1
/*****************************************************************************
2
* mc.c: h264 encoder library (Motion Compensation)
3
*****************************************************************************
4
* Copyright (C) 2003 Laurent Aimar
5
* $Id: mc.c,v 1.1 2004/06/03 19:27:07 fenrir Exp $
7
* Authors: Laurent Aimar <fenrir@via.ecp.fr>
9
* This program is free software; you can redistribute it and/or modify
10
* it under the terms of the GNU General Public License as published by
11
* the Free Software Foundation; either version 2 of the License, or
12
* (at your option) any later version.
14
* This program is distributed in the hope that it will be useful,
15
* but WITHOUT ANY WARRANTY; without even the implied warranty of
16
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17
* GNU General Public License for more details.
19
* You should have received a copy of the GNU General Public License
20
* along with this program; if not, write to the Free Software
21
* Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111, USA.
22
*****************************************************************************/
35
static inline int x264_tapfilter( uint8_t *pix, int i_pix_next )
37
return pix[-2*i_pix_next] - 5*pix[-1*i_pix_next] + 20*(pix[0] + pix[1*i_pix_next]) - 5*pix[ 2*i_pix_next] + pix[ 3*i_pix_next];
39
static inline int x264_tapfilter1( uint8_t *pix )
41
return pix[-2] - 5*pix[-1] + 20*(pix[0] + pix[1]) - 5*pix[ 2] + pix[ 3];
44
static inline void pixel_avg( uint8_t *dst, int i_dst_stride,
45
uint8_t *src1, int i_src1_stride,
46
uint8_t *src2, int i_src2_stride,
47
int i_width, int i_height )
50
for( y = 0; y < i_height; y++ )
52
for( x = 0; x < i_width; x++ )
54
dst[x] = ( src1[x] + src2[x] + 1 ) >> 1;
57
src1 += i_src1_stride;
58
src2 += i_src2_stride;
62
static inline void pixel_avg_wxh( uint8_t *dst, int i_dst, uint8_t *src, int i_src, int width, int height )
65
for( y = 0; y < height; y++ )
67
for( x = 0; x < width; x++ )
69
dst[x] = ( dst[x] + src[x] + 1 ) >> 1;
76
#define PIXEL_AVG_C( name, width, height ) \
77
static void name( uint8_t *pix1, int i_stride_pix1, \
78
uint8_t *pix2, int i_stride_pix2 ) \
80
pixel_avg_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height ); \
82
PIXEL_AVG_C( pixel_avg_16x16, 16, 16 )
83
PIXEL_AVG_C( pixel_avg_16x8, 16, 8 )
84
PIXEL_AVG_C( pixel_avg_8x16, 8, 16 )
85
PIXEL_AVG_C( pixel_avg_8x8, 8, 8 )
86
PIXEL_AVG_C( pixel_avg_8x4, 8, 4 )
87
PIXEL_AVG_C( pixel_avg_4x8, 4, 8 )
88
PIXEL_AVG_C( pixel_avg_4x4, 4, 4 )
89
PIXEL_AVG_C( pixel_avg_4x2, 4, 2 )
90
PIXEL_AVG_C( pixel_avg_2x4, 2, 4 )
91
PIXEL_AVG_C( pixel_avg_2x2, 2, 2 )
94
/* Implicit weighted bipred only:
95
* assumes log2_denom = 5, offset = 0, weight1 + weight2 = 64 */
96
#define op_scale2(x) dst[x] = x264_clip_uint8( (dst[x]*i_weight1 + src[x]*i_weight2 + (1<<5)) >> 6 )
97
static inline void pixel_avg_weight_wxh( uint8_t *dst, int i_dst, uint8_t *src, int i_src, int width, int height, int i_weight1 ){
99
const int i_weight2 = 64 - i_weight1;
100
for(y=0; y<height; y++, dst += i_dst, src += i_src){
103
if(width==2) continue;
106
if(width==4) continue;
111
if(width==8) continue;
123
#define PIXEL_AVG_WEIGHT_C( width, height ) \
124
static void pixel_avg_weight_##width##x##height( \
125
uint8_t *pix1, int i_stride_pix1, \
126
uint8_t *pix2, int i_stride_pix2, int i_weight1 ) \
128
pixel_avg_weight_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height, i_weight1 ); \
