mc.c 25.8 KB
Newer Older
1
/*****************************************************************************
2
 * mc.c: motion compensation
3
 *****************************************************************************
Henrik Gramner's avatar
Henrik Gramner committed
4
 * Copyright (C) 2003-2016 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
 *
 * This program is also available under a commercial proprietary license.
 * For more information, contact us at licensing@x264.com.
25 26
 *****************************************************************************/

27
#include "common.h"
28

Steven Walters's avatar
Steven Walters committed
29
#if HAVE_MMX
30
#include "x86/mc.h"
31
#endif
Steven Walters's avatar
Steven Walters committed
32
#if ARCH_PPC
33
#include "ppc/mc.h"
34
#endif
Steven Walters's avatar
Steven Walters committed
35
#if ARCH_ARM
36 37
#include "arm/mc.h"
#endif
38 39 40
#if ARCH_AARCH64
#include "aarch64/mc.h"
#endif
41 42 43
#if ARCH_MIPS
#include "mips/mc.h"
#endif
44 45


46 47 48
static inline void pixel_avg( pixel *dst,  intptr_t i_dst_stride,
                              pixel *src1, intptr_t i_src1_stride,
                              pixel *src2, intptr_t i_src2_stride, int i_width, int i_height )
49
{
50
    for( int y = 0; y < i_height; y++ )
51
    {
52
        for( int x = 0; x < i_width; x++ )
53 54 55 56 57 58 59
            dst[x] = ( src1[x] + src2[x] + 1 ) >> 1;
        dst  += i_dst_stride;
        src1 += i_src1_stride;
        src2 += i_src2_stride;
    }
}

60 61 62
static inline void pixel_avg_wxh( pixel *dst,  intptr_t i_dst,
                                  pixel *src1, intptr_t i_src1,
                                  pixel *src2, intptr_t i_src2, int width, int height )
63
{
64
    for( int y = 0; y < height; y++ )
65
    {
66
        for( int x = 0; x < width; x++ )
67 68 69
            dst[x] = ( src1[x] + src2[x] + 1 ) >> 1;
        src1 += i_src1;
        src2 += i_src2;
70 71 72 73 74 75
        dst += i_dst;
    }
}

/* Implicit weighted bipred only:
 * assumes log2_denom = 5, offset = 0, weight1 + weight2 = 64 */
76 77 78
static inline void pixel_avg_weight_wxh( pixel *dst,  intptr_t i_dst,
                                         pixel *src1, intptr_t i_src1,
                                         pixel *src2, intptr_t i_src2, int width, int height, int i_weight1 )
79
{
80
    int i_weight2 = 64 - i_weight1;
81
    for( int y = 0; y<height; y++, dst += i_dst, src1 += i_src1, src2 += i_src2 )
82 83
        for( int x = 0; x<width; x++ )
            dst[x] = x264_clip_pixel( (src1[x]*i_weight1 + src2[x]*i_weight2 + (1<<5)) >> 6 );
84
}
Fiona Glaser's avatar
Fiona Glaser committed
85
#undef op_scale2
86

Fiona Glaser's avatar
Fiona Glaser committed
87
#define PIXEL_AVG_C( name, width, height ) \
88 89 90
static void name( pixel *pix1, intptr_t i_stride_pix1, \
                  pixel *pix2, intptr_t i_stride_pix2, \
                  pixel *pix3, intptr_t i_stride_pix3, int weight ) \
91
{ \
Anton Mitrofanov's avatar
Anton Mitrofanov committed
92
    if( weight == 32 ) \
Fiona Glaser's avatar
Fiona Glaser committed
93
        pixel_avg_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, pix3, i_stride_pix3, width, height ); \
Anton Mitrofanov's avatar
Anton Mitrofanov committed
94
    else \
Fiona Glaser's avatar
Fiona Glaser committed
95
        pixel_avg_weight_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, pix3, i_stride_pix3, width, height, weight ); \
96
}
Fiona Glaser's avatar
Fiona Glaser committed
97 98 99 100 101
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 )
Henrik Gramner's avatar
Henrik Gramner committed
102
PIXEL_AVG_C( pixel_avg_4x16,  4, 16 )
Fiona Glaser's avatar
Fiona Glaser committed
103 104 105
PIXEL_AVG_C( pixel_avg_4x8,   4, 8 )
PIXEL_AVG_C( pixel_avg_4x4,   4, 4 )
PIXEL_AVG_C( pixel_avg_4x2,   4, 2 )
Henrik Gramner's avatar
Henrik Gramner committed
106
PIXEL_AVG_C( pixel_avg_2x8,   2, 8 )
Fiona Glaser's avatar
Fiona Glaser committed
107 108
PIXEL_AVG_C( pixel_avg_2x4,   2, 4 )
PIXEL_AVG_C( pixel_avg_2x2,   2, 2 )
109

