mc.c 19.6 KB
Newer Older
Laurent Aimar's avatar
Laurent Aimar committed
1
/*****************************************************************************
Fiona Glaser's avatar
Fiona Glaser committed
2
 * mc.c: motion compensation
Laurent Aimar's avatar
Laurent Aimar committed
3
 *****************************************************************************
Fiona Glaser's avatar
Fiona Glaser committed
4
 * Copyright (C) 2003-2010 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.
Fiona Glaser's avatar
Fiona Glaser committed
22 23 24
 *
 * This program is also available under a commercial proprietary license.
 * For more information, contact us at licensing@x264.com.
Laurent Aimar's avatar
Laurent Aimar committed
25 26
 *****************************************************************************/

Loren Merritt's avatar
Loren Merritt committed
27
#include "common.h"
Laurent Aimar's avatar
Laurent Aimar committed
28

Steven Walters's avatar
Steven Walters committed
29
#if HAVE_MMX
30
#include "x86/mc.h"
Laurent Aimar's avatar
Laurent Aimar committed
31
#endif
Steven Walters's avatar
Steven Walters committed
32
#if ARCH_PPC
33
#include "ppc/mc.h"
Laurent Aimar's avatar
Laurent Aimar committed
34
#endif
Steven Walters's avatar
Steven Walters committed
35
#if ARCH_ARM
36 37
#include "arm/mc.h"
#endif
Laurent Aimar's avatar
Laurent Aimar committed
38 39


40 41 42
static inline void pixel_avg( pixel *dst,  int i_dst_stride,
                              pixel *src1, int i_src1_stride,
                              pixel *src2, int i_src2_stride,
Laurent Aimar's avatar
Laurent Aimar committed
43 44
                              int i_width, int i_height )
{
45
    for( int y = 0; y < i_height; y++ )
Laurent Aimar's avatar
Laurent Aimar committed
46
    {
47
        for( int x = 0; x < i_width; x++ )
Laurent Aimar's avatar
Laurent Aimar committed
48 49 50 51 52 53 54
            dst[x] = ( src1[x] + src2[x] + 1 ) >> 1;
        dst  += i_dst_stride;
        src1 += i_src1_stride;
        src2 += i_src2_stride;
    }
}

55
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 )
56
{
57
    for( int y = 0; y < height; y++ )
58
    {
59
        for( int x = 0; x < width; x++ )
60 61 62
            dst[x] = ( src1[x] + src2[x] + 1 ) >> 1;
        src1 += i_src1;
        src2 += i_src2;
63 64 65 66 67 68
        dst += i_dst;
    }
}

/* Implicit weighted bipred only:
 * assumes log2_denom = 5, offset = 0, weight1 + weight2 = 64 */
69
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 )
70
{
71
    const int i_weight2 = 64 - i_weight1;
72
    for( int y = 0; y<height; y++, dst += i_dst, src1 += i_src1, src2 += i_src2 )
73 74
        for( int x = 0; x<width; x++ )
            dst[x] = x264_clip_pixel( (src1[x]*i_weight1 + src2[x]*i_weight2 + (1<<5)) >> 6 );
75
}
Fiona Glaser's avatar
Fiona Glaser committed
76
#undef op_scale2
77

Fiona Glaser's avatar
Fiona Glaser committed
78
#define PIXEL_AVG_C( name, width, height ) \
79 80 81
static void name( pixel *pix1, int i_stride_pix1, \
                  pixel *pix2, int i_stride_pix2, \
                  pixel *pix3, int i_stride_pix3, int weight ) \
82
{ \
Anton Mitrofanov's avatar
Anton Mitrofanov committed
83
    if( weight == 32 ) \
Fiona Glaser's avatar
Fiona Glaser committed
84
        pixel_avg_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, pix3, i_stride_pix3, width, height ); \
Anton Mitrofanov's avatar
Anton Mitrofanov committed
85
    else \
Fiona Glaser's avatar
Fiona Glaser committed
86
        pixel_avg_weight_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, pix3, i_stride_pix3, width, height, weight ); \
87
}
Fiona Glaser's avatar
Fiona Glaser committed
88 89 90 91 92 93 94 95 96 97
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 )
98

