < i_height; y++, dst += i_dst_stride, src += i_src_stride )
            for( int x = 0; x < i_width; x++ )
137 138 139 140
                opscale_noden( x );
    }
}

141
#define MC_WEIGHT_C( name, width ) \
142
    static void name( pixel *dst, int i_dst_stride, pixel *src, int i_src_stride, const x264_weight_t *weight, int height ) \
143
{ \
144
    mc_weight( dst, i_dst_stride, src, i_src_stride, weight, width, height );\
145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163
}

MC_WEIGHT_C( mc_weight_w20, 20 )
MC_WEIGHT_C( mc_weight_w16, 16 )
MC_WEIGHT_C( mc_weight_w12, 12 )
MC_WEIGHT_C( mc_weight_w8,   8 )
MC_WEIGHT_C( mc_weight_w4,   4 )
MC_WEIGHT_C( mc_weight_w2,   2 )

static weight_fn_t x264_mc_weight_wtab[6] =
{
    mc_weight_w2,
    mc_weight_w4,
    mc_weight_w8,
    mc_weight_w12,
    mc_weight_w16,
    mc_weight_w20,
};
const x264_weight_t weight_none[3] = { {{0}} };
164
static void mc_copy( pixel *src, int i_src_stride, pixel *dst, int i_dst_stride, int i_width, int i_height )
165
{
166
    for( int y = 0; y < i_height; y++ )
167
    {
168
        memcpy( dst, src, i_width * sizeof(pixel) );
169 170 171 172 173 174

        src += i_src_stride;
        dst += i_dst_stride;
    }
}

Loren Merritt's avatar
Loren Merritt committed
175
#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]))
176
static void hpel_filter( pixel *dsth, pixel *dstv, pixel *dstc, pixel *src,
177
                         int stride, int width, int height, dctcoef *buf )
178
{
179
    for( int y = 0; y < height; y++ )
180
    {
181
        for( int x = -2; x < width+3; x++ )
182
        {
Loren Merritt's avatar
Loren Merritt committed
183
            int v = TAPFILTER(src,stride);
184
            dstv[x] = x264_clip_pixel( (v + 16) >> 5 );
Loren Merritt's avatar
Loren Merritt committed
185
            buf[x+2] = v;
186
        }
187
        for( int x = 0; x < width; x++ )
188
            dstc[x] = x264_clip_pixel( (TAPFILTER(buf+2,1) + 512) >> 10 );
189
        for( int x = 0; x < width; x++ )
190
            dsth[x] = x264_clip_pixel( (TAPFILTER(src,1) + 16) >> 5 );
Loren Merritt's avatar
Loren Merritt committed
191 192 193 194
        dsth += stride;
        dstv += stride;
        dstc += stride;
        src += stride;
195
    }
196 197
}

198 199
static const uint8_t hpel_ref0[16] = {0,1,1,1,0,1,1,1,2,3,3,3,0,1,1,1};
static const uint8_t hpel_ref1[16] = {0,0,0,0,2,2,3,2,2,2,3,2,2,2,3,2};
200

201 202
static void mc_luma( pixel *dst,    int i_dst_stride,
                     pixel *src[4], int i_src_stride,
Loren Merritt's avatar
Loren Merritt committed
203
                     int mvx, int mvy,
204
                     int i_width, int i_height, const x264_weight_t *weight )
205
{
206 207
    int qpel_idx = ((mvy&3)<<2) + (mvx&3);
    int offset = (mvy>>2)*i_src_stride + (mvx>>2);
208
    pixel *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
209

210
    if( qpel_idx & 5 ) /* qpel interpolation needed */
211
    {
212
        pixel *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
213 214
        pixel_avg( dst, i_dst_stride, src1, i_src_stride,
                   src2, i_src_stride, i_width, i_height );
215 216
        if( weight->weightfn )
            mc_weight( dst, i_dst_stride, dst, i_dst_stride, weight, i_width, i_height );
217
    }
218 219
    else if( weight->weightfn )
        mc_weight( dst, i_dst_stride, src1, i_src_stride, weight, i_width, i_height );
220 221 222 223
    else
        mc_copy( src1, i_src_stride, dst, i_dst_stride, i_width, i_height );
}

224 225 226 227
static pixel *get_ref( pixel *dst,   int *i_dst_stride,
                       pixel *src[4], int i_src_stride,
                       int mvx, int mvy,
                       int i_width, int i_height, const x264_weight_t *weight )
228
{
229 230
    int qpel_idx = ((mvy&3)<<2) + (mvx&3);
    int offset = (mvy>>2)*i_src_stride + (mvx>>2);
231
    pixel *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
232

