1 /*****************************************************************************
2 * mc.c: h264 encoder library (Motion Compensation)
3 *****************************************************************************
4 * Copyright (C) 2003-2008 x264 project
6 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7 * Loren Merritt <lorenm@u.washington.edu>
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., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
22 *****************************************************************************/
34 static inline void pixel_avg( uint8_t *dst, int i_dst_stride,
35 uint8_t *src1, int i_src1_stride,
36 uint8_t *src2, int i_src2_stride,
37 int i_width, int i_height )
40 for( y = 0; y < i_height; y++ )
42 for( x = 0; x < i_width; x++ )
44 dst[x] = ( src1[x] + src2[x] + 1 ) >> 1;
47 src1 += i_src1_stride;
48 src2 += i_src2_stride;
52 static inline void pixel_avg_wxh( uint8_t *dst, int i_dst, uint8_t *src, int i_src, int width, int height )
55 for( y = 0; y < height; y++ )
57 for( x = 0; x < width; x++ )
59 dst[x] = ( dst[x] + src[x] + 1 ) >> 1;
66 #define PIXEL_AVG_C( name, width, height ) \
67 static void name( uint8_t *pix1, int i_stride_pix1, \
68 uint8_t *pix2, int i_stride_pix2 ) \
70 pixel_avg_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height ); \
72 PIXEL_AVG_C( pixel_avg_16x16, 16, 16 )
73 PIXEL_AVG_C( pixel_avg_16x8, 16, 8 )
74 PIXEL_AVG_C( pixel_avg_8x16, 8, 16 )
75 PIXEL_AVG_C( pixel_avg_8x8, 8, 8 )
76 PIXEL_AVG_C( pixel_avg_8x4, 8, 4 )
77 PIXEL_AVG_C( pixel_avg_4x8, 4, 8 )
78 PIXEL_AVG_C( pixel_avg_4x4, 4, 4 )
79 PIXEL_AVG_C( pixel_avg_4x2, 4, 2 )
80 PIXEL_AVG_C( pixel_avg_2x4, 2, 4 )
81 PIXEL_AVG_C( pixel_avg_2x2, 2, 2 )
84 /* Implicit weighted bipred only:
85 * assumes log2_denom = 5, offset = 0, weight1 + weight2 = 64 */
86 #define op_scale2(x) dst[x] = x264_clip_uint8( (dst[x]*i_weight1 + src[x]*i_weight2 + (1<<5)) >> 6 )
87 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 ){
89 const int i_weight2 = 64 - i_weight1;
90 for(y=0; y<height; y++, dst += i_dst, src += i_src){
93 if(width==2) continue;
96 if(width==4) continue;
101 if(width==8) continue;
113 #define PIXEL_AVG_WEIGHT_C( width, height ) \
114 static void pixel_avg_weight_##width##x##height( \
115 uint8_t *pix1, int i_stride_pix1, \
116 uint8_t *pix2, int i_stride_pix2, int i_weight1 ) \
118 pixel_avg_weight_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height, i_weight1 ); \
121 PIXEL_AVG_WEIGHT_C(16,16)
122 PIXEL_AVG_WEIGHT_C(16,8)
123 PIXEL_AVG_WEIGHT_C(8,16)
124 PIXEL_AVG_WEIGHT_C(8,8)
125 PIXEL_AVG_WEIGHT_C(8,4)
126 PIXEL_AVG_WEIGHT_C(4,8)
127 PIXEL_AVG_WEIGHT_C(4,4)
128 PIXEL_AVG_WEIGHT_C(4,2)
129 PIXEL_AVG_WEIGHT_C(2,4)
130 PIXEL_AVG_WEIGHT_C(2,2)
132 #undef PIXEL_AVG_WEIGHT_C
134 static void mc_copy( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_width, int i_height )
138 for( y = 0; y < i_height; y++ )
140 memcpy( dst, src, i_width );
147 #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]))
148 static void hpel_filter( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src,
149 int stride, int width, int height )
151 int16_t *buf = x264_malloc((width+5)*sizeof(int16_t));
153 for( y=0; y<height; y++ )
155 for( x=-2; x<width+3; x++ )
157 int v = TAPFILTER(src,stride);
158 dstv[x] = x264_clip_uint8((v + 16) >> 5);
161 for( x=0; x<width; x++ )
162 dstc[x] = x264_clip_uint8((TAPFILTER(buf+2,1) + 512) >> 10);
163 for( x=0; x<width; x++ )
164 dsth[x] = x264_clip_uint8((TAPFILTER(src,1) + 16) >> 5);
173 static const int hpel_ref0[16] = {0,1,1,1,0,1,1,1,2,3,3,3,0,1,1,1};