131
PIXEL_AVG_WEIGHT_C(16,16)
132
PIXEL_AVG_WEIGHT_C(16,8)
133
PIXEL_AVG_WEIGHT_C(8,16)
134
PIXEL_AVG_WEIGHT_C(8,8)
135
PIXEL_AVG_WEIGHT_C(8,4)
136
PIXEL_AVG_WEIGHT_C(4,8)
137
PIXEL_AVG_WEIGHT_C(4,4)
138
PIXEL_AVG_WEIGHT_C(4,2)
139
PIXEL_AVG_WEIGHT_C(2,4)
140
PIXEL_AVG_WEIGHT_C(2,2)
142
#undef PIXEL_AVG_WEIGHT_C
144
typedef void (*pf_mc_t)(uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_width, int i_height );
146
static void mc_copy( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_width, int i_height )
150
for( y = 0; y < i_height; y++ )
152
memcpy( dst, src, i_width );
158
static inline void mc_hh( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_width, int i_height )
162
for( y = 0; y < i_height; y++ )
164
for( x = 0; x < i_width; x++ )
166
dst[x] = x264_mc_clip1( ( x264_tapfilter1( &src[x] ) + 16 ) >> 5 );
172
static inline void mc_hv( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_width, int i_height )
176
for( y = 0; y < i_height; y++ )
178
for( x = 0; x < i_width; x++ )
180
dst[x] = x264_mc_clip1( ( x264_tapfilter( &src[x], i_src_stride ) + 16 ) >> 5 );
186
static inline void mc_hc( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_width, int i_height )
192
for( x = 0; x < i_width; x++ )
199
tap[0] = x264_tapfilter1( &pix[-2*i_src_stride] );
200
tap[1] = x264_tapfilter1( &pix[-1*i_src_stride] );
201
tap[2] = x264_tapfilter1( &pix[ 0*i_src_stride] );
202
tap[3] = x264_tapfilter1( &pix[ 1*i_src_stride] );
203
tap[4] = x264_tapfilter1( &pix[ 2*i_src_stride] );
205
for( y = 0; y < i_height; y++ )
207
tap[5] = x264_tapfilter1( &pix[ 3*i_src_stride] );
209
*out = x264_mc_clip1( ( tap[0] - 5*tap[1] + 20 * tap[2] + 20 * tap[3] -5*tap[4] + tap[5] + 512 ) >> 10 );
223
static void hpel_filter( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src,
224
int stride, int width, int height )
226
mc_hh( src, stride, dsth, stride, width, height );
227
mc_hv( src, stride, dstv, stride, width, height );
228
mc_hc( src, stride, dstc, stride, width, height );
231
static const int hpel_ref0[16] = {0,1,1,1,0,1,1,1,2,3,3,3,0,1,1,1};
232
static const int hpel_ref1[16] = {0,0,0,0,2,2,3,2,2,2,3,2,2,2,3,2};
234
static void mc_luma( uint8_t *dst, int i_dst_stride,
235
uint8_t *src[4], int i_src_stride,
237
int i_width, int i_height )
239
int qpel_idx = ((mvy&3)<<2) + (mvx&3);
240
int offset = (mvy>>2)*i_src_stride + (mvx>>2);
241
uint8_t *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
243
if( qpel_idx & 5 ) /* qpel interpolation needed */
245
uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
246
pixel_avg( dst, i_dst_stride, src1, i_src_stride,
247
src2, i_src_stride, i_width, i_height );
251
mc_copy( src1, i_src_stride, dst, i_dst_stride, i_width, i_height );
255
static uint8_t *get_ref( uint8_t *dst, int *i_dst_stride,
256
uint8_t *src[4], int i_src_stride,
258
int i_width, int i_height )
260
int qpel_idx = ((mvy&3)<<2) + (mvx&3);
261
int offset = (mvy>>2)*i_src_stride + (mvx>>2);
262
uint8_t *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
264
if( qpel_idx & 5 ) /* qpel interpolation needed */
266
uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
267
pixel_avg( dst, *i_dst_stride, src1, i_src_stride,
268
src2, i_src_stride, i_width, i_height );
273
*i_dst_stride = i_src_stride;
278
/* full chroma mc (ie until 1/8 pixel)*/
279
static void mc_chroma( uint8_t *dst, int i_dst_stride,
280