233
    if( qpel_idx & 5 ) /* qpel interpolation needed */
234
    {
235
        pixel *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
236 237
        pixel_avg( dst, *i_dst_stride, src1, i_src_stride,
                   src2, i_src_stride, i_width, i_height );
238 239 240 241 242 243 244
        if( weight->weightfn )
            mc_weight( dst, *i_dst_stride, dst, *i_dst_stride, weight, i_width, i_height );
        return dst;
    }
    else if( weight->weightfn )
    {
        mc_weight( dst, *i_dst_stride, src1, i_src_stride, weight, i_width, i_height );
245 246 247 248 249 250 251 252 253
        return dst;
    }
    else
    {
        *i_dst_stride = i_src_stride;
        return src1;
    }
}

254
/* full chroma mc (ie until 1/8 pixel)*/
255 256
static void mc_chroma( pixel *dst, int i_dst_stride,
                       pixel *src, int i_src_stride,
257 258
                       int mvx, int mvy,
                       int i_width, int i_height )
259
{
260
    pixel *srcp;
261

262 263 264 265 266 267
    int d8x = mvx&0x07;
    int d8y = mvy&0x07;
    int cA = (8-d8x)*(8-d8y);
    int cB = d8x    *(8-d8y);
    int cC = (8-d8x)*d8y;
    int cD = d8x    *d8y;
268

269
    src += (mvy >> 3) * i_src_stride + (mvx >> 3);
270 271
    srcp = &src[i_src_stride];

272
    for( int y = 0; y < i_height; y++ )
273
    {
274 275
        for( int x = 0; x < i_width; x++ )
            dst[x] = ( cA*src[x]  + cB*src[x+1] + cC*srcp[x] + cD*srcp[x+1] + 32 ) >> 6;
276 277 278 279 280 281
        dst  += i_dst_stride;
        src   = srcp;
        srcp += i_src_stride;
    }
}

Loren Merritt's avatar
Loren Merritt committed
282
#define MC_COPY(W) \
283
static void mc_copy_w##W( pixel *dst, int i_dst, pixel *src, int i_src, int i_height ) \
Loren Merritt's avatar
Loren Merritt committed
284 285 286 287 288 289 290
{ \
    mc_copy( src, i_src, dst, i_dst, W, i_height ); \
}
MC_COPY( 16 )
MC_COPY( 8 )
MC_COPY( 4 )

291
void x264_plane_copy_c( pixel *dst, int i_dst,
292 293 294 295
                        uint8_t *src, int i_src, int w, int h)
{
    while( h-- )
    {
296 297 298 299
#if X264_HIGH_BIT_DEPTH
        for( int i = 0; i < w; i++ )
            dst[i] = src[i] << (BIT_DEPTH-8);
#else
300
        memcpy( dst, src, w );
301
#endif
302 303 304 305 306
        dst += i_dst;
        src += i_src;
    }
}

307 308
static void prefetch_fenc_null( pixel *pix_y, int stride_y,
                                pixel *pix_uv, int stride_uv, int mb_x )
309 310
{}

311
static void prefetch_ref_null( pixel *pix, int stride, int parity )
312 313
{}

Fiona Glaser's avatar
Fiona Glaser committed
314 315 316 317 318
static void memzero_aligned( void * dst, int n )
{
    memset( dst, 0, n );
}

319
static void integral_init4h( uint16_t *sum, pixel *pix, int stride )
Loren Merritt's avatar
Loren Merritt committed
320
{
321 322
    int v = pix[0]+pix[1]+pix[2]+pix[3];
    for( int x = 0; x < stride-4; x++ )
Loren Merritt's avatar
Loren Merritt committed
323 324 325 326 327 328
    {
        sum[x] = v + sum[x-stride];
        v += pix[x+4] - pix[x];
    }
}

329
static void integral_init8h( uint16_t *sum, pixel *pix, int stride )
Loren Merritt's avatar
Loren Merritt committed
330
{
331 332
    int v = pix[0]+pix[1]+pix[2]+pix[3]+pix[4]+pix[5]+pix[6]+pix[7];
    for( int x = 0; x < stride-8; x++ )
Loren Merritt's avatar
Loren Merritt committed
333 334 335 336 337 338 339 340
    {
        sum[x] = v + sum[x-stride];
        v += pix[x+8] - pix[x];
    }
}

static void integral_init4v( uint16_t *sum8, uint16_t *sum4, int stride )
{
341
    for( int x = 0; x < stride-8; x++ )
Loren Merritt's avatar
Loren Merritt committed
342
        sum4[x] = sum8[x+4*stride] - sum8[x];
343
    for( int x = 0; x < stride-8; x++ )
Loren Merritt's avatar
Loren Merritt committed
344 345 346 347 348
        sum8[x] = sum8[x+8*stride] + sum8[x+8*stride+4] - sum8[x] - sum8[x+4];
}

static void integral_init8v( uint16_t *sum8, int stride )
{
349
    for( int x = 0; x < stride-8; x++ )
Loren Merritt's avatar
Loren Merritt committed
350 351 352
        sum8[x] = sum8[x+8*stride] - sum8[x];
}