110 111 112 113
static void x264_weight_cache( x264_t *h, x264_weight_t *w )
{
    w->weightfn = h->mc.weight;
}
114 115
#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 )
116 117
static void mc_weight( pixel *dst, intptr_t i_dst_stride, pixel *src, intptr_t i_src_stride,
                       const x264_weight_t *weight, int i_width, int i_height )
118
{
119 120 121 122
    int offset = weight->i_offset << (BIT_DEPTH-8);
    int scale = weight->i_scale;
    int denom = weight->i_denom;
    if( denom >= 1 )
123
    {
124 125
        for( int y = 0; y < i_height; y++, dst += i_dst_stride, src += i_src_stride )
            for( int x = 0; x < i_width; x++ )
126 127 128 129
                opscale( x );
    }
    else
    {
130 131
        for( int y = 0; y < i_height; y++, dst += i_dst_stride, src += i_src_stride )
            for( int x = 0; x < i_width; x++ )
132 133 134 135
                opscale_noden( x );
    }
}

136
#define MC_WEIGHT_C( name, width ) \
137
    static void name( pixel *dst, intptr_t i_dst_stride, pixel *src, intptr_t i_src_stride, const x264_weight_t *weight, int height ) \
138
{ \
139
    mc_weight( dst, i_dst_stride, src, i_src_stride, weight, width, height );\
140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157
}

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,
};
Anton Mitrofanov's avatar
Anton Mitrofanov committed
158
const x264_weight_t x264_weight_none[3] = { {{0}} };
159
static void mc_copy( pixel *src, intptr_t i_src_stride, pixel *dst, intptr_t i_dst_stride, int i_width, int i_height )
160
{
161
    for( int y = 0; y < i_height; y++ )
162
    {
163
        memcpy( dst, src, i_width * sizeof(pixel) );
164 165 166 167 168 169

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

Loren Merritt's avatar
Loren Merritt committed
170
#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]))
171
static void hpel_filter( pixel *dsth, pixel *dstv, pixel *dstc, pixel *src,
172
                         intptr_t stride, int width, int height, int16_t *buf )
173
{
174
    const int pad = (BIT_DEPTH > 9) ? (-10 * PIXEL_MAX) : 0;
175
    for( int y = 0; y < height; y++ )
176
    {
177
        for( int x = -2; x < width+3; x++ )
178
        {
Loren Merritt's avatar
Loren Merritt committed
179
            int v = TAPFILTER(src,stride);
180
            dstv[x] = x264_clip_pixel( (v + 16) >> 5 );
181 182
            /* transform v for storage in a 16-bit integer */
            buf[x+2] = v + pad;
183
        }
184
        for( int x = 0; x < width; x++ )
185
            dstc[x] = x264_clip_pixel( (TAPFILTER(buf+2,1) - 32*pad + 512) >> 10 );
186
        for( int x = 0; x < width; x++ )
187
            dsth[x] = x264_clip_pixel( (TAPFILTER(src,1) + 16) >> 5 );
Loren Merritt's avatar
Loren Merritt committed
188 189 190 191
        dsth += stride;
        dstv += stride;
        dstc += stride;
        src += stride;
192
    }
193 194
}

195 196
const uint8_t x264_hpel_ref0[16] = {0,1,1,1,0,1,1,1,2,3,3,3,0,1,1,1};
const uint8_t x264_hpel_ref1[16] = {0,0,1,0,2,2,3,2,2,2,3,2,2,2,3,2};
197