Dylan Yudaken's avatar
Dylan Yudaken committed
99 100 101 102
static void x264_weight_cache( x264_t *h, x264_weight_t *w )
{
    w->weightfn = h->mc.weight;
}
103 104 105
#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 )
Dylan Yudaken's avatar
Dylan Yudaken committed
106
{
107 108 109 110
    int offset = weight->i_offset << (BIT_DEPTH-8);
    int scale = weight->i_scale;
    int denom = weight->i_denom;
    if( denom >= 1 )
Dylan Yudaken's avatar
Dylan Yudaken committed
111
    {
112 113
        for( int y = 0; y < i_height; y++, dst += i_dst_stride, src += i_src_stride )
            for( int x = 0; x < i_width; x++ )
Dylan Yudaken's avatar
Dylan Yudaken committed
114 115 116 117
                opscale( x );
    }
    else
    {
118 119
        for( int y = 0; y < i_height; y++, dst += i_dst_stride, src += i_src_stride )
            for( int x = 0; x < i_width; x++ )
Dylan Yudaken's avatar
Dylan Yudaken committed
120 121 122 123
                opscale_noden( x );
    }
}

124
#define MC_WEIGHT_C( name, width ) \
125
    static void name( pixel *dst, int i_dst_stride, pixel *src, int i_src_stride, const x264_weight_t *weight, int height ) \
Dylan Yudaken's avatar
Dylan Yudaken committed
126
{ \
127
    mc_weight( dst, i_dst_stride, src, i_src_stride, weight, width, height );\
Dylan Yudaken's avatar
Dylan Yudaken committed
128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146
}

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}} };
147
static void mc_copy( pixel *src, int i_src_stride, pixel *dst, int i_dst_stride, int i_width, int i_height )
Laurent Aimar's avatar
Laurent Aimar committed
148
{
149
    for( int y = 0; y < i_height; y++ )
Laurent Aimar's avatar
Laurent Aimar committed
150
    {
151
        memcpy( dst, src, i_width * sizeof(pixel) );
Laurent Aimar's avatar
Laurent Aimar committed
152 153 154 155 156 157

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

Loren Merritt's avatar
Loren Merritt committed
158
#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]))
159
static void hpel_filter( pixel *dsth, pixel *dstv, pixel *dstc, pixel *src,
160
                         int stride, int width, int height, int16_t *buf )
Laurent Aimar's avatar
Laurent Aimar committed
161
{
162
    const int pad = (BIT_DEPTH > 9) ? (-10 * PIXEL_MAX) : 0;
163
    for( int y = 0; y < height; y++ )
Laurent Aimar's avatar
Laurent Aimar committed
164
    {
165
        for( int x = -2; x < width+3; x++ )
Laurent Aimar's avatar
Laurent Aimar committed
166
        {
Loren Merritt's avatar
Loren Merritt committed
167
            int v = TAPFILTER(src,stride);
168
            dstv[x] = x264_clip_pixel( (v + 16) >> 5 );
169 170
            /* transform v for storage in a 16-bit integer */
            buf[x+2] = v + pad;
Laurent Aimar's avatar
Laurent Aimar committed
171
        }
172
        for( int x = 0; x < width; x++ )
173
            dstc[x] = x264_clip_pixel( (TAPFILTER(buf+2,1) - 32*pad + 512) >> 10 );
174
        for( int x = 0; x < width; x++ )
175
            dsth[x] = x264_clip_pixel( (TAPFILTER(src,1) + 16) >> 5 );
Loren Merritt's avatar
Loren Merritt committed
176 177 178 179
        dsth += stride;
        dstv += stride;
        dstc += stride;
        src += stride;
Laurent Aimar's avatar
Laurent Aimar committed
180
    }
181 182
}

