filter_avdeinterlace.c: bugfix (kdenlive-672) deinterlace only works on left half...
[melted] / src / modules / avformat / filter_avdeinterlace.c
1 /*
2 * filter_avdeinterlace.c -- deinterlace filter
3 * Copyright (C) 2003-2004 Ushodaya Enterprises Limited
4 * Author: Charles Yates <charles.yates@pandora.be>
5 *
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 */
20
21 #include <framework/mlt_filter.h>
22 #include <framework/mlt_frame.h>
23
24 #include <string.h>
25 #include <stdlib.h>
26
27 // ffmpeg Header files
28 #include <avformat.h>
29
30 #ifdef USE_MMX
31 #include "mmx.h"
32 #else
33 #define MAX_NEG_CROP 1024
34 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
35 #endif
36
37 #ifdef USE_MMX
38 #define DEINT_INPLACE_LINE_LUM \
39 movd_m2r(lum_m4[0],mm0);\
40 movd_m2r(lum_m3[0],mm1);\
41 movd_m2r(lum_m2[0],mm2);\
42 movd_m2r(lum_m1[0],mm3);\
43 movd_m2r(lum[0],mm4);\
44 punpcklbw_r2r(mm7,mm0);\
45 movd_r2m(mm2,lum_m4[0]);\
46 punpcklbw_r2r(mm7,mm1);\
47 punpcklbw_r2r(mm7,mm2);\
48 punpcklbw_r2r(mm7,mm3);\
49 punpcklbw_r2r(mm7,mm4);\
50 paddw_r2r(mm3,mm1);\
51 psllw_i2r(1,mm2);\
52 paddw_r2r(mm4,mm0);\
53 psllw_i2r(2,mm1);\
54 paddw_r2r(mm6,mm2);\
55 paddw_r2r(mm2,mm1);\
56 psubusw_r2r(mm0,mm1);\
57 psrlw_i2r(3,mm1);\
58 packuswb_r2r(mm7,mm1);\
59 movd_r2m(mm1,lum_m2[0]);
60
61 #define DEINT_LINE_LUM \
62 movd_m2r(lum_m4[0],mm0);\
63 movd_m2r(lum_m3[0],mm1);\
64 movd_m2r(lum_m2[0],mm2);\
65 movd_m2r(lum_m1[0],mm3);\
66 movd_m2r(lum[0],mm4);\
67 punpcklbw_r2r(mm7,mm0);\
68 punpcklbw_r2r(mm7,mm1);\
69 punpcklbw_r2r(mm7,mm2);\
70 punpcklbw_r2r(mm7,mm3);\
71 punpcklbw_r2r(mm7,mm4);\
72 paddw_r2r(mm3,mm1);\
73 psllw_i2r(1,mm2);\
74 paddw_r2r(mm4,mm0);\
75 psllw_i2r(2,mm1);\
76 paddw_r2r(mm6,mm2);\
77 paddw_r2r(mm2,mm1);\
78 psubusw_r2r(mm0,mm1);\
79 psrlw_i2r(3,mm1);\
80 packuswb_r2r(mm7,mm1);\
81 movd_r2m(mm1,dst[0]);
82 #endif
83
84 /* filter parameters: [-1 4 2 4 -1] // 8 */
85 static inline void deinterlace_line(uint8_t *dst,
86 const uint8_t *lum_m4, const uint8_t *lum_m3,
87 const uint8_t *lum_m2, const uint8_t *lum_m1,
88 const uint8_t *lum,
89 int size)
90 {
91 #ifndef USE_MMX
92 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
93 int sum;
94
95 for(;size > 0;size--) {
96 sum = -lum_m4[0];
97 sum += lum_m3[0] << 2;
98 sum += lum_m2[0] << 1;
99 sum += lum_m1[0] << 2;
100 sum += -lum[0];
101 dst[0] = cm[(sum + 4) >> 3];
102 lum_m4++;
103 lum_m3++;
104 lum_m2++;
105 lum_m1++;
106 lum++;
107 dst++;
108 }
109 #else
110
111 {
112 mmx_t rounder;
113 rounder.uw[0]=4;
114 rounder.uw[1]=4;
115 rounder.uw[2]=4;
116 rounder.uw[3]=4;
117 pxor_r2r(mm7,mm7);
118 movq_m2r(rounder,mm6);