198 199
static void mc_luma( pixel *dst,    intptr_t i_dst_stride,
                     pixel *src[4], intptr_t i_src_stride,
Loren Merritt's avatar
Loren Merritt committed
200
                     int mvx, int mvy,
201
                     int i_width, int i_height, const x264_weight_t *weight )
202
{
203 204
    int qpel_idx = ((mvy&3)<<2) + (mvx&3);
    int offset = (mvy>>2)*i_src_stride + (mvx>>2);
205
    pixel *src1 = src[x264_hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
206

207
    if( qpel_idx & 5 ) /* qpel interpolation needed */
208
    {
209
        pixel *src2 = src[x264_hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
210 211
        pixel_avg( dst, i_dst_stride, src1, i_src_stride,
                   src2, i_src_stride, i_width, i_height );
212 213
        if( weight->weightfn )
            mc_weight( dst, i_dst_stride, dst, i_dst_stride, weight, i_width, i_height );
214
    }
215 216
    else if( weight->weightfn )
        mc_weight( dst, i_dst_stride, src1, i_src_stride, weight, i_width, i_height );
217 218 219 220
    else
        mc_copy( src1, i_src_stride, dst, i_dst_stride, i_width, i_height );
}

221 222
static pixel *get_ref( pixel *dst,   intptr_t *i_dst_stride,
                       pixel *src[4], intptr_t i_src_stride,
223 224
                       int mvx, int mvy,
                       int i_width, int i_height, const x264_weight_t *weight )
225
{
226 227
    int qpel_idx = ((mvy&3)<<2) + (mvx&3);
    int offset = (mvy>>2)*i_src_stride + (mvx>>2);
228
    pixel *src1 = src[x264_hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
229

230
    if( qpel_idx & 5 ) /* qpel interpolation needed */
231
    {
232
        pixel *src2 = src[x264_hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
233 234
        pixel_avg( dst, *i_dst_stride, src1, i_src_stride,
                   src2, i_src_stride, i_width, i_height );
235 236 237 238 239 240 241
        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 );
242 243 244 245 246 247 248 249 250
        return dst;
    }
    else
    {
        *i_dst_stride = i_src_stride;
        return src1;
    }
}

251
/* full chroma mc (ie until 1/8 pixel)*/
252 253
static void mc_chroma( pixel *dstu, pixel *dstv, intptr_t i_dst_stride,
                       pixel *src, intptr_t i_src_stride,
254 255
                       int mvx, int mvy,
                       int i_width, int i_height )
256
{
257
    pixel *srcp;
258

259 260 261 262 263 264
    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;
265

266
    src += (mvy >> 3) * i_src_stride + (mvx >> 3)*2;
267 268
    srcp = &src[i_src_stride];

269
    for( int y = 0; y < i_height; y++ )
270
    {
271
        for( int x = 0; x < i_width; x++ )
272 273 274 275 276 277 278 279
        {
            dstu[x] = ( cA*src[2*x]  + cB*src[2*x+2] +
                        cC*srcp[2*x] + cD*srcp[2*x+2] + 32 ) >> 6;
            dstv[x] = ( cA*src[2*x+1]  + cB*src[2*x+3] +
                        cC*srcp[2*x+1] + cD*srcp[2*x+3] + 32 ) >> 6;
        }
        dstu += i_dst_stride;
        dstv += i_dst_stride;
280 281 282 283 284
        src   = srcp;
        srcp += i_src_stride;
    }
}

Loren Merritt's avatar
Loren Merritt committed
285
#define MC_COPY(W) \
286
static void mc_copy_w##W( pixel *dst, intptr_t i_dst, pixel *src, intptr_t i_src, int i_height ) \
Loren Merritt's avatar
Loren Merritt committed
287 288 289 290 291 292 293
{ \
    mc_copy( src, i_src, dst, i_dst, W, i_height ); \
}
MC_COPY( 16 )
MC_COPY( 8 )
MC_COPY( 4 )

294 295
void x264_plane_copy_c( pixel *dst, intptr_t i_dst,
                        pixel *src, intptr_t i_src, int w, int h )
296 297 298
{
    while( h-- )
    {
299
        memcpy( dst, src, w * sizeof(pixel) );
300 301 302 303 304
        dst += i_dst;
        src += i_src;
    }
}

