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 *****************************************************************************/
38 static inline int x264_tapfilter( uint8_t *pix, int i_pix_next )
40 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];
42 static inline int x264_tapfilter1( uint8_t *pix )
44 return pix[-2] - 5*pix[-1] + 20*(pix[0] + pix[1]) - 5*pix[ 2] + pix[ 3];
47 static inline void pixel_avg( uint8_t *dst, int i_dst_stride,
48 uint8_t *src1, int i_src1_stride,
49 uint8_t *src2, int i_src2_stride,
50 int i_width, int i_height )
53 for( y = 0; y < i_height; y++ )
55 for( x = 0; x < i_width; x++ )
57 dst[x] = ( src1[x] + src2[x] + 1 ) >> 1;
60 src1 += i_src1_stride;
61 src2 += i_src2_stride;
65 static inline void pixel_avg_wxh( uint8_t *dst, int i_dst, uint8_t *src, int i_src, int width, int height )
68 for( y = 0; y < height; y++ )
70 for( x = 0; x < width; x++ )
72 dst[x] = ( dst[x] + src[x] + 1 ) >> 1;
79 #define PIXEL_AVG_C( name, width, height ) \
80 static void name( uint8_t *pix1, int i_stride_pix1, \
81 uint8_t *pix2, int i_stride_pix2 ) \
83 pixel_avg_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height ); \
85 PIXEL_AVG_C( pixel_avg_16x16, 16, 16 )
86 PIXEL_AVG_C( pixel_avg_16x8, 16, 8 )
87 PIXEL_AVG_C( pixel_avg_8x16, 8, 16 )
88 PIXEL_AVG_C( pixel_avg_8x8, 8, 8 )
89 PIXEL_AVG_C( pixel_avg_8x4, 8, 4 )
90 PIXEL_AVG_C( pixel_avg_4x8, 4, 8 )
91 PIXEL_AVG_C( pixel_avg_4x4, 4, 4 )
92 PIXEL_AVG_C( pixel_avg_4x2, 4, 2 )
93 PIXEL_AVG_C( pixel_avg_2x4, 2, 4 )
94 PIXEL_AVG_C( pixel_avg_2x2, 2, 2 )
97 /* Implicit weighted bipred only:
98 * assumes log2_denom = 5, offset = 0, weight1 + weight2 = 64 */
99 #define op_scale2(x) dst[x] = x264_clip_uint8( (dst[x]*i_weight1 + src[x]*i_weight2 + (1<<5)) >> 6 )
100 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 ){
102 const int i_weight2 = 64 - i_weight1;
103 for(y=0; y<height; y++, dst += i_dst, src += i_src){
106 if(width==2) continue;
109 if(width==4) continue;
114 if(width==8) continue;
126 #define PIXEL_AVG_WEIGHT_C( width, height ) \
127 static void pixel_avg_weight_##width##x##height( \
128 uint8_t *pix1, int i_stride_pix1, \
129 uint8_t *pix2, int i_stride_pix2, int i_weight1 ) \
131 pixel_avg_weight_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height, i_weight1 ); \
134 PIXEL_AVG_WEIGHT_C(16,16)
135 PIXEL_AVG_WEIGHT_C(16,8)
136 PIXEL_AVG_WEIGHT_C(8,16)
137 PIXEL_AVG_WEIGHT_C(8,8)
138 PIXEL_AVG_WEIGHT_C(8,4)
139 PIXEL_AVG_WEIGHT_C(4,8)
140 PIXEL_AVG_WEIGHT_C(4,4)
141 PIXEL_AVG_WEIGHT_C(4,2)
142 PIXEL_AVG_WEIGHT_C(2,4)
143 PIXEL_AVG_WEIGHT_C(2,2)
145 #undef PIXEL_AVG_WEIGHT_C
147 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 );
149 static void mc_copy( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_width, int i_height )
153 for( y = 0; y < i_height; y++ )
155 memcpy( dst, src, i_width );
161 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 )
165 for( y = 0; y < i_height; y++ )
167 for( x = 0; x < i_width; x++ )
169 dst[x] = x264_mc_clip1( ( x264_tapfilter1( &src[x] ) + 16 ) >> 5 );
175 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 )
179 for( y = 0; y < i_height; y++ )
181 for( x = 0; x < i_width; x++ )
