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 *****************************************************************************/
34 static inline int x264_tapfilter( uint8_t *pix, int i_pix_next )
36 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];
38 static inline int x264_tapfilter1( uint8_t *pix )
40 return pix[-2] - 5*pix[-1] + 20*(pix[0] + pix[1]) - 5*pix[ 2] + pix[ 3];
43 static inline void pixel_avg( uint8_t *dst, int i_dst_stride,
44 uint8_t *src1, int i_src1_stride,
45 uint8_t *src2, int i_src2_stride,
46 int i_width, int i_height )
49 for( y = 0; y < i_height; y++ )
51 for( x = 0; x < i_width; x++ )
53 dst[x] = ( src1[x] + src2[x] + 1 ) >> 1;
56 src1 += i_src1_stride;
57 src2 += i_src2_stride;
61 static inline void pixel_avg_wxh( uint8_t *dst, int i_dst, uint8_t *src, int i_src, int width, int height )
64 for( y = 0; y < height; y++ )
66 for( x = 0; x < width; x++ )
68 dst[x] = ( dst[x] + src[x] + 1 ) >> 1;
75 #define PIXEL_AVG_C( name, width, height ) \
76 static void name( uint8_t *pix1, int i_stride_pix1, \
77 uint8_t *pix2, int i_stride_pix2 ) \
79 pixel_avg_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height ); \
81 PIXEL_AVG_C( pixel_avg_16x16, 16, 16 )
82 PIXEL_AVG_C( pixel_avg_16x8, 16, 8 )
83 PIXEL_AVG_C( pixel_avg_8x16, 8, 16 )
84 PIXEL_AVG_C( pixel_avg_8x8, 8, 8 )
85 PIXEL_AVG_C( pixel_avg_8x4, 8, 4 )
86 PIXEL_AVG_C( pixel_avg_4x8, 4, 8 )
87 PIXEL_AVG_C( pixel_avg_4x4, 4, 4 )
88 PIXEL_AVG_C( pixel_avg_4x2, 4, 2 )
89 PIXEL_AVG_C( pixel_avg_2x4, 2, 4 )
90 PIXEL_AVG_C( pixel_avg_2x2, 2, 2 )
93 /* Implicit weighted bipred only:
94 * assumes log2_denom = 5, offset = 0, weight1 + weight2 = 64 */
95 #define op_scale2(x) dst[x] = x264_clip_uint8( (dst[x]*i_weight1 + src[x]*i_weight2 + (1<<5)) >> 6 )
96 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 ){
98 const int i_weight2 = 64 - i_weight1;
99 for(y=0; y<height; y++, dst += i_dst, src += i_src){
102 if(width==2) continue;
105 if(width==4) continue;
110 if(width==8) continue;
122 #define PIXEL_AVG_WEIGHT_C( width, height ) \
123 static void pixel_avg_weight_##width##x##height( \
124 uint8_t *pix1, int i_stride_pix1, \
125 uint8_t *pix2, int i_stride_pix2, int i_weight1 ) \
127 pixel_avg_weight_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height, i_weight1 ); \
130 PIXEL_AVG_WEIGHT_C(16,16)
131 PIXEL_AVG_WEIGHT_C(16,8)
132 PIXEL_AVG_WEIGHT_C(8,16)
133 PIXEL_AVG_WEIGHT_C(8,8)
134 PIXEL_AVG_WEIGHT_C(8,4)
135 PIXEL_AVG_WEIGHT_C(4,8)
136 PIXEL_AVG_WEIGHT_C(4,4)
137 PIXEL_AVG_WEIGHT_C(4,2)
138 PIXEL_AVG_WEIGHT_C(2,4)
139 PIXEL_AVG_WEIGHT_C(2,2)
141 #undef PIXEL_AVG_WEIGHT_C
143 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 );
145 static void mc_copy( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_width, int i_height )
149 for( y = 0; y < i_height; y++ )
151 memcpy( dst, src, i_width );
157 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 )
161 for( y = 0; y < i_height; y++ )
163 for( x = 0; x < i_width; x++ )