119 }
120 for (;size > 3; size-=4) {
121 DEINT_LINE_LUM
122 lum_m4+=4;
123 lum_m3+=4;
124 lum_m2+=4;
125 lum_m1+=4;
126 lum+=4;
127 dst+=4;
128 }
129 #endif
130 }
131 static inline void deinterlace_line_inplace(uint8_t *lum_m4, uint8_t *lum_m3, uint8_t *lum_m2, uint8_t *lum_m1, uint8_t *lum,
132 int size)
133 {
134 #ifndef USE_MMX
135 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
136 int sum;
137
138 for(;size > 0;size--) {
139 sum = -lum_m4[0];
140 sum += lum_m3[0] << 2;
141 sum += lum_m2[0] << 1;
142 lum_m4[0]=lum_m2[0];
143 sum += lum_m1[0] << 2;
144 sum += -lum[0];
145 lum_m2[0] = cm[(sum + 4) >> 3];
146 lum_m4++;
147 lum_m3++;
148 lum_m2++;
149 lum_m1++;
150 lum++;
151 }
152 #else
153
154 {
155 mmx_t rounder;
156 rounder.uw[0]=4;
157 rounder.uw[1]=4;
158 rounder.uw[2]=4;
159 rounder.uw[3]=4;
160 pxor_r2r(mm7,mm7);
161 movq_m2r(rounder,mm6);
162 }
163 for (;size > 3; size-=4) {
164 DEINT_INPLACE_LINE_LUM
165 lum_m4+=4;
166 lum_m3+=4;
167 lum_m2+=4;
168 lum_m1+=4;
169 lum+=4;
170 }
171 #endif
172 }
173
174 /* deinterlacing : 2 temporal taps, 3 spatial taps linear filter. The
175 top field is copied as is, but the bottom field is deinterlaced
176 against the top field. */
177 static inline void deinterlace_bottom_field(uint8_t *dst, int dst_wrap,
178 const uint8_t *src1, int src_wrap,
179 int width, int height)
180 {
181 const uint8_t *src_m2, *src_m1, *src_0, *src_p1, *src_p2;
182 int y;
183
184 src_m2 = src1;
185 src_m1 = src1;
186 src_0=&src_m1[src_wrap];
187 src_p1=&src_0[src_wrap];
188 src_p2=&src_p1[src_wrap];
189 for(y=0;y<(height-2);y+=2) {
190 memcpy(dst,src_m1,width);
191 dst += dst_wrap;
192 deinterlace_line(dst,src_m2,src_m1,src_0,src_p1,src_p2,width);
193 src_m2 = src_0;
194 src_m1 = src_p1;
195 src_0 = src_p2;
196 src_p1 += 2*src_wrap;
197 src_p2 += 2*src_wrap;
198 dst += dst_wrap;
199 }
200 memcpy(dst,src_m1,width);
201 dst += dst_wrap;
202 /* do last line */
203 deinterlace_line(dst,src_m2,src_m1,src_0,src_0,src_0,width);
204 }
205
206 static inline void deinterlace_bottom_field_inplace(uint8_t *src1, int src_wrap,
207 int width, int height)
208 {
209 uint8_t *src_m1, *src_0, *src_p1, *src_p2;
210 int y;
211 uint8_t *buf;
212 buf = (uint8_t*)av_malloc(width);
213
214 src_m1 = src1;
215 memcpy(buf,src_m1,width);
216 src_0=&src_m1[src_wrap];
217 src_p1=&src_0[src_wrap];
218 src_p2=&src_p1[src_wrap];
219 for(y=0;y<(height-2);y+=2) {
220 deinterlace_line_inplace(buf,src_m1,src_0,src_p1,src_p2,width);
221 src_m1 = src_p1;
222 src_0 = src_p2;
223 src_p1 += 2*src_wrap;
224 src_p2 += 2*src_wrap;
225 }
226 /* do last line */
227 deinterlace_line_inplace(buf,src_m1,src_0,src_0,src_0,width);