174 static const int hpel_ref1[16] = {0,0,0,0,2,2,3,2,2,2,3,2,2,2,3,2};
176 static void mc_luma( uint8_t *dst, int i_dst_stride,
177 uint8_t *src[4], int i_src_stride,
179 int i_width, int i_height )
181 int qpel_idx = ((mvy&3)<<2) + (mvx&3);
182 int offset = (mvy>>2)*i_src_stride + (mvx>>2);
183 uint8_t *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
185 if( qpel_idx & 5 ) /* qpel interpolation needed */
187 uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
188 pixel_avg( dst, i_dst_stride, src1, i_src_stride,
189 src2, i_src_stride, i_width, i_height );
193 mc_copy( src1, i_src_stride, dst, i_dst_stride, i_width, i_height );
197 static uint8_t *get_ref( uint8_t *dst, int *i_dst_stride,
198 uint8_t *src[4], int i_src_stride,
200 int i_width, int i_height )
202 int qpel_idx = ((mvy&3)<<2) + (mvx&3);
203 int offset = (mvy>>2)*i_src_stride + (mvx>>2);
204 uint8_t *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
206 if( qpel_idx & 5 ) /* qpel interpolation needed */
208 uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
209 pixel_avg( dst, *i_dst_stride, src1, i_src_stride,
210 src2, i_src_stride, i_width, i_height );
215 *i_dst_stride = i_src_stride;
220 /* full chroma mc (ie until 1/8 pixel)*/
221 static void mc_chroma( uint8_t *dst, int i_dst_stride,
222 uint8_t *src, int i_src_stride,
224 int i_width, int i_height )
229 const int d8x = mvx&0x07;
230 const int d8y = mvy&0x07;
232 const int cA = (8-d8x)*(8-d8y);
233 const int cB = d8x *(8-d8y);
234 const int cC = (8-d8x)*d8y;
235 const int cD = d8x *d8y;
237 src += (mvy >> 3) * i_src_stride + (mvx >> 3);
238 srcp = &src[i_src_stride];
240 for( y = 0; y < i_height; y++ )
242 for( x = 0; x < i_width; x++ )
244 dst[x] = ( cA*src[x] + cB*src[x+1] +
245 cC*srcp[x] + cD*srcp[x+1] + 32 ) >> 6;
250 srcp += i_src_stride;
255 static void mc_copy_w##W( uint8_t *dst, int i_dst, uint8_t *src, int i_src, int i_height ) \
257 mc_copy( src, i_src, dst, i_dst, W, i_height ); \
263 static void plane_copy( uint8_t *dst, int i_dst,
264 uint8_t *src, int i_src, int w, int h)
268 memcpy( dst, src, w );
274 static void prefetch_fenc_null( uint8_t *pix_y, int stride_y,
275 uint8_t *pix_uv, int stride_uv, int mb_x )
278 static void prefetch_ref_null( uint8_t *pix, int stride, int parity )
281 static void memzero_aligned( void * dst, int n )
286 void x264_frame_init_lowres( x264_t *h, x264_frame_t *frame )
288 uint8_t *src = frame->plane[0];
289 int i_stride = frame->i_stride[0];
290 int i_height = frame->i_lines[0];
291 int i_width = frame->i_width[0];
294 // duplicate last row and column so that their interpolation doesn't have to be special-cased
295 for( y=0; y<i_height; y++ )
296 src[i_width+y*i_stride] = src[i_width-1+y*i_stride];
297 h->mc.memcpy_aligned( src+i_stride*i_height, src+i_stride*(i_height-1), i_width );
298 h->mc.frame_init_lowres_core( src, frame->lowres[0], frame->lowres[1], frame->lowres[2], frame->lowres[3],
299 i_stride, frame->i_stride_lowres, frame->i_width_lowres, frame->i_lines_lowres );
300 x264_frame_expand_border_lowres( frame );
302 for( y=0; y<16; y++ )
303 for( x=0; x<16; x++ )
304 frame->i_cost_est[y][x] = -1;
307 static void frame_init_lowres_core( uint8_t *src0, uint8_t *dst0, uint8_t *dsth, uint8_t *dstv, uint8_t *dstc,
308 int src_stride, int dst_stride, int width, int height )
311 for( y=0; y<height; y++ )
313 uint8_t *src1 = src0+src_stride;
314 uint8_t *src2 = src1+src_stride;
315 for( x=0; x<width; x++ )
317 // slower than naive bilinear, but matches asm
318 #define FILTER(a,b,c,d) ((((a+b+1)>>1)+((c+d+1)>>1)+1)>>1)
