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 *src1, int i_src1, uint8_t *src2, int i_src2, int width, int height )
55 for( y = 0; y < height; y++ )
57 for( x = 0; x < width; x++ )
59 dst[x] = ( src1[x] + src2[x] + 1 ) >> 1;
67 /* Implicit weighted bipred only:
68 * assumes log2_denom = 5, offset = 0, weight1 + weight2 = 64 */
69 #define op_scale2(x) dst[x] = x264_clip_uint8( (src1[x]*i_weight1 + src2[x]*i_weight2 + (1<<5)) >> 6 )
70 static inline void pixel_avg_weight_wxh( uint8_t *dst, int i_dst, uint8_t *src1, int i_src1, uint8_t *src2, int i_src2, int width, int height, int i_weight1 )
73 const int i_weight2 = 64 - i_weight1;
74 for( y = 0; y<height; y++, dst += i_dst, src1 += i_src1, src2 += i_src2 )
78 if(width==2) continue;
81 if(width==4) continue;
86 if(width==8) continue;
99 #define PIXEL_AVG_C( name, width, height ) \
100 static void name( uint8_t *pix1, int i_stride_pix1, \
101 uint8_t *pix2, int i_stride_pix2, \
102 uint8_t *pix3, int i_stride_pix3, int weight ) \
105 pixel_avg_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, pix3, i_stride_pix3, width, height ); \
107 pixel_avg_weight_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, pix3, i_stride_pix3, width, height, weight ); \
109 PIXEL_AVG_C( pixel_avg_16x16, 16, 16 )
110 PIXEL_AVG_C( pixel_avg_16x8, 16, 8 )
111 PIXEL_AVG_C( pixel_avg_8x16, 8, 16 )
112 PIXEL_AVG_C( pixel_avg_8x8, 8, 8 )
113 PIXEL_AVG_C( pixel_avg_8x4, 8, 4 )
114 PIXEL_AVG_C( pixel_avg_4x8, 4, 8 )
115 PIXEL_AVG_C( pixel_avg_4x4, 4, 4 )
116 PIXEL_AVG_C( pixel_avg_4x2, 4, 2 )
117 PIXEL_AVG_C( pixel_avg_2x4, 2, 4 )
118 PIXEL_AVG_C( pixel_avg_2x2, 2, 2 )
120 static void mc_copy( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_width, int i_height )
124 for( y = 0; y < i_height; y++ )
126 memcpy( dst, src, i_width );
133 #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]))
134 static void hpel_filter( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src,
135 int stride, int width, int height )
137 int16_t *buf = x264_malloc((width+5)*sizeof(int16_t));
139 for( y=0; y<height; y++ )
141 for( x=-2; x<width+3; x++ )
143 int v = TAPFILTER(src,stride);
144 dstv[x] = x264_clip_uint8((v + 16) >> 5);
147 for( x=0; x<width; x++ )
148 dstc[x] = x264_clip_uint8((TAPFILTER(buf+2,1) + 512) >> 10);
149 for( x=0; x<width; x++ )
150 dsth[x] = x264_clip_uint8((TAPFILTER(src,1) + 16) >> 5);
159 static const int hpel_ref0[16] = {0,1,1,1,0,1,1,1,2,3,3,3,0,1,1,1};
160 static const int hpel_ref1[16] = {0,0,0,0,2,2,3,2,2,2,3,2,2,2,3,2};
162 static void mc_luma( uint8_t *dst, int i_dst_stride,
163 uint8_t *src[4], int i_src_stride,
165 int i_width, int i_height )
167 int qpel_idx = ((mvy&3)<<2) + (mvx&3);
168 int offset = (mvy>>2)*i_src_stride + (mvx>>2);
169 uint8_t *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
171 if( qpel_idx & 5 ) /* qpel interpolation needed */
173 uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
174 pixel_avg( dst, i_dst_stride, src1, i_src_stride,
175 src2, i_src_stride, i_width, i_height );
179 mc_copy( src1, i_src_stride, dst, i_dst_stride, i_width, i_height );
183 static uint8_t *get_ref( uint8_t *dst, int *i_dst_stride,
184 uint8_t *src[4], int i_src_stride,
186 int i_width, int i_height )
188 int qpel_idx = ((mvy&3)<<2) + (mvx&3);
189 int offset = (mvy>>2)*i_src_stride + (mvx>>2);
190 uint8_t *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
192 if( qpel_idx & 5 ) /* qpel interpolation needed */