uint8_t *src, int i_src_stride,
282
int i_width, int i_height )
287
const int d8x = mvx&0x07;
288
const int d8y = mvy&0x07;
290
const int cA = (8-d8x)*(8-d8y);
291
const int cB = d8x *(8-d8y);
292
const int cC = (8-d8x)*d8y;
293
const int cD = d8x *d8y;
295
src += (mvy >> 3) * i_src_stride + (mvx >> 3);
296
srcp = &src[i_src_stride];
298
for( y = 0; y < i_height; y++ )
300
for( x = 0; x < i_width; x++ )
302
dst[x] = ( cA*src[x] + cB*src[x+1] +
303
cC*srcp[x] + cD*srcp[x+1] + 32 ) >> 6;
308
srcp += i_src_stride;
313
static void mc_copy_w##W( uint8_t *dst, int i_dst, uint8_t *src, int i_src, int i_height ) \
315
mc_copy( src, i_src, dst, i_dst, W, i_height ); \
321
static void plane_copy( uint8_t *dst, int i_dst,
322
uint8_t *src, int i_src, int w, int h)
326
memcpy( dst, src, w );
332
void prefetch_fenc_null( uint8_t *pix_y, int stride_y,
333
uint8_t *pix_uv, int stride_uv, int mb_x )
336
void prefetch_ref_null( uint8_t *pix, int stride, int parity )
339
void x264_mc_init( int cpu, x264_mc_functions_t *pf )
341
pf->mc_luma = mc_luma;
342
pf->get_ref = get_ref;
343
pf->mc_chroma = mc_chroma;
345
pf->avg[PIXEL_16x16]= pixel_avg_16x16;
346
pf->avg[PIXEL_16x8] = pixel_avg_16x8;
347
pf->avg[PIXEL_8x16] = pixel_avg_8x16;
348
pf->avg[PIXEL_8x8] = pixel_avg_8x8;
349
pf->avg[PIXEL_8x4] = pixel_avg_8x4;
350
pf->avg[PIXEL_4x8] = pixel_avg_4x8;
351
pf->avg[PIXEL_4x4] = pixel_avg_4x4;
352
pf->avg[PIXEL_4x2] = pixel_avg_4x2;
353
pf->avg[PIXEL_2x4] = pixel_avg_2x4;
354
pf->avg[PIXEL_2x2] = pixel_avg_2x2;
356
pf->avg_weight[PIXEL_16x16]= pixel_avg_weight_16x16;
357
pf->avg_weight[PIXEL_16x8] = pixel_avg_weight_16x8;
358
pf->avg_weight[PIXEL_8x16] = pixel_avg_weight_8x16;
359
pf->avg_weight[PIXEL_8x8] = pixel_avg_weight_8x8;
360
pf->avg_weight[PIXEL_8x4] = pixel_avg_weight_8x4;
361
pf->avg_weight[PIXEL_4x8] = pixel_avg_weight_4x8;
362
pf->avg_weight[PIXEL_4x4] = pixel_avg_weight_4x4;
363
pf->avg_weight[PIXEL_4x2] = pixel_avg_weight_4x2;
364
pf->avg_weight[PIXEL_2x4] = pixel_avg_weight_2x4;
365
pf->avg_weight[PIXEL_2x2] = pixel_avg_weight_2x2;
367
pf->copy[PIXEL_16x16] = mc_copy_w16;
368
pf->copy[PIXEL_8x8] = mc_copy_w8;
369
pf->copy[PIXEL_4x4] = mc_copy_w4;
371
pf->plane_copy = plane_copy;
372
pf->hpel_filter = hpel_filter;
374
pf->prefetch_fenc = prefetch_fenc_null;
375
pf->prefetch_ref = prefetch_ref_null;
378
if( cpu&X264_CPU_MMXEXT ) {
379
x264_mc_mmxext_init( pf );
380
pf->mc_chroma = x264_mc_chroma_mmxext;
382
if( cpu&X264_CPU_SSE2 )
383
x264_mc_sse2_init( pf );
386
if( cpu&X264_CPU_ALTIVEC )
387
x264_mc_altivec_init( pf );
391
void x264_frame_filter( x264_t *h, x264_frame_t *frame, int mb_y, int b_end )
393
const int b_interlaced = h->sh.b_mbaff;
394
const int stride = frame->i_stride[0] << b_interlaced;
395
const int width = frame->i_width[0];
396
int start = (mb_y*16 >> b_interlaced) - 8;
397
int height = ((b_end ? frame->i_lines[0] : mb_y*16) >> b_interlaced) + 8;
398
int offs = start*stride - 8; // buffer = 4 for deblock + 3 for 6tap, rounded to 8
401
if( mb_y & b_interlaced )
404
for( y=0; y<=b_interlaced; y++, offs+=frame->i_stride[0] )
407
frame->filtered[1] + offs,
408
frame->filtered[2] + offs,
409
frame->filtered[3] + offs,
410
frame->plane[0] + offs,
411
stride, width + 16, height - start );
414
/* generate integral image:
415
* frame->integral contains 2 planes. in the upper plane, each element is
416
* the sum of an 8x8 pixel region with top-left corner on that point.