Xiaolei Yu's avatar
Xiaolei Yu committed
305 306 307 308 309 310 311 312 313 314 315
void x264_plane_copy_swap_c( pixel *dst, intptr_t i_dst,
                             pixel *src, intptr_t i_src, int w, int h )
{
    for( int y=0; y<h; y++, dst+=i_dst, src+=i_src )
        for( int x=0; x<2*w; x+=2 )
        {
            dst[x]   = src[x+1];
            dst[x+1] = src[x];
        }
}

316 317 318
void x264_plane_copy_interleave_c( pixel *dst,  intptr_t i_dst,
                                   pixel *srcu, intptr_t i_srcu,
                                   pixel *srcv, intptr_t i_srcv, int w, int h )
319 320 321 322
{
    for( int y=0; y<h; y++, dst+=i_dst, srcu+=i_srcu, srcv+=i_srcv )
        for( int x=0; x<w; x++ )
        {
323 324
            dst[2*x]   = srcu[x];
            dst[2*x+1] = srcv[x];
325 326 327
        }
}

328 329 330
static void x264_plane_copy_deinterleave_c( pixel *dstu, intptr_t i_dstu,
                                            pixel *dstv, intptr_t i_dstv,
                                            pixel *src,  intptr_t i_src, int w, int h )
331 332 333 334 335 336 337 338 339
{
    for( int y=0; y<h; y++, dstu+=i_dstu, dstv+=i_dstv, src+=i_src )
        for( int x=0; x<w; x++ )
        {
            dstu[x] = src[2*x];
            dstv[x] = src[2*x+1];
        }
}

340 341 342 343
static void x264_plane_copy_deinterleave_rgb_c( pixel *dsta, intptr_t i_dsta,
                                                pixel *dstb, intptr_t i_dstb,
                                                pixel *dstc, intptr_t i_dstc,
                                                pixel *src,  intptr_t i_src, int pw, int w, int h )
xvidfan's avatar
xvidfan committed
344 345 346 347 348 349 350 351 352 353 354 355
{
    for( int y=0; y<h; y++, dsta+=i_dsta, dstb+=i_dstb, dstc+=i_dstc, src+=i_src )
    {
        for( int x=0; x<w; x++ )
        {
            dsta[x] = src[x*pw];
            dstb[x] = src[x*pw+1];
            dstc[x] = src[x*pw+2];
        }
    }
}

James Weaver's avatar
James Weaver committed
356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383
void x264_plane_copy_deinterleave_v210_c( pixel *dsty, intptr_t i_dsty,
                                          pixel *dstc, intptr_t i_dstc,
                                          uint32_t *src, intptr_t i_src, int w, int h )
{
    for( int l = 0; l < h; l++ )
    {
        pixel *dsty0 = dsty;
        pixel *dstc0 = dstc;
        uint32_t *src0 = src;

        for( int n = 0; n < w; n += 3 )
        {
            *(dstc0++) = *src0 & 0x03FF;
            *(dsty0++) = ( *src0 >> 10 ) & 0x03FF;
            *(dstc0++) = ( *src0 >> 20 ) & 0x03FF;
            src0++;
            *(dsty0++) = *src0 & 0x03FF;
            *(dstc0++) = ( *src0 >> 10 ) & 0x03FF;
            *(dsty0++) = ( *src0 >> 20 ) & 0x03FF;
            src0++;
        }

        dsty += i_dsty;
        dstc += i_dstc;
        src  += i_src;
    }
}

384
static void store_interleave_chroma( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height )
385
{
Henrik Gramner's avatar
Henrik Gramner committed
386
    for( int y=0; y<height; y++, dst+=i_dst, srcu+=FDEC_STRIDE, srcv+=FDEC_STRIDE )
387 388 389 390 391 392 393
        for( int x=0; x<8; x++ )
        {
            dst[2*x]   = srcu[x];
            dst[2*x+1] = srcv[x];
        }
}