Loren Merritt's avatar
Loren Merritt committed
353 354
void x264_frame_init_lowres( x264_t *h, x264_frame_t *frame )
{
355
    pixel *src = frame->plane[0];
Loren Merritt's avatar
Loren Merritt committed
356 357 358 359 360
    int i_stride = frame->i_stride[0];
    int i_height = frame->i_lines[0];
    int i_width  = frame->i_width[0];

    // duplicate last row and column so that their interpolation doesn't have to be special-cased
361
    for( int y = 0; y < i_height; y++ )
Loren Merritt's avatar
Loren Merritt committed
362
        src[i_width+y*i_stride] = src[i_width-1+y*i_stride];
363
    memcpy( src+i_stride*i_height, src+i_stride*(i_height-1), (i_width+1) * sizeof(pixel) );
Loren Merritt's avatar
Loren Merritt committed
364 365 366 367
    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 );

368 369
    memset( frame->i_cost_est, -1, sizeof(frame->i_cost_est) );

370 371
    for( int y = 0; y < h->param.i_bframe + 2; y++ )
        for( int x = 0; x < h->param.i_bframe + 2; x++ )
372
            frame->i_row_satds[y][x][0] = -1;
373

374 375
    for( int y = 0; y <= !!h->param.i_bframe; y++ )
        for( int x = 0; x <= h->param.i_bframe; x++ )
376
            frame->lowres_mvs[y][x][0][0] = 0x7FFF;
Loren Merritt's avatar
Loren Merritt committed
377 378
}

379
static void frame_init_lowres_core( pixel *src0, pixel *dst0, pixel *dsth, pixel *dstv, pixel *dstc,
Loren Merritt's avatar
Loren Merritt committed
380 381
                                    int src_stride, int dst_stride, int width, int height )
{
382
    for( int y = 0; y < height; y++ )
Loren Merritt's avatar
Loren Merritt committed
383
    {
384 385
        pixel *src1 = src0+src_stride;
        pixel *src2 = src1+src_stride;
386
        for( int x = 0; x<width; x++ )
Loren Merritt's avatar
Loren Merritt committed
387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403
        {
            // 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;
    }
}

Steven Walters's avatar
Steven Walters committed
404
#if defined(__GNUC__) && (ARCH_X86 || ARCH_X86_64)
405
// gcc isn't smart enough to use the "idiv" instruction
406 407
static ALWAYS_INLINE int32_t div_64_32(int64_t x, int32_t y)
{
408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423
    int32_t quotient, remainder;
    asm("idiv %4"
        :"=a"(quotient), "=d"(remainder)
        :"a"((uint32_t)x), "d"((int32_t)(x>>32)), "r"(y)
    );
    return quotient;
}
#else
#define div_64_32(x,y) ((x)/(y))
#endif

/* Estimate the total amount of influence on future quality that could be had if we
 * were to improve the reference samples used to inter predict any given macroblock. */
static void mbtree_propagate_cost( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
                                   uint16_t *inter_costs, uint16_t *inv_qscales, int len )
{
424
    for( int i = 0; i < len; i++ )
425 426
    {
        int propagate_amount = propagate_in[i] + ((intra_costs[i] * inv_qscales[i] + 128)>>8);
427
        dst[i] = div_64_32((int64_t)propagate_amount * (intra_costs[i] - (inter_costs[i] & LOWRES_COST_MASK)), intra_costs[i]);
428 429 430
    }
}

431
void x264_mc_init( int cpu, x264_mc_functions_t *pf )
432
{
433 434
    pf->mc_luma   = mc_luma;
    pf->get_ref   = get_ref;
435
    pf->mc_chroma = mc_chroma;
436

437 438 439 440 441 442 443 444 445 446
    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
447

448 449 450 451 452
    pf->weight    = x264_mc_weight_wtab;
    pf->offsetadd = x264_mc_weight_wtab;
    pf->offsetsub = x264_mc_weight_wtab;
    pf->weight_cache = x264_weight_cache;