417
* in the lower plane, 4x4 sums (needed only with --partitions p4x4). */
419
if( frame->integral )
423
memset( frame->integral - PADV * stride - PADH, 0, stride * sizeof(uint16_t) );
428
for( y = start; y < height; y++ )
430
uint8_t *ref = frame->plane[0] + y * stride - PADH;
431
uint16_t *line = frame->integral + (y+1) * stride - PADH + 1;
432
uint16_t v = line[0] = 0;
433
for( x = 1; x < stride-1; x++ )
434
line[x] = v += ref[x] + line[x-stride] - line[x-stride-1];
438
uint16_t *sum4 = line + stride * (frame->i_lines[0] + PADV*2);
439
for( x = 1; x < stride-8; x++, line++, sum4++ )
441
sum4[0] = line[4+4*stride] - line[4] - line[4*stride] + line[0];
442
line[0] += line[8+8*stride] - line[8] - line[8*stride];
449
void x264_frame_init_lowres( x264_t *h, x264_frame_t *frame )
452
const int i_stride = frame->i_stride[0];
453
const int i_stride2 = frame->i_stride_lowres;
454
const int i_width2 = frame->i_width_lowres;
456
for( y = 0; y < frame->i_lines_lowres - 1; y++ )
458
uint8_t *src0 = &frame->plane[0][2*y*i_stride];
459
uint8_t *src1 = src0+i_stride;
460
uint8_t *src2 = src1+i_stride;
461
uint8_t *dst0 = &frame->lowres[0][y*i_stride2];
462
uint8_t *dsth = &frame->lowres[1][y*i_stride2];
463
uint8_t *dstv = &frame->lowres[2][y*i_stride2];
464
uint8_t *dstc = &frame->lowres[3][y*i_stride2];
465
for( x = 0; x < i_width2 - 1; x++ )
467
dst0[x] = (src0[2*x ] + src0[2*x+1] + src1[2*x ] + src1[2*x+1] + 2) >> 2;
468
dsth[x] = (src0[2*x+1] + src0[2*x+2] + src1[2*x+1] + src1[2*x+2] + 2) >> 2;
469
dstv[x] = (src1[2*x ] + src1[2*x+1] + src2[2*x ] + src2[2*x+1] + 2) >> 2;
470
dstc[x] = (src1[2*x+1] + src1[2*x+2] + src2[2*x+1] + src2[2*x+2] + 2) >> 2;
472
dst0[x] = (src0[2*x ] + src0[2*x+1] + src1[2*x ] + src1[2*x+1] + 2) >> 2;
473
dstv[x] = (src1[2*x ] + src1[2*x+1] + src2[2*x ] + src2[2*x+1] + 2) >> 2;
474
dsth[x] = (src0[2*x+1] + src1[2*x+1] + 1) >> 1;
475
dstc[x] = (src1[2*x+1] + src2[2*x+1] + 1) >> 1;
477
for( i = 0; i < 4; i++ )
478
memcpy( &frame->lowres[i][y*i_stride2], &frame->lowres[i][(y-1)*i_stride2], i_width2 );
480
for( y = 0; y < 16; y++ )
481
for( x = 0; x < 16; x++ )
482
frame->i_cost_est[x][y] = -1;
484
x264_frame_expand_border_lowres( frame );