183 dst[x] = x264_mc_clip1( ( x264_tapfilter( &src[x], i_src_stride ) + 16 ) >> 5 );
189 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 )
195 for( x = 0; x < i_width; x++ )
202 tap[0] = x264_tapfilter1( &pix[-2*i_src_stride] );
203 tap[1] = x264_tapfilter1( &pix[-1*i_src_stride] );
204 tap[2] = x264_tapfilter1( &pix[ 0*i_src_stride] );
205 tap[3] = x264_tapfilter1( &pix[ 1*i_src_stride] );
206 tap[4] = x264_tapfilter1( &pix[ 2*i_src_stride] );
208 for( y = 0; y < i_height; y++ )
210 tap[5] = x264_tapfilter1( &pix[ 3*i_src_stride] );
212 *out = x264_mc_clip1( ( tap[0] - 5*tap[1] + 20 * tap[2] + 20 * tap[3] -5*tap[4] + tap[5] + 512 ) >> 10 );
226 static void mc_luma( uint8_t *src[4], int i_src_stride,
227 uint8_t *dst, int i_dst_stride,
229 int i_width, int i_height )
231 uint8_t *src1, *src2;
233 int correction = (mvx&1) && (mvy&1) && ((mvx&2) ^ (mvy&2));
235 int hpel1y = (mvy+1-correction)>>1;
236 int filter1 = (hpel1x & 1) + ( (hpel1y & 1) << 1 );
238 src1 = src[filter1] + (hpel1y >> 1) * i_src_stride + (hpel1x >> 1);
240 if ( (mvx|mvy) & 1 ) /* qpel interpolation needed */
242 int hpel2x = (mvx+1)>>1;
243 int hpel2y = (mvy+correction)>>1;
244 int filter2 = (hpel2x & 1) + ( (hpel2y & 1) <<1 );
246 src2 = src[filter2] + (hpel2y >> 1) * i_src_stride + (hpel2x >> 1);
248 pixel_avg( dst, i_dst_stride, src1, i_src_stride,
249 src2, i_src_stride, i_width, i_height );
253 mc_copy( src1, i_src_stride, dst, i_dst_stride, i_width, i_height );
257 static uint8_t *get_ref( uint8_t *src[4], int i_src_stride,
258 uint8_t *dst, int * i_dst_stride,
260 int i_width, int i_height )
262 uint8_t *src1, *src2;
264 int correction = (mvx&1) && (mvy&1) && ((mvx&2) ^ (mvy&2));
266 int hpel1y = (mvy+1-correction)>>1;
267 int filter1 = (hpel1x & 1) + ( (hpel1y & 1) << 1 );
269 src1 = src[filter1] + (hpel1y >> 1) * i_src_stride + (hpel1x >> 1);
271 if ( (mvx|mvy) & 1 ) /* qpel interpolation needed */
273 int hpel2x = (mvx+1)>>1;
274 int hpel2y = (mvy+correction)>>1;
275 int filter2 = (hpel2x & 1) + ( (hpel2y & 1) <<1 );
277 src2 = src[filter2] + (hpel2y >> 1) * i_src_stride + (hpel2x >> 1);
279 pixel_avg( dst, *i_dst_stride, src1, i_src_stride,
280 src2, i_src_stride, i_width, i_height );
286 *i_dst_stride = i_src_stride;
291 /* full chroma mc (ie until 1/8 pixel)*/
292 static void motion_compensation_chroma( uint8_t *src, int i_src_stride,
293 uint8_t *dst, int i_dst_stride,
295 int i_width, int i_height )
300 const int d8x = mvx&0x07;
301 const int d8y = mvy&0x07;
303 const int cA = (8-d8x)*(8-d8y);
304 const int cB = d8x *(8-d8y);
305 const int cC = (8-d8x)*d8y;
306 const int cD = d8x *d8y;
308 src += (mvy >> 3) * i_src_stride + (mvx >> 3);
309 srcp = &src[i_src_stride];
311 for( y = 0; y < i_height; y++ )
313 for( x = 0; x < i_width; x++ )
315 dst[x] = ( cA*src[x] + cB*src[x+1] +
316 cC*srcp[x] + cD*srcp[x+1] + 32 ) >> 6;
321 srcp += i_src_stride;
326 static void motion_compensation_chroma_sse( uint8_t *src, int i_src_stride,
327 uint8_t *dst, int i_dst_stride,
329 int i_width, int i_height )
332 motion_compensation_chroma(src, i_src_stride, dst, i_dst_stride,
333 mvx, mvy, i_width, i_height);