194 uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
195 pixel_avg( dst, *i_dst_stride, src1, i_src_stride,
196 src2, i_src_stride, i_width, i_height );
201 *i_dst_stride = i_src_stride;
206 /* full chroma mc (ie until 1/8 pixel)*/
207 static void mc_chroma( uint8_t *dst, int i_dst_stride,
208 uint8_t *src, int i_src_stride,
210 int i_width, int i_height )
215 const int d8x = mvx&0x07;
216 const int d8y = mvy&0x07;
218 const int cA = (8-d8x)*(8-d8y);
219 const int cB = d8x *(8-d8y);
220 const int cC = (8-d8x)*d8y;
221 const int cD = d8x *d8y;
223 src += (mvy >> 3) * i_src_stride + (mvx >> 3);
224 srcp = &src[i_src_stride];
226 for( y = 0; y < i_height; y++ )
228 for( x = 0; x < i_width; x++ )
230 dst[x] = ( cA*src[x] + cB*src[x+1] +
231 cC*srcp[x] + cD*srcp[x+1] + 32 ) >> 6;
236 srcp += i_src_stride;
241 static void mc_copy_w##W( uint8_t *dst, int i_dst, uint8_t *src, int i_src, int i_height ) \
243 mc_copy( src, i_src, dst, i_dst, W, i_height ); \
249 static void plane_copy( uint8_t *dst, int i_dst,
250 uint8_t *src, int i_src, int w, int h)
254 memcpy( dst, src, w );
260 static void prefetch_fenc_null( uint8_t *pix_y, int stride_y,
261 uint8_t *pix_uv, int stride_uv, int mb_x )
264 static void prefetch_ref_null( uint8_t *pix, int stride, int parity )
267 static void memzero_aligned( void * dst, int n )
272 static void integral_init4h( uint16_t *sum, uint8_t *pix, int stride )
274 int x, v = pix[0]+pix[1]+pix[2]+pix[3];
275 for( x=0; x<stride-4; x++ )
277 sum[x] = v + sum[x-stride];
278 v += pix[x+4] - pix[x];
282 static void integral_init8h( uint16_t *sum, uint8_t *pix, int stride )
284 int x, v = pix[0]+pix[1]+pix[2]+pix[3]+pix[4]+pix[5]+pix[6]+pix[7];
285 for( x=0; x<stride-8; x++ )
287 sum[x] = v + sum[x-stride];
288 v += pix[x+8] - pix[x];
292 static void integral_init4v( uint16_t *sum8, uint16_t *sum4, int stride )
295 for( x=0; x<stride-8; x++ )
296 sum4[x] = sum8[x+4*stride] - sum8[x];
297 for( x=0; x<stride-8; x++ )
298 sum8[x] = sum8[x+8*stride] + sum8[x+8*stride+4] - sum8[x] - sum8[x+4];
301 static void integral_init8v( uint16_t *sum8, int stride )
304 for( x=0; x<stride-8; x++ )
305 sum8[x] = sum8[x+8*stride] - sum8[x];
308 void x264_frame_init_lowres( x264_t *h, x264_frame_t *frame )
310 uint8_t *src = frame->plane[0];
311 int i_stride = frame->i_stride[0];
312 int i_height = frame->i_lines[0];
313 int i_width = frame->i_width[0];
316 // duplicate last row and column so that their interpolation doesn't have to be special-cased
317 for( y=0; y<i_height; y++ )
318 src[i_width+y*i_stride] = src[i_width-1+y*i_stride];
319 h->mc.memcpy_aligned( src+i_stride*i_height, src+i_stride*(i_height-1), i_width );
320 h->mc.frame_init_lowres_core( src, frame->lowres[0], frame->lowres[1], frame->lowres[2], frame->lowres[3],
321 i_stride, frame->i_stride_lowres, frame->i_width_lowres, frame->i_lines_lowres );
322 x264_frame_expand_border_lowres( frame );
324 memset( frame->i_cost_est, -1, sizeof(frame->i_cost_est) );
326 for( x = 0; x < h->param.i_bframe + 2; x++ )
327 for( y = 0; y < h->param.i_bframe + 2; y++ )
328 frame->i_row_satds[y][x][0] = -1;
330 for( y = 0; y <= !!h->param.i_bframe; y++ )
331 for( x = 0; x <= h->param.i_bframe; x++ )
332 frame->lowres_mvs[y][x][0][0] = 0x7FFF;
335 static void frame_init_lowres_core( uint8_t *src0, uint8_t *dst0, uint8_t *dsth, uint8_t *dstv, uint8_t *dstc,