453
    pf->copy_16x16_unaligned = mc_copy_w16;
Loren Merritt's avatar
Loren Merritt committed
454 455 456 457
    pf->copy[PIXEL_16x16] = mc_copy_w16;
    pf->copy[PIXEL_8x8]   = mc_copy_w8;
    pf->copy[PIXEL_4x4]   = mc_copy_w4;

458
    pf->plane_copy = x264_plane_copy_c;
459
    pf->hpel_filter = hpel_filter;
460

461 462
    pf->prefetch_fenc = prefetch_fenc_null;
    pf->prefetch_ref  = prefetch_ref_null;
Fiona Glaser's avatar
Fiona Glaser committed
463
    pf->memcpy_aligned = memcpy;
Fiona Glaser's avatar
Fiona Glaser committed
464
    pf->memzero_aligned = memzero_aligned;
Loren Merritt's avatar
Loren Merritt committed
465
    pf->frame_init_lowres_core = frame_init_lowres_core;
466

Loren Merritt's avatar
Loren Merritt committed
467 468 469 470 471
    pf->integral_init4h = integral_init4h;
    pf->integral_init8h = integral_init8h;
    pf->integral_init4v = integral_init4v;
    pf->integral_init8v = integral_init8v;

472 473
    pf->mbtree_propagate_cost = mbtree_propagate_cost;

Steven Walters's avatar
Steven Walters committed
474
#if HAVE_MMX
Loren Merritt's avatar
Loren Merritt committed
475
    x264_mc_init_mmx( cpu, pf );
476
#endif
Steven Walters's avatar
Steven Walters committed
477
#if HAVE_ALTIVEC
478 479 480
    if( cpu&X264_CPU_ALTIVEC )
        x264_mc_altivec_init( pf );
#endif
Steven Walters's avatar
Steven Walters committed
481
#if HAVE_ARMV6
482 483
    x264_mc_init_arm( cpu, pf );
#endif
484 485
}

486
void x264_frame_filter( x264_t *h, x264_frame_t *frame, int mb_y, int b_end )
487
{
488
    const int b_interlaced = h->sh.b_mbaff;
489
    const int stride = frame->i_stride[0] << b_interlaced;
490
    const int width = frame->i_width[0];
491
    int start = (mb_y*16 >> b_interlaced) - 8; // buffer = 4 for deblock + 3 for 6tap, rounded to 8
492
    int height = ((b_end ? frame->i_lines[0] : mb_y*16) >> b_interlaced) + 8;
493
    int offs = start*stride - 8; // buffer = 3 for 6tap, aligned to 8 for simd
494

Loren Merritt's avatar
Loren Merritt committed
495 496
    if( mb_y & b_interlaced )
        return;
497

498
    for( int y = 0; y <= b_interlaced; y++, offs += frame->i_stride[0] )
499
    {
500
        h->mc.hpel_filter(
501 502 503 504
            frame->filtered[1] + offs,
            frame->filtered[2] + offs,
            frame->filtered[3] + offs,
            frame->plane[0] + offs,
505 506
            stride, width + 16, height - start,
            h->scratch_buffer );
507
    }
508 509

    /* generate integral image:
Loren Merritt's avatar
Loren Merritt committed
510 511
     * 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.
512
     * in the lower plane, 4x4 sums (needed only with --partitions p4x4). */
513

514
    if( frame->integral )
515
    {
516 517
        if( start < 0 )
        {
518 519
            memset( frame->integral - PADV * stride - PADH, 0, stride * sizeof(uint16_t) );
            start = -PADV;
520 521
        }
        if( b_end )
522
            height += PADV-9;
523
        for( int y = start; y < height; y++ )
524
        {
525
            pixel    *pix  = frame->plane[0] + y * stride - PADH;
Loren Merritt's avatar
Loren Merritt committed
526 527 528 529 530 531 532 533 534 535 536
            uint16_t *sum8 = frame->integral + (y+1) * stride - PADH;
            uint16_t *sum4;
            if( h->frames.b_have_sub8x8_esa )
            {
                h->mc.integral_init4h( sum8, pix, stride );
                sum8 -= 8*stride;
                sum4 = sum8 + stride * (frame->i_lines[0] + PADV*2);
                if( y >= 8-PADV )
                    h->mc.integral_init4v( sum8, sum4, stride );
            }
            else
Loren Merritt's avatar
Loren Merritt committed
537
            {
Loren Merritt's avatar
Loren Merritt committed
538 539 540
                h->mc.integral_init8h( sum8, pix, stride );
                if( y >= 8-PADV )
                    h->mc.integral_init8v( sum8-8*stride, stride );
Loren Merritt's avatar
Loren Merritt committed
541 542
            }
        }
543
    }
544
}