394
static void load_deinterleave_chroma_fenc( pixel *dst, pixel *src, intptr_t i_src, int height )
395
{
Henrik Gramner's avatar
Henrik Gramner committed
396
    x264_plane_copy_deinterleave_c( dst, FENC_STRIDE, dst+FENC_STRIDE/2, FENC_STRIDE, src, i_src, 8, height );
397 398
}

399
static void load_deinterleave_chroma_fdec( pixel *dst, pixel *src, intptr_t i_src, int height )
400
{
Henrik Gramner's avatar
Henrik Gramner committed
401
    x264_plane_copy_deinterleave_c( dst, FDEC_STRIDE, dst+FDEC_STRIDE/2, FDEC_STRIDE, src, i_src, 8, height );
402 403
}

404 405
static void prefetch_fenc_null( pixel *pix_y,  intptr_t stride_y,
                                pixel *pix_uv, intptr_t stride_uv, int mb_x )
406 407
{}

408
static void prefetch_ref_null( pixel *pix, intptr_t stride, int parity )
409 410
{}

411
static void memzero_aligned( void * dst, size_t n )
Fiona Glaser's avatar
Fiona Glaser committed
412 413 414 415
{
    memset( dst, 0, n );
}

416
static void integral_init4h( uint16_t *sum, pixel *pix, intptr_t stride )
Loren Merritt's avatar
Loren Merritt committed
417
{
418 419
    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
420 421 422 423 424 425
    {
        sum[x] = v + sum[x-stride];
        v += pix[x+4] - pix[x];
    }
}

426
static void integral_init8h( uint16_t *sum, pixel *pix, intptr_t stride )
Loren Merritt's avatar
Loren Merritt committed
427
{
428 429
    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
430 431 432 433 434 435
    {
        sum[x] = v + sum[x-stride];
        v += pix[x+8] - pix[x];
    }
}

436
static void integral_init4v( uint16_t *sum8, uint16_t *sum4, intptr_t stride )
Loren Merritt's avatar
Loren Merritt committed
437
{
438
    for( int x = 0; x < stride-8; x++ )
Loren Merritt's avatar
Loren Merritt committed
439
        sum4[x] = sum8[x+4*stride] - sum8[x];
440
    for( int x = 0; x < stride-8; x++ )
Loren Merritt's avatar
Loren Merritt committed
441 442 443
        sum8[x] = sum8[x+8*stride] + sum8[x+8*stride+4] - sum8[x] - sum8[x+4];
}

444
static void integral_init8v( uint16_t *sum8, intptr_t stride )
Loren Merritt's avatar
Loren Merritt committed
445
{
446
    for( int x = 0; x < stride-8; x++ )
Loren Merritt's avatar
Loren Merritt committed
447 448 449
        sum8[x] = sum8[x+8*stride] - sum8[x];
}

Loren Merritt's avatar
Loren Merritt committed
450 451
void x264_frame_init_lowres( x264_t *h, x264_frame_t *frame )
{
452
    pixel *src = frame->plane[0];
Loren Merritt's avatar
Loren Merritt committed
453 454 455 456 457
    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
458
    for( int y = 0; y < i_height; y++ )
Loren Merritt's avatar
Loren Merritt committed
459
        src[i_width+y*i_stride] = src[i_width-1+y*i_stride];
460
    memcpy( src+i_stride*i_height, src+i_stride*(i_height-1), (i_width+1) * sizeof(pixel) );
Loren Merritt's avatar
Loren Merritt committed
461 462 463 464
    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 );

465 466
    memset( frame->i_cost_est, -1, sizeof(frame->i_cost_est) );

467 468
    for( int y = 0; y < h->param.i_bframe + 2; y++ )
        for( int x = 0; x < h->param.i_bframe + 2; x++ )
469
            frame->i_row_satds[y][x][0] = -1;
470

471 472
    for( int y = 0; y <= !!h->param.i_bframe; y++ )
        for( int x = 0; x <= h->param.i_bframe; x++ )
473
            frame->lowres_mvs[y][x][0][0] = 0x7FFF;
Loren Merritt's avatar
Loren Merritt committed
474 475
}