183 184
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};
Loren Merritt's avatar
Loren Merritt committed
185

186 187
static void mc_luma( pixel *dst,    int i_dst_stride,
                     pixel *src[4], int i_src_stride,
Loren Merritt's avatar
Loren Merritt committed
188
                     int mvx, int mvy,
Dylan Yudaken's avatar
Dylan Yudaken committed
189
                     int i_width, int i_height, const x264_weight_t *weight )
190
{
Loren Merritt's avatar
Loren Merritt committed
191 192
    int qpel_idx = ((mvy&3)<<2) + (mvx&3);
    int offset = (mvy>>2)*i_src_stride + (mvx>>2);
193
    pixel *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
194

Loren Merritt's avatar
Loren Merritt committed
195
    if( qpel_idx & 5 ) /* qpel interpolation needed */
196
    {
197
        pixel *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
198 199
        pixel_avg( dst, i_dst_stride, src1, i_src_stride,
                   src2, i_src_stride, i_width, i_height );
Dylan Yudaken's avatar
Dylan Yudaken committed
200 201
        if( weight->weightfn )
            mc_weight( dst, i_dst_stride, dst, i_dst_stride, weight, i_width, i_height );
202
    }
Dylan Yudaken's avatar
Dylan Yudaken committed
203 204
    else if( weight->weightfn )
        mc_weight( dst, i_dst_stride, src1, i_src_stride, weight, i_width, i_height );
205 206 207 208
    else
        mc_copy( src1, i_src_stride, dst, i_dst_stride, i_width, i_height );
}

209 210 211 212
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 )
213
{
Loren Merritt's avatar
Loren Merritt committed
214 215
    int qpel_idx = ((mvy&3)<<2) + (mvx&3);
    int offset = (mvy>>2)*i_src_stride + (mvx>>2);
216
    pixel *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
217

Loren Merritt's avatar
Loren Merritt committed
218
    if( qpel_idx & 5 ) /* qpel interpolation needed */
219
    {
220
        pixel *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
221 222
        pixel_avg( dst, *i_dst_stride, src1, i_src_stride,
                   src2, i_src_stride, i_width, i_height );
Dylan Yudaken's avatar
Dylan Yudaken committed
223 224 225 226 227 228 229
        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 );
230 231 232 233 234 235 236 237 238
        return dst;
    }
    else
    {
        *i_dst_stride = i_src_stride;
        return src1;
    }
}

Laurent Aimar's avatar
Laurent Aimar committed
239
/* full chroma mc (ie until 1/8 pixel)*/
240
static void mc_chroma( pixel *dstu, pixel *dstv, int i_dst_stride,
241
                       pixel *src, int i_src_stride,
242 243
                       int mvx, int mvy,
                       int i_width, int i_height )
Laurent Aimar's avatar
Laurent Aimar committed
244
{
245
    pixel *srcp;
Laurent Aimar's avatar
Laurent Aimar committed
246

247 248 249 250 251 252
    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;
Laurent Aimar's avatar
Laurent Aimar committed
253

254
    src += (mvy >> 3) * i_src_stride + (mvx >> 3)*2;
Laurent Aimar's avatar
Laurent Aimar committed
255 256
    srcp = &src[i_src_stride];

257
    for( int y = 0; y < i_height; y++ )
Laurent Aimar's avatar
Laurent Aimar committed
258
    {
259
        for( int x = 0; x < i_width; x++ )
260 261 262 263 264 265 266 267
        {
            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;
Laurent Aimar's avatar
Laurent Aimar committed
268 269 270 271 272
        src   = srcp;
        srcp += i_src_stride;
    }
}

Loren Merritt's avatar
Loren Merritt committed
273
#define MC_COPY(W) \
274
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
275 276 277 278 279 280 281
{ \
    mc_copy( src, i_src, dst, i_dst, W, i_height ); \
}
MC_COPY( 16 )
MC_COPY( 8 )
MC_COPY( 4 )