228 av_free(buf);
229 }
230
231
232 /* deinterlace - if not supported return -1 */
233 static int mlt_avpicture_deinterlace(AVPicture *dst, const AVPicture *src,
234 int pix_fmt, int width, int height)
235 {
236 int i;
237
238 if (pix_fmt != PIX_FMT_YUV420P &&
239 pix_fmt != PIX_FMT_YUV422P &&
240 pix_fmt != PIX_FMT_YUV422 &&
241 pix_fmt != PIX_FMT_YUV444P &&
242 pix_fmt != PIX_FMT_YUV411P)
243 return -1;
244 if ((width & 3) != 0 || (height & 3) != 0)
245 return -1;
246
247 if ( pix_fmt != PIX_FMT_YUV422 )
248 {
249 for(i=0;i<3;i++) {
250 if (i == 1) {
251 switch(pix_fmt) {
252 case PIX_FMT_YUV420P:
253 width >>= 1;
254 height >>= 1;
255 break;
256 case PIX_FMT_YUV422P:
257 width >>= 1;
258 break;
259 case PIX_FMT_YUV411P:
260 width >>= 2;
261 break;
262 default:
263 break;
264 }
265 }
266 if (src == dst) {
267 deinterlace_bottom_field_inplace(dst->data[i], dst->linesize[i],
268 width, height);
269 } else {
270 deinterlace_bottom_field(dst->data[i],dst->linesize[i],
271 src->data[i], src->linesize[i],
272 width, height);
273 }
274 }
275 }
276 else {
277 if (src == dst) {
278 deinterlace_bottom_field_inplace(dst->data[0], dst->linesize[0],
279 width<<1, height);
280 } else {
281 deinterlace_bottom_field(dst->data[0],dst->linesize[0],
282 src->data[0], src->linesize[0],
283 width<<1, height);
284 }
285 }
286
287 #ifdef USE_MMX
288 emms();
289 #endif
290 return 0;
291 }
292
293 /** Do it :-).
294 */
295
296 static int filter_get_image( mlt_frame this, uint8_t **image, mlt_image_format *format, int *width, int *height, int writable )
297 {
298 int error = 0;
299 int deinterlace = mlt_properties_get_int( MLT_FRAME_PROPERTIES( this ), "consumer_deinterlace" );
300
301 // Determine if we need a writable version or not
302 if ( deinterlace && !writable )
303 writable = !mlt_properties_get_int( MLT_FRAME_PROPERTIES( this ), "progressive" );
304
305 // Get the input image
306 error = mlt_frame_get_image( this, image, format, width, height, writable );
307
308 // Check that we want progressive and we aren't already progressive
309 if ( deinterlace && *format == mlt_image_yuv422 && *image != NULL && !mlt_properties_get_int( MLT_FRAME_PROPERTIES( this ), "progressive" ) )
310 {
311 // Create a picture
312 AVPicture *output = mlt_pool_alloc( sizeof( AVPicture ) );
313
314 // Fill the picture
315 if ( *format == mlt_image_yuv422 )
316 {
317 avpicture_fill( output, *image, PIX_FMT_YUV422, *width, *height );
318 mlt_avpicture_deinterlace( output, output, PIX_FMT_YUV422, *width, *height );
319 }
320
321 // Free the picture
322 mlt_pool_release( output );
323
324 // Make sure that others know the