476
static void frame_init_lowres_core( pixel *src0, pixel *dst0, pixel *dsth, pixel *dstv, pixel *dstc,
477
                                    intptr_t src_stride, intptr_t dst_stride, int width, int height )
Loren Merritt's avatar
Loren Merritt committed
478
{
479
    for( int y = 0; y < height; y++ )
Loren Merritt's avatar
Loren Merritt committed
480
    {
481 482
        pixel *src1 = src0+src_stride;
        pixel *src2 = src1+src_stride;
483
        for( int x = 0; x<width; x++ )
Loren Merritt's avatar
Loren Merritt committed
484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500
        {
            // 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;
    }
}

501 502
/* 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. */
503
static void mbtree_propagate_cost( int16_t *dst, uint16_t *propagate_in, uint16_t *intra_costs,
504
                                   uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len )
505
{
506
    float fps = *fps_factor;
507
    for( int i = 0; i < len; i++ )
508
    {
509 510 511 512 513 514 515
        int intra_cost = intra_costs[i];
        int inter_cost = X264_MIN(intra_costs[i], inter_costs[i] & LOWRES_COST_MASK);
        float propagate_intra  = intra_cost * inv_qscales[i];
        float propagate_amount = propagate_in[i] + propagate_intra*fps;
        float propagate_num    = intra_cost - inter_cost;
        float propagate_denom  = intra_cost;
        dst[i] = X264_MIN((int)(propagate_amount * propagate_num / propagate_denom + 0.5f), 32767);
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
static void mbtree_propagate_list( x264_t *h, uint16_t *ref_costs, int16_t (*mvs)[2],
                                   int16_t *propagate_amount, uint16_t *lowres_costs,
                                   int bipred_weight, int mb_y, int len, int list )
{
    unsigned stride = h->mb.i_mb_stride;
    unsigned width = h->mb.i_mb_width;
    unsigned height = h->mb.i_mb_height;

    for( unsigned i = 0; i < len; i++ )
    {
        int lists_used = lowres_costs[i]>>LOWRES_COST_SHIFT;

        if( !(lists_used & (1 << list)) )
            continue;

        int listamount = propagate_amount[i];
        /* Apply bipred weighting. */
        if( lists_used == 3 )
            listamount = (listamount * bipred_weight + 32) >> 6;

        /* Early termination for simple case of mv0. */
        if( !M32( mvs[i] ) )
        {
542
            MC_CLIP_ADD( ref_costs[mb_y*stride + i], listamount );
543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564
            continue;
        }

        int x = mvs[i][0];
        int y = mvs[i][1];
        unsigned mbx = (x>>5)+i;
        unsigned mby = (y>>5)+mb_y;
        unsigned idx0 = mbx + mby * stride;
        unsigned idx2 = idx0 + stride;
        x &= 31;
        y &= 31;
        int idx0weight = (32-y)*(32-x);
        int idx1weight = (32-y)*x;
        int idx2weight = y*(32-x);
        int idx3weight = y*x;
        idx0weight = (idx0weight * listamount + 512) >> 10;
        idx1weight = (idx1weight * listamount + 512) >> 10;
        idx2weight = (idx2weight * listamount + 512) >> 10;
        idx3weight = (idx3weight * listamount + 512) >> 10;

        if( mbx < width-1 && mby < height-1 )
        {
565 566 567 568
            MC_CLIP_ADD( ref_costs[idx0+0], idx0weight );
            MC_CLIP_ADD( ref_costs[idx0+1], idx1weight );
            MC_CLIP_ADD( ref_costs[idx2+0], idx2weight );
            MC_CLIP_ADD( ref_costs[idx2+1], idx3weight );
569 570 571 572 573 574 575 576
        }
        else
        {
            /* Note: this takes advantage of unsigned representation to
             * catch negative mbx/mby. */
            if( mby < height )
            {
                if( mbx < width )
577
                    MC_CLIP_ADD( ref_costs[idx0+0], idx0weight );
578
                if( mbx+1 < width )
579
                    MC_CLIP_ADD( ref_costs[idx0+1], idx1weight );
580 581 582 583
            }
            if( mby+1 < height )
            {
                if( mbx < width )
584
                    MC_CLIP_ADD( ref_costs[idx2+0], idx2weight );
585
                if( mbx+1 < width )
586
                    MC_CLIP_ADD( ref_costs[idx2+1], idx3weight );
587 588 589 590 591
            }
        }
    }
}