282
void x264_plane_copy_c( pixel *dst, int i_dst,
283
                        pixel *src, int i_src, int w, int h )
284 285 286
{
    while( h-- )
    {
287
        memcpy( dst, src, w * sizeof(pixel) );
288 289 290 291 292
        dst += i_dst;
        src += i_src;
    }
}

293
void x264_plane_copy_interleave_c( pixel *dst, int i_dst,
294 295
                                   pixel *srcu, int i_srcu,
                                   pixel *srcv, int i_srcv, int w, int h )
296 297 298 299
{
    for( int y=0; y<h; y++, dst+=i_dst, srcu+=i_srcu, srcv+=i_srcv )
        for( int x=0; x<w; x++ )
        {
300 301
            dst[2*x]   = srcu[x];
            dst[2*x+1] = srcv[x];
302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336
        }
}

void x264_plane_copy_deinterleave_c( pixel *dstu, int i_dstu,
                                     pixel *dstv, int i_dstv,
                                     pixel *src, int i_src, int w, int h )
{
    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];
        }
}

static void store_interleave_8x8x2( pixel *dst, int i_dst, pixel *srcu, pixel *srcv )
{
    for( int y=0; y<8; y++, dst+=i_dst, srcu+=FDEC_STRIDE, srcv+=FDEC_STRIDE )
        for( int x=0; x<8; x++ )
        {
            dst[2*x]   = srcu[x];
            dst[2*x+1] = srcv[x];
        }
}

static void load_deinterleave_8x8x2_fenc( pixel *dst, pixel *src, int i_src )
{
    x264_plane_copy_deinterleave_c( dst, FENC_STRIDE, dst+FENC_STRIDE/2, FENC_STRIDE, src, i_src, 8, 8 );
}

static void load_deinterleave_8x8x2_fdec( pixel *dst, pixel *src, int i_src )
{
    x264_plane_copy_deinterleave_c( dst, FDEC_STRIDE, dst+FDEC_STRIDE/2, FDEC_STRIDE, src, i_src, 8, 8 );
}

337 338
static void prefetch_fenc_null( pixel *pix_y, int stride_y,
                                pixel *pix_uv, int stride_uv, int mb_x )
339 340
{}

341
static void prefetch_ref_null( pixel *pix, int stride, int parity )
342 343
{}

Fiona Glaser's avatar
Fiona Glaser committed
344 345 346 347 348
static void memzero_aligned( void * dst, int n )
{
    memset( dst, 0, n );
}

349
static void integral_init4h( uint16_t *sum, pixel *pix, int stride )
Loren Merritt's avatar
Loren Merritt committed
350
{
351 352
    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
353 354 355 356 357 358
    {
        sum[x] = v + sum[x-stride];
        v += pix[x+4] - pix[x];
    }
}

359
static void integral_init8h( uint16_t *sum, pixel *pix, int stride )
Loren Merritt's avatar
Loren Merritt committed
360
{
361 362
    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
363 364 365 366 367 368 369 370
    {
        sum[x] = v + sum[x-stride];
        v += pix[x+8] - pix[x];
    }
}

static void integral_init4v( uint16_t *sum8, uint16_t *sum4, int stride )
{
371
    for( int x = 0; x < stride-8; x++ )
Loren Merritt's avatar
Loren Merritt committed
372
        sum4[x] = sum8[x+4*stride] - sum8[x];
373
    for( int x = 0; x < stride-8; x++ )
Loren Merritt's avatar
Loren Merritt committed
374 375 376 377 378
        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 )
{
379
    for( int x = 0; x < stride-8; x++ )
Loren Merritt's avatar
Loren Merritt committed
380 381 382
        sum8[x] = sum8[x+8*stride] - sum8[x];
}