165 dst[x] = x264_clip_uint8( ( x264_tapfilter1( &src[x] ) + 16 ) >> 5 );
171 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 )
175 for( y = 0; y < i_height; y++ )
177 for( x = 0; x < i_width; x++ )
179 dst[x] = x264_clip_uint8( ( x264_tapfilter( &src[x], i_src_stride ) + 16 ) >> 5 );
185 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 )
191 for( x = 0; x < i_width; x++ )
198 tap[0] = x264_tapfilter1( &pix[-2*i_src_stride] );
199 tap[1] = x264_tapfilter1( &pix[-1*i_src_stride] );
200 tap[2] = x264_tapfilter1( &pix[ 0*i_src_stride] );
201 tap[3] = x264_tapfilter1( &pix[ 1*i_src_stride] );
202 tap[4] = x264_tapfilter1( &pix[ 2*i_src_stride] );
204 for( y = 0; y < i_height; y++ )
206 tap[5] = x264_tapfilter1( &pix[ 3*i_src_stride] );
208 *out = x264_clip_uint8( ( tap[0] - 5*tap[1] + 20 * tap[2] + 20 * tap[3] -5*tap[4] + tap[5] + 512 ) >> 10 );
222 static void hpel_filter( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src,
223 int stride, int width, int height )
225 mc_hh( src, stride, dsth, stride, width, height );
226 mc_hv( src, stride, dstv, stride, width, height );
227 mc_hc( src, stride, dstc, stride, width, height );
230 static const int hpel_ref0[16] = {0,1,1,1,0,1,1,1,2,3,3,3,0,1,1,1};
231 static const int hpel_ref1[16] = {0,0,0,0,2,2,3,2,2,2,3,2,2,2,3,2};
233 static void mc_luma( uint8_t *dst, int i_dst_stride,
234 uint8_t *src[4], int i_src_stride,
236 int i_width, int i_height )
238 int qpel_idx = ((mvy&3)<<2) + (mvx&3);
239 int offset = (mvy>>2)*i_src_stride + (mvx>>2);
240 uint8_t *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
242 if( qpel_idx & 5 ) /* qpel interpolation needed */
244 uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
245 pixel_avg( dst, i_dst_stride, src1, i_src_stride,
246 src2, i_src_stride, i_width, i_height );
250 mc_copy( src1, i_src_stride, dst, i_dst_stride, i_width, i_height );
254 static uint8_t *get_ref( uint8_t *dst, int *i_dst_stride,
255 uint8_t *src[4], int i_src_stride,
257 int i_width, int i_height )
259 int qpel_idx = ((mvy&3)<<2) + (mvx&3);
260 int offset = (mvy>>2)*i_src_stride + (mvx>>2);
261 uint8_t *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
263 if( qpel_idx & 5 ) /* qpel interpolation needed */
265 uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
266 pixel_avg( dst, *i_dst_stride, src1, i_src_stride,
267 src2, i_src_stride, i_width, i_height );
272 *i_dst_stride = i_src_stride;
277 /* full chroma mc (ie until 1/8 pixel)*/
278 static void mc_chroma( uint8_t *dst, int i_dst_stride,
279 uint8_t *src, int i_src_stride,
281 int i_width, int i_height )
286 const int d8x = mvx&0x07;
287 const int d8y = mvy&0x07;
289 const int cA = (8-d8x)*(8-d8y);
290 const int cB = d8x *(8-d8y);
291 const int cC = (8-d8x)*d8y;
292 const int cD = d8x *d8y;
294 src += (mvy >> 3) * i_src_stride + (mvx >> 3);
295 srcp = &src[i_src_stride];
297 for( y = 0; y < i_height; y++ )
299 for( x = 0; x < i_width; x++ )