592
void x264_mc_init( int cpu, x264_mc_functions_t *pf, int cpu_independent )
593
{
594 595
    pf->mc_luma   = mc_luma;
    pf->get_ref   = get_ref;
Henrik Gramner's avatar
Henrik Gramner committed
596

597
    pf->mc_chroma = mc_chroma;
598

599 600 601 602 603
    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;
Henrik Gramner's avatar
Henrik Gramner committed
604
    pf->avg[PIXEL_4x16] = pixel_avg_4x16;
605 606 607
    pf->avg[PIXEL_4x8]  = pixel_avg_4x8;
    pf->avg[PIXEL_4x4]  = pixel_avg_4x4;
    pf->avg[PIXEL_4x2]  = pixel_avg_4x2;
Henrik Gramner's avatar
Henrik Gramner committed
608
    pf->avg[PIXEL_2x8]  = pixel_avg_2x8;
609 610
    pf->avg[PIXEL_2x4]  = pixel_avg_2x4;
    pf->avg[PIXEL_2x2]  = pixel_avg_2x2;
Loren Merritt's avatar
Loren Merritt committed
611

612 613 614 615 616
    pf->weight    = x264_mc_weight_wtab;
    pf->offsetadd = x264_mc_weight_wtab;
    pf->offsetsub = x264_mc_weight_wtab;
    pf->weight_cache = x264_weight_cache;

617
    pf->copy_16x16_unaligned = mc_copy_w16;
Loren Merritt's avatar
Loren Merritt committed
618 619 620 621
    pf->copy[PIXEL_16x16] = mc_copy_w16;
    pf->copy[PIXEL_8x8]   = mc_copy_w8;
    pf->copy[PIXEL_4x4]   = mc_copy_w4;

Henrik Gramner's avatar
Henrik Gramner committed
622 623 624
    pf->store_interleave_chroma       = store_interleave_chroma;
    pf->load_deinterleave_chroma_fenc = load_deinterleave_chroma_fenc;
    pf->load_deinterleave_chroma_fdec = load_deinterleave_chroma_fdec;
625

626
    pf->plane_copy = x264_plane_copy_c;
Xiaolei Yu's avatar
Xiaolei Yu committed
627
    pf->plane_copy_swap = x264_plane_copy_swap_c;
628 629
    pf->plane_copy_interleave = x264_plane_copy_interleave_c;
    pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_c;
xvidfan's avatar
xvidfan committed
630
    pf->plane_copy_deinterleave_rgb = x264_plane_copy_deinterleave_rgb_c;
James Weaver's avatar
James Weaver committed
631
    pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_c;
632

633
    pf->hpel_filter = hpel_filter;
634

635 636
    pf->prefetch_fenc_420 = prefetch_fenc_null;
    pf->prefetch_fenc_422 = prefetch_fenc_null;
637
    pf->prefetch_ref  = prefetch_ref_null;
Fiona Glaser's avatar
Fiona Glaser committed
638
    pf->memcpy_aligned = memcpy;
Fiona Glaser's avatar
Fiona Glaser committed
639
    pf->memzero_aligned = memzero_aligned;
Loren Merritt's avatar
Loren Merritt committed
640
    pf->frame_init_lowres_core = frame_init_lowres_core;
641

Loren Merritt's avatar
Loren Merritt committed
642 643 644 645 646
    pf->integral_init4h = integral_init4h;
    pf->integral_init8h = integral_init8h;
    pf->integral_init4v = integral_init4v;
    pf->integral_init8v = integral_init8v;

647
    pf->mbtree_propagate_cost = mbtree_propagate_cost;
648
    pf->mbtree_propagate_list = mbtree_propagate_list;
649