Loren Merritt's avatar
Loren Merritt committed
383 384
void x264_frame_init_lowres( x264_t *h, x264_frame_t *frame )
{
385
    pixel *src = frame->plane[0];
Loren Merritt's avatar
Loren Merritt committed
386 387 388 389 390
    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
391
    for( int y = 0; y < i_height; y++ )
Loren Merritt's avatar
Loren Merritt committed
392
        src[i_width+y*i_stride] = src[i_width-1+y*i_stride];
393
    memcpy( src+i_stride*i_height, src+i_stride*(i_height-1), (i_width+1) * sizeof(pixel) );
Loren Merritt's avatar
Loren Merritt committed
394 395 396 397
    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 );

398 399
    memset( frame->i_cost_est, -1, sizeof(frame->i_cost_est) );

400 401
    for( int y = 0; y < h->param.i_bframe + 2; y++ )
        for( int x = 0; x < h->param.i_bframe + 2; x++ )
402
            frame->i_row_satds[y][x][0] = -1;
403

404 405
    for( int y = 0; y <= !!h->param.i_bframe; y++ )
        for( int x = 0; x <= h->param.i_bframe; x++ )
406
            frame->lowres_mvs[y][x][0][0] = 0x7FFF;
Loren Merritt's avatar
Loren Merritt committed
407 408
}

409
static void frame_init_lowres_core( pixel *src0, pixel *dst0, pixel *dsth, pixel *dstv, pixel *dstc,
Loren Merritt's avatar
Loren Merritt committed
410 411
                                    int src_stride, int dst_stride, int width, int height )
{
412
    for( int y = 0; y < height; y++ )
Loren Merritt's avatar
Loren Merritt committed
413
    {
414 415
        pixel *src1 = src0+src_stride;
        pixel *src2 = src1+src_stride;
416
        for( int x = 0; x<width; x++ )
Loren Merritt's avatar
Loren Merritt committed
417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433
        {
            // 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
434
#if defined(__GNUC__) && (ARCH_X86 || ARCH_X86_64)
435
// gcc isn't smart enough to use the "idiv" instruction
436 437
static ALWAYS_INLINE int32_t div_64_32(int64_t x, int32_t y)
{
438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453
    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 )
{
454
    for( int i = 0; i < len; i++ )
455 456
    {
        int propagate_amount = propagate_in[i] + ((intra_costs[i] * inv_qscales[i] + 128)>>8);
457
        dst[i] = div_64_32((int64_t)propagate_amount * (intra_costs[i] - (inter_costs[i] & LOWRES_COST_MASK)), intra_costs[i]);
458 459 460
    }
}

461
void x264_mc_init( int cpu, x264_mc_functions_t *pf )
Laurent Aimar's avatar
Laurent Aimar committed
462
{
463 464
    pf->mc_luma   = mc_luma;
    pf->get_ref   = get_ref;
465
    pf->mc_chroma = mc_chroma;
Laurent Aimar's avatar
Laurent Aimar committed
466

467 468 469 470 471 472 473 474 475 476
    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
477

Dylan Yudaken's avatar
Dylan Yudaken committed
478 479 480 481 482
    pf->weight    = x264_mc_weight_wtab;
    pf->offsetadd = x264_mc_weight_wtab;
    pf->offsetsub = x264_mc_weight_wtab;
    pf->weight_cache = x264_weight_cache;

483
    pf->copy_16x16_unaligned = mc_copy_w16;
Loren Merritt's avatar
Loren Merritt committed
484 485 486 487
    pf->copy[PIXEL_16x16] = mc_copy_w16;
    pf->copy[PIXEL_8x8]   = mc_copy_w8;
    pf->copy[PIXEL_4x4]   = mc_copy_w4;

488 489 490 491
    pf->store_interleave_8x8x2  = store_interleave_8x8x2;
    pf->load_deinterleave_8x8x2_fenc = load_deinterleave_8x8x2_fenc;
    pf->load_deinterleave_8x8x2_fdec = load_deinterleave_8x8x2_fdec;