301 dst[x] = ( cA*src[x] + cB*src[x+1] +
302 cC*srcp[x] + cD*srcp[x+1] + 32 ) >> 6;
307 srcp += i_src_stride;
312 static void mc_copy_w##W( uint8_t *dst, int i_dst, uint8_t *src, int i_src, int i_height ) \
314 mc_copy( src, i_src, dst, i_dst, W, i_height ); \
320 static void plane_copy( uint8_t *dst, int i_dst,
321 uint8_t *src, int i_src, int w, int h)
325 memcpy( dst, src, w );
331 void prefetch_fenc_null( uint8_t *pix_y, int stride_y,
332 uint8_t *pix_uv, int stride_uv, int mb_x )
335 void prefetch_ref_null( uint8_t *pix, int stride, int parity )
338 void x264_mc_init( int cpu, x264_mc_functions_t *pf )
340 pf->mc_luma = mc_luma;
341 pf->get_ref = get_ref;
342 pf->mc_chroma = mc_chroma;
344 pf->avg[PIXEL_16x16]= pixel_avg_16x16;
345 pf->avg[PIXEL_16x8] = pixel_avg_16x8;
346 pf->avg[PIXEL_8x16] = pixel_avg_8x16;
347 pf->avg[PIXEL_8x8] = pixel_avg_8x8;
348 pf->avg[PIXEL_8x4] = pixel_avg_8x4;
349 pf->avg[PIXEL_4x8] = pixel_avg_4x8;
350 pf->avg[PIXEL_4x4] = pixel_avg_4x4;
351 pf->avg[PIXEL_4x2] = pixel_avg_4x2;
352 pf->avg[PIXEL_2x4] = pixel_avg_2x4;
353 pf->avg[PIXEL_2x2] = pixel_avg_2x2;
355 pf->avg_weight[PIXEL_16x16]= pixel_avg_weight_16x16;
356 pf->avg_weight[PIXEL_16x8] = pixel_avg_weight_16x8;
357 pf->avg_weight[PIXEL_8x16] = pixel_avg_weight_8x16;
358 pf->avg_weight[PIXEL_8x8] = pixel_avg_weight_8x8;
359 pf->avg_weight[PIXEL_8x4] = pixel_avg_weight_8x4;
360 pf->avg_weight[PIXEL_4x8] = pixel_avg_weight_4x8;
361 pf->avg_weight[PIXEL_4x4] = pixel_avg_weight_4x4;
362 pf->avg_weight[PIXEL_4x2] = pixel_avg_weight_4x2;
363 pf->avg_weight[PIXEL_2x4] = pixel_avg_weight_2x4;
364 pf->avg_weight[PIXEL_2x2] = pixel_avg_weight_2x2;
366 pf->copy[PIXEL_16x16] = mc_copy_w16;
367 pf->copy[PIXEL_8x8] = mc_copy_w8;
368 pf->copy[PIXEL_4x4] = mc_copy_w4;
370 pf->plane_copy = plane_copy;
371 pf->hpel_filter = hpel_filter;
373 pf->prefetch_fenc = prefetch_fenc_null;
374 pf->prefetch_ref = prefetch_ref_null;
375 pf->memcpy_aligned = memcpy;
378 x264_mc_init_mmx( cpu, pf );
381 if( cpu&X264_CPU_ALTIVEC )
382 x264_mc_altivec_init( pf );
386 void x264_frame_filter( x264_t *h, x264_frame_t *frame, int mb_y, int b_end )
388 const int b_interlaced = h->sh.b_mbaff;
389 const int stride = frame->i_stride[0] << b_interlaced;
390 const int width = frame->i_width[0];
391 int start = (mb_y*16 >> b_interlaced) - 8; // buffer = 4 for deblock + 3 for 6tap, rounded to 8
392 int height = ((b_end ? frame->i_lines[0] : mb_y*16) >> b_interlaced) + 8;
393 int offs = start*stride - 8; // buffer = 3 for 6tap, aligned to 8 for simd
396 if( mb_y & b_interlaced )
399 for( y=0; y<=b_interlaced; y++, offs+=frame->i_stride[0] )
402 frame->filtered[1] + offs,
403 frame->filtered[2] + offs,
404 frame->filtered[3] + offs,
405 frame->plane[0] + offs,
406 stride, width + 16, height - start );
409 /* generate integral image:
410 * frame->integral contains 2 planes. in the upper plane, each element is
411 * the sum of an 8x8 pixel region with top-left corner on that point.