335 const int d8x = mvx&0x07;
336 const int d8y = mvy&0x07;
338 src += (mvy >> 3) * i_src_stride + (mvx >> 3);
340 x264_mc_chroma_sse(src, i_src_stride, dst, i_dst_stride,
341 d8x, d8y, i_height, i_width);
346 void x264_mc_init( int cpu, x264_mc_functions_t *pf )
348 pf->mc_luma = mc_luma;
349 pf->get_ref = get_ref;
350 pf->mc_chroma = motion_compensation_chroma;
352 pf->avg[PIXEL_16x16]= pixel_avg_16x16;
353 pf->avg[PIXEL_16x8] = pixel_avg_16x8;
354 pf->avg[PIXEL_8x16] = pixel_avg_8x16;
355 pf->avg[PIXEL_8x8] = pixel_avg_8x8;
356 pf->avg[PIXEL_8x4] = pixel_avg_8x4;
357 pf->avg[PIXEL_4x8] = pixel_avg_4x8;
358 pf->avg[PIXEL_4x4] = pixel_avg_4x4;
359 pf->avg[PIXEL_4x2] = pixel_avg_4x2;
360 pf->avg[PIXEL_2x4] = pixel_avg_2x4;
361 pf->avg[PIXEL_2x2] = pixel_avg_2x2;
363 pf->avg_weight[PIXEL_16x16]= pixel_avg_weight_16x16;
364 pf->avg_weight[PIXEL_16x8] = pixel_avg_weight_16x8;
365 pf->avg_weight[PIXEL_8x16] = pixel_avg_weight_8x16;
366 pf->avg_weight[PIXEL_8x8] = pixel_avg_weight_8x8;
367 pf->avg_weight[PIXEL_8x4] = pixel_avg_weight_8x4;
368 pf->avg_weight[PIXEL_4x8] = pixel_avg_weight_4x8;
369 pf->avg_weight[PIXEL_4x4] = pixel_avg_weight_4x4;
370 pf->avg_weight[PIXEL_4x2] = pixel_avg_weight_4x2;
371 pf->avg_weight[PIXEL_2x4] = pixel_avg_weight_2x4;
372 pf->avg_weight[PIXEL_2x2] = pixel_avg_weight_2x2;
375 if( cpu&X264_CPU_MMXEXT ) {
376 x264_mc_mmxext_init( pf );
377 pf->mc_chroma = motion_compensation_chroma_sse;
381 if( cpu&X264_CPU_SSE2 )
382 x264_mc_sse2_init( pf );
385 if( cpu&X264_CPU_ALTIVEC )
386 x264_mc_altivec_init( pf );
390 extern void x264_horizontal_filter_mmxext( uint8_t *dst, int i_dst_stride,
391 uint8_t *src, int i_src_stride,
392 int i_width, int i_height );
393 extern void x264_center_filter_mmxext( uint8_t *dst1, int i_dst1_stride,
394 uint8_t *dst2, int i_dst2_stride,
395 uint8_t *src, int i_src_stride,
396 int i_width, int i_height );
398 void x264_frame_filter( int cpu, x264_frame_t *frame )
400 const int x_inc = 16, y_inc = 16;
401 const int stride = frame->i_stride[0];
404 pf_mc_t int_h = mc_hh;
405 pf_mc_t int_v = mc_hv;
406 pf_mc_t int_hv = mc_hc;
409 if ( cpu & X264_CPU_MMXEXT )
411 x264_horizontal_filter_mmxext(frame->filtered[1] - 8 * stride - 8, stride,
412 frame->plane[0] - 8 * stride - 8, stride,
413 stride - 48, frame->i_lines[0] + 16);
414 x264_center_filter_mmxext(frame->filtered[2] - 8 * stride - 8, stride,
415 frame->filtered[3] - 8 * stride - 8, stride,
416 frame->plane[0] - 8 * stride - 8, stride,
417 stride - 48, frame->i_lines[0] + 16);
422 for( y = -8; y < frame->i_lines[0]+8; y += y_inc )
424 uint8_t *p_in = frame->plane[0] + y * stride - 8;
425 uint8_t *p_h = frame->filtered[1] + y * stride - 8;
426 uint8_t *p_v = frame->filtered[2] + y * stride - 8;
427 uint8_t *p_hv = frame->filtered[3] + y * stride - 8;
428 for( x = -8; x < stride - 64 + 8; x += x_inc )
430 int_h( p_in, stride, p_h, stride, x_inc, y_inc );
431 int_v( p_in, stride, p_v, stride, x_inc, y_inc );
432 int_hv( p_in, stride, p_hv, stride, x_inc, y_inc );
442 /* generate integral image:
443 * each entry in frame->integral is the sum of all luma samples above and
444 * to the left of its location (inclusive).