336 int src_stride, int dst_stride, int width, int height )
339 for( y=0; y<height; y++ )
341 uint8_t *src1 = src0+src_stride;
342 uint8_t *src2 = src1+src_stride;
343 for( x=0; x<width; x++ )
345 // slower than naive bilinear, but matches asm
346 #define FILTER(a,b,c,d) ((((a+b+1)>>1)+((c+d+1)>>1)+1)>>1)
347 dst0[x] = FILTER(src0[2*x ], src1[2*x ], src0[2*x+1], src1[2*x+1]);
348 dsth[x] = FILTER(src0[2*x+1], src1[2*x+1], src0[2*x+2], src1[2*x+2]);
349 dstv[x] = FILTER(src1[2*x ], src2[2*x ], src1[2*x+1], src2[2*x+1]);
350 dstc[x] = FILTER(src1[2*x+1], src2[2*x+1], src1[2*x+2], src2[2*x+2]);
353 src0 += src_stride*2;
361 void x264_mc_init( int cpu, x264_mc_functions_t *pf )
363 pf->mc_luma = mc_luma;
364 pf->get_ref = get_ref;
365 pf->mc_chroma = mc_chroma;
367 pf->avg[PIXEL_16x16]= pixel_avg_16x16;
368 pf->avg[PIXEL_16x8] = pixel_avg_16x8;
369 pf->avg[PIXEL_8x16] = pixel_avg_8x16;
370 pf->avg[PIXEL_8x8] = pixel_avg_8x8;
371 pf->avg[PIXEL_8x4] = pixel_avg_8x4;
372 pf->avg[PIXEL_4x8] = pixel_avg_4x8;
373 pf->avg[PIXEL_4x4] = pixel_avg_4x4;
374 pf->avg[PIXEL_4x2] = pixel_avg_4x2;
375 pf->avg[PIXEL_2x4] = pixel_avg_2x4;
376 pf->avg[PIXEL_2x2] = pixel_avg_2x2;
378 pf->copy_16x16_unaligned = mc_copy_w16;
379 pf->copy[PIXEL_16x16] = mc_copy_w16;
380 pf->copy[PIXEL_8x8] = mc_copy_w8;
381 pf->copy[PIXEL_4x4] = mc_copy_w4;
383 pf->plane_copy = plane_copy;
384 pf->hpel_filter = hpel_filter;
386 pf->prefetch_fenc = prefetch_fenc_null;
387 pf->prefetch_ref = prefetch_ref_null;
388 pf->memcpy_aligned = memcpy;
389 pf->memzero_aligned = memzero_aligned;
390 pf->frame_init_lowres_core = frame_init_lowres_core;
392 pf->integral_init4h = integral_init4h;
393 pf->integral_init8h = integral_init8h;
394 pf->integral_init4v = integral_init4v;
395 pf->integral_init8v = integral_init8v;
398 x264_mc_init_mmx( cpu, pf );
401 if( cpu&X264_CPU_ALTIVEC )
402 x264_mc_altivec_init( pf );
406 void x264_frame_filter( x264_t *h, x264_frame_t *frame, int mb_y, int b_end )
408 const int b_interlaced = h->sh.b_mbaff;
409 const int stride = frame->i_stride[0] << b_interlaced;
410 const int width = frame->i_width[0];
411 int start = (mb_y*16 >> b_interlaced) - 8; // buffer = 4 for deblock + 3 for 6tap, rounded to 8
412 int height = ((b_end ? frame->i_lines[0] : mb_y*16) >> b_interlaced) + 8;
413 int offs = start*stride - 8; // buffer = 3 for 6tap, aligned to 8 for simd
416 if( mb_y & b_interlaced )
419 for( y=0; y<=b_interlaced; y++, offs+=frame->i_stride[0] )
422 frame->filtered[1] + offs,
423 frame->filtered[2] + offs,
424 frame->filtered[3] + offs,
425 frame->plane[0] + offs,
426 stride, width + 16, height - start );
429 /* generate integral image:
430 * frame->integral contains 2 planes. in the upper plane, each element is
431 * the sum of an 8x8 pixel region with top-left corner on that point.
432 * in the lower plane, 4x4 sums (needed only with --partitions p4x4). */
434 if( frame->integral )
438 memset( frame->integral - PADV * stride - PADH, 0, stride * sizeof(uint16_t) );
443 for( y = start; y < height; y++ )
445 uint8_t *pix = frame->plane[0] + y * stride - PADH;
446 uint16_t *sum8 = frame->integral + (y+1) * stride - PADH;
448 if( h->frames.b_have_sub8x8_esa )
450 h->mc.integral_init4h( sum8, pix, stride );
452 sum4 = sum8 + stride * (frame->i_lines[0] + PADV*2);
454 h->mc.integral_init4v( sum8, sum4, stride );
458 h->mc.integral_init8h( sum8, pix, stride );
460 h->mc.integral_init8v( sum8-8*stride, stride );