412 * in the lower plane, 4x4 sums (needed only with --partitions p4x4). */
414 if( frame->integral )
418 memset( frame->integral - PADV * stride - PADH, 0, stride * sizeof(uint16_t) );
423 for( y = start; y < height; y++ )
425 uint8_t *ref = frame->plane[0] + y * stride - PADH;
426 uint16_t *line = frame->integral + (y+1) * stride - PADH + 1;
427 uint16_t v = line[0] = 0;
428 for( x = 1; x < stride-1; x++ )
429 line[x] = v += ref[x] + line[x-stride] - line[x-stride-1];
433 uint16_t *sum4 = line + stride * (frame->i_lines[0] + PADV*2);
434 for( x = 1; x < stride-8; x++, line++, sum4++ )
436 sum4[0] = line[4+4*stride] - line[4] - line[4*stride] + line[0];
437 line[0] += line[8+8*stride] - line[8] - line[8*stride];
444 void x264_frame_init_lowres( x264_t *h, x264_frame_t *frame )
447 const int i_stride = frame->i_stride[0];
448 const int i_stride2 = frame->i_stride_lowres;
449 const int i_width2 = frame->i_width_lowres;
451 for( y = 0; y < frame->i_lines_lowres - 1; y++ )
453 uint8_t *src0 = &frame->plane[0][2*y*i_stride];
454 uint8_t *src1 = src0+i_stride;
455 uint8_t *src2 = src1+i_stride;
456 uint8_t *dst0 = &frame->lowres[0][y*i_stride2];
457 uint8_t *dsth = &frame->lowres[1][y*i_stride2];
458 uint8_t *dstv = &frame->lowres[2][y*i_stride2];
459 uint8_t *dstc = &frame->lowres[3][y*i_stride2];
460 for( x = 0; x < i_width2 - 1; x++ )
462 dst0[x] = (src0[2*x ] + src0[2*x+1] + src1[2*x ] + src1[2*x+1] + 2) >> 2;
463 dsth[x] = (src0[2*x+1] + src0[2*x+2] + src1[2*x+1] + src1[2*x+2] + 2) >> 2;
464 dstv[x] = (src1[2*x ] + src1[2*x+1] + src2[2*x ] + src2[2*x+1] + 2) >> 2;
465 dstc[x] = (src1[2*x+1] + src1[2*x+2] + src2[2*x+1] + src2[2*x+2] + 2) >> 2;
467 dst0[x] = (src0[2*x ] + src0[2*x+1] + src1[2*x ] + src1[2*x+1] + 2) >> 2;
468 dstv[x] = (src1[2*x ] + src1[2*x+1] + src2[2*x ] + src2[2*x+1] + 2) >> 2;
469 dsth[x] = (src0[2*x+1] + src1[2*x+1] + 1) >> 1;
470 dstc[x] = (src1[2*x+1] + src2[2*x+1] + 1) >> 1;
472 for( i = 0; i < 4; i++ )
473 memcpy( &frame->lowres[i][y*i_stride2], &frame->lowres[i][(y-1)*i_stride2], i_width2 );
475 for( y = 0; y < 16; y++ )
476 for( x = 0; x < 16; x++ )
477 frame->i_cost_est[x][y] = -1;
479 x264_frame_expand_border_lowres( frame );