445 * this allows us to calculate the DC of any rectangle by looking only
446 * at the corner entries.
447 * individual entries will overflow 16 bits, but that's ok:
448 * we only need the differences between entries, and those will be correct
449 * as long as we don't try to evaluate a rectangle bigger than 16x16.
450 * likewise, we don't really have to init the edges to 0, leaving garbage
451 * there wouldn't affect the results.*/
453 if( frame->integral )
455 memset( frame->integral - 32 * stride - 32, 0, stride * sizeof(uint16_t) );
456 for( y = -31; y < frame->i_lines[0] + 32; y++ )
458 uint8_t *ref = frame->plane[0] + y * stride - 32;
459 uint16_t *line = frame->integral + y * stride - 32;
460 uint16_t v = line[0] = 0;
461 for( x = 1; x < stride; x++ )
462 line[x] = v += ref[x] + line[x-stride] - line[x-stride-1];
467 void x264_frame_init_lowres( int cpu, x264_frame_t *frame )
470 const int i_stride = frame->i_stride[0];
471 const int i_stride2 = frame->i_stride_lowres;
472 const int i_width2 = i_stride2 - 64;
474 for( y = 0; y < frame->i_lines_lowres - 1; y++ )
476 uint8_t *src0 = &frame->plane[0][2*y*i_stride];
477 uint8_t *src1 = src0+i_stride;
478 uint8_t *src2 = src1+i_stride;
479 uint8_t *dst0 = &frame->lowres[0][y*i_stride2];
480 uint8_t *dsth = &frame->lowres[1][y*i_stride2];
481 uint8_t *dstv = &frame->lowres[2][y*i_stride2];
482 uint8_t *dstc = &frame->lowres[3][y*i_stride2];
483 for( x = 0; x < i_width2 - 1; x++ )
485 dst0[x] = (src0[2*x ] + src0[2*x+1] + src1[2*x ] + src1[2*x+1] + 2) >> 2;
486 dsth[x] = (src0[2*x+1] + src0[2*x+2] + src1[2*x+1] + src1[2*x+2] + 2) >> 2;
487 dstv[x] = (src1[2*x ] + src1[2*x+1] + src2[2*x ] + src2[2*x+1] + 2) >> 2;
488 dstc[x] = (src1[2*x+1] + src1[2*x+2] + src2[2*x+1] + src2[2*x+2] + 2) >> 2;
490 dst0[x] = (src0[2*x ] + src0[2*x+1] + src1[2*x ] + src1[2*x+1] + 2) >> 2;
491 dstv[x] = (src1[2*x ] + src1[2*x+1] + src2[2*x ] + src2[2*x+1] + 2) >> 2;
492 dsth[x] = (src0[2*x+1] + src1[2*x+1] + 1) >> 1;
493 dstc[x] = (src1[2*x+1] + src2[2*x+1] + 1) >> 1;
495 for( i = 0; i < 4; i++ )
496 memcpy( &frame->lowres[i][y*i_stride2], &frame->lowres[i][(y-1)*i_stride2], i_width2 );
498 for( y = 0; y < 16; y++ )
499 for( x = 0; x < 16; x++ )
500 frame->i_cost_est[x][y] = -1;
502 x264_frame_expand_border_lowres( frame );