319 dst0[x] = FILTER(src0[2*x ], src1[2*x ], src0[2*x+1], src1[2*x+1]);
320 dsth[x] = FILTER(src0[2*x+1], src1[2*x+1], src0[2*x+2], src1[2*x+2]);
321 dstv[x] = FILTER(src1[2*x ], src2[2*x ], src1[2*x+1], src2[2*x+1]);
322 dstc[x] = FILTER(src1[2*x+1], src2[2*x+1], src1[2*x+2], src2[2*x+2]);
325 src0 += src_stride*2;
333 void x264_mc_init( int cpu, x264_mc_functions_t *pf )
335 pf->mc_luma = mc_luma;
336 pf->get_ref = get_ref;
337 pf->mc_chroma = mc_chroma;
339 pf->avg[PIXEL_16x16]= pixel_avg_16x16;
340 pf->avg[PIXEL_16x8] = pixel_avg_16x8;
341 pf->avg[PIXEL_8x16] = pixel_avg_8x16;
342 pf->avg[PIXEL_8x8] = pixel_avg_8x8;
343 pf->avg[PIXEL_8x4] = pixel_avg_8x4;
344 pf->avg[PIXEL_4x8] = pixel_avg_4x8;
345 pf->avg[PIXEL_4x4] = pixel_avg_4x4;
346 pf->avg[PIXEL_4x2] = pixel_avg_4x2;
347 pf->avg[PIXEL_2x4] = pixel_avg_2x4;
348 pf->avg[PIXEL_2x2] = pixel_avg_2x2;
350 pf->avg_weight[PIXEL_16x16]= pixel_avg_weight_16x16;
351 pf->avg_weight[PIXEL_16x8] = pixel_avg_weight_16x8;
352 pf->avg_weight[PIXEL_8x16] = pixel_avg_weight_8x16;
353 pf->avg_weight[PIXEL_8x8] = pixel_avg_weight_8x8;
354 pf->avg_weight[PIXEL_8x4] = pixel_avg_weight_8x4;
355 pf->avg_weight[PIXEL_4x8] = pixel_avg_weight_4x8;
356 pf->avg_weight[PIXEL_4x4] = pixel_avg_weight_4x4;
357 pf->avg_weight[PIXEL_4x2] = pixel_avg_weight_4x2;
358 pf->avg_weight[PIXEL_2x4] = pixel_avg_weight_2x4;
359 pf->avg_weight[PIXEL_2x2] = pixel_avg_weight_2x2;
361 pf->copy[PIXEL_16x16] = mc_copy_w16;
362 pf->copy[PIXEL_8x8] = mc_copy_w8;
363 pf->copy[PIXEL_4x4] = mc_copy_w4;
365 pf->plane_copy = plane_copy;
366 pf->hpel_filter = hpel_filter;
368 pf->prefetch_fenc = prefetch_fenc_null;
369 pf->prefetch_ref = prefetch_ref_null;
370 pf->memcpy_aligned = memcpy;
371 pf->memzero_aligned = memzero_aligned;
372 pf->frame_init_lowres_core = frame_init_lowres_core;
375 x264_mc_init_mmx( cpu, pf );
378 if( cpu&X264_CPU_ALTIVEC )
379 x264_mc_altivec_init( pf );
383 void x264_frame_filter( x264_t *h, x264_frame_t *frame, int mb_y, int b_end )
385 const int b_interlaced = h->sh.b_mbaff;
386 const int stride = frame->i_stride[0] << b_interlaced;
387 const int width = frame->i_width[0];
388 int start = (mb_y*16 >> b_interlaced) - 8; // buffer = 4 for deblock + 3 for 6tap, rounded to 8
389 int height = ((b_end ? frame->i_lines[0] : mb_y*16) >> b_interlaced) + 8;
390 int offs = start*stride - 8; // buffer = 3 for 6tap, aligned to 8 for simd
393 if( mb_y & b_interlaced )
396 for( y=0; y<=b_interlaced; y++, offs+=frame->i_stride[0] )
399 frame->filtered[1] + offs,
400 frame->filtered[2] + offs,
401 frame->filtered[3] + offs,
402 frame->plane[0] + offs,
403 stride, width + 16, height - start );
406 /* generate integral image:
407 * frame->integral contains 2 planes. in the upper plane, each element is
408 * the sum of an 8x8 pixel region with top-left corner on that point.
409 * in the lower plane, 4x4 sums (needed only with --partitions p4x4). */
411 if( frame->integral )
415 memset( frame->integral - PADV * stride - PADH, 0, stride * sizeof(uint16_t) );
420 for( y = start; y < height; y++ )
422 uint8_t *ref = frame->plane[0] + y * stride - PADH;
423 uint16_t *line = frame->integral + (y+1) * stride - PADH + 1;
424 uint16_t v = line[0] = 0;
425 for( x = 1; x < stride-1; x++ )
426 line[x] = v += ref[x] + line[x-stride] - line[x-stride-1];
430 uint16_t *sum4 = line + stride * (frame->i_lines[0] + PADV*2);
431 for( x = 1; x < stride-8; x++, line++, sum4++ )
433 sum4[0] = line[4+4*stride] - line[4] - line[4*stride] + line[0];
434 line[0] += line[8+8*stride] - line[8] - line[8*stride];