Steven Walters's avatar
Steven Walters committed
650
#if HAVE_MMX
Loren Merritt's avatar
Loren Merritt committed
651
    x264_mc_init_mmx( cpu, pf );
652
#endif
Steven Walters's avatar
Steven Walters committed
653
#if HAVE_ALTIVEC
654 655 656
    if( cpu&X264_CPU_ALTIVEC )
        x264_mc_altivec_init( pf );
#endif
Steven Walters's avatar
Steven Walters committed
657
#if HAVE_ARMV6
658 659
    x264_mc_init_arm( cpu, pf );
#endif
660 661 662
#if ARCH_AARCH64
    x264_mc_init_aarch64( cpu, pf );
#endif
663 664 665 666
#if HAVE_MSA
    if( cpu&X264_CPU_MSA )
        x264_mc_init_mips( cpu, pf );
#endif
667 668

    if( cpu_independent )
669
    {
670
        pf->mbtree_propagate_cost = mbtree_propagate_cost;
671 672
        pf->mbtree_propagate_list = mbtree_propagate_list;
    }
673 674
}

675
void x264_frame_filter( x264_t *h, x264_frame_t *frame, int mb_y, int b_end )
676
{
677
    const int b_interlaced = PARAM_INTERLACED;
Simon Horlick's avatar
Simon Horlick committed
678
    int start = mb_y*16 - 8; // buffer = 4 for deblock + 3 for 6tap, rounded to 8
679
    int height = (b_end ? frame->i_lines[0] + 16*PARAM_INTERLACED : (mb_y+b_interlaced)*16) + 8;
680

Loren Merritt's avatar
Loren Merritt committed
681 682
    if( mb_y & b_interlaced )
        return;
683

Fiona Glaser's avatar
Fiona Glaser committed
684
    for( int p = 0; p < (CHROMA444 ? 3 : 1); p++ )
Simon Horlick's avatar
Simon Horlick committed
685
    {
Fiona Glaser's avatar
Fiona Glaser committed
686 687 688 689 690
        int stride = frame->i_stride[p];
        const int width = frame->i_width[p];
        int offs = start*stride - 8; // buffer = 3 for 6tap, aligned to 8 for simd

        if( !b_interlaced || h->mb.b_adaptive_mbaff )
Simon Horlick's avatar
Simon Horlick committed
691
            h->mc.hpel_filter(
Fiona Glaser's avatar
Fiona Glaser committed
692 693 694 695 696
                frame->filtered[p][1] + offs,
                frame->filtered[p][2] + offs,
                frame->filtered[p][3] + offs,
                frame->plane[p] + offs,
                stride, width + 16, height - start,
Simon Horlick's avatar
Simon Horlick committed
697
                h->scratch_buffer );
Fiona Glaser's avatar
Fiona Glaser committed
698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715

        if( b_interlaced )
        {
            /* MC must happen between pixels in the same field. */
            stride = frame->i_stride[p] << 1;
            start = (mb_y*16 >> 1) - 8;
            int height_fld = ((b_end ? frame->i_lines[p] : mb_y*16) >> 1) + 8;
            offs = start*stride - 8;
            for( int i = 0; i < 2; i++, offs += frame->i_stride[p] )
            {
                h->mc.hpel_filter(
                    frame->filtered_fld[p][1] + offs,
                    frame->filtered_fld[p][2] + offs,
                    frame->filtered_fld[p][3] + offs,
                    frame->plane_fld[p] + offs,
                    stride, width + 16, height_fld - start,
                    h->scratch_buffer );
            }
Simon Horlick's avatar
Simon Horlick committed
716
        }
717
    }
718 719

    /* generate integral image:
Loren Merritt's avatar
Loren Merritt committed
720 721
     * 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.
722
     * in the lower plane, 4x4 sums (needed only with --partitions p4x4). */
723

724
    if( frame->integral )
725
    {
Fiona Glaser's avatar
Fiona Glaser committed
726
        int stride = frame->i_stride[0];
727 728
        if( start < 0 )
        {
729 730
            memset( frame->integral - PADV * stride - PADH, 0, stride * sizeof(uint16_t) );
            start = -PADV;
731 732
        }
        if( b_end )
733
            height += PADV-9;
734
        for( int y = start; y < height; y++ )
735
        {
736
            pixel    *pix  = frame->plane[0] + y * stride - PADH;
Loren Merritt's avatar
Loren Merritt committed
737 738 739 740 741 742 743 744 745 746 747
            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
748
            {
Loren Merritt's avatar
Loren Merritt committed
749 750 751
                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
752 753
            }
        }
754
    }
755
}