492
    pf->plane_copy = x264_plane_copy_c;
493 494 495
    pf->plane_copy_interleave = x264_plane_copy_interleave_c;
    pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_c;

496
    pf->hpel_filter = hpel_filter;
497

498 499
    pf->prefetch_fenc = prefetch_fenc_null;
    pf->prefetch_ref  = prefetch_ref_null;
Fiona Glaser's avatar
Fiona Glaser committed
500
    pf->memcpy_aligned = memcpy;
Fiona Glaser's avatar
Fiona Glaser committed
501
    pf->memzero_aligned = memzero_aligned;
Loren Merritt's avatar
Loren Merritt committed
502
    pf->frame_init_lowres_core = frame_init_lowres_core;
503

Loren Merritt's avatar
Loren Merritt committed
504 505 506 507 508
    pf->integral_init4h = integral_init4h;
    pf->integral_init8h = integral_init8h;
    pf->integral_init4v = integral_init4v;
    pf->integral_init8v = integral_init8v;

509 510
    pf->mbtree_propagate_cost = mbtree_propagate_cost;

Steven Walters's avatar
Steven Walters committed
511
#if HAVE_MMX
Loren Merritt's avatar
Loren Merritt committed
512
    x264_mc_init_mmx( cpu, pf );
513
#endif
Steven Walters's avatar
Steven Walters committed
514
#if HAVE_ALTIVEC
Laurent Aimar's avatar
Laurent Aimar committed
515 516 517
    if( cpu&X264_CPU_ALTIVEC )
        x264_mc_altivec_init( pf );
#endif
Steven Walters's avatar
Steven Walters committed
518
#if HAVE_ARMV6
519 520
    x264_mc_init_arm( cpu, pf );
#endif
Laurent Aimar's avatar
Laurent Aimar committed
521 522
}

523
void x264_frame_filter( x264_t *h, x264_frame_t *frame, int mb_y, int b_end )
524
{
525
    const int b_interlaced = h->sh.b_mbaff;
526
    const int stride = frame->i_stride[0] << b_interlaced;
527
    const int width = frame->i_width[0];
528
    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
529
    int height = ((b_end ? frame->i_lines[0] : mb_y*16) >> b_interlaced) + 8;
530
    int offs = start*stride - 8; // buffer = 3 for 6tap, aligned to 8 for simd
531

Loren Merritt's avatar
Loren Merritt committed
532 533
    if( mb_y & b_interlaced )
        return;
534

535
    for( int y = 0; y <= b_interlaced; y++, offs += frame->i_stride[0] )
536
    {
537
        h->mc.hpel_filter(
538 539 540 541
            frame->filtered[1] + offs,
            frame->filtered[2] + offs,
            frame->filtered[3] + offs,
            frame->plane[0] + offs,
542 543
            stride, width + 16, height - start,
            h->scratch_buffer );
544
    }
545 546

    /* generate integral image:
Loren Merritt's avatar
Loren Merritt committed
547 548
     * 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
549
     * in the lower plane, 4x4 sums (needed only with --partitions p4x4). */
550

Loren Merritt's avatar
Loren Merritt committed
551
    if( frame->integral )
552
    {
Loren Merritt's avatar
Loren Merritt committed
553 554
        if( start < 0 )
        {
555 556
            memset( frame->integral - PADV * stride - PADH, 0, stride * sizeof(uint16_t) );
            start = -PADV;
Loren Merritt's avatar
Loren Merritt committed
557 558
        }
        if( b_end )
559
            height += PADV-9;
560
        for( int y = start; y < height; y++ )
561
        {
562
            pixel    *pix  = frame->plane[0] + y * stride - PADH;
Loren Merritt's avatar
Loren Merritt committed
563 564 565 566 567 568 569 570 571 572 573
            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
574
            {
Loren Merritt's avatar
Loren Merritt committed
575 576 577
                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
578 579
            }
        }
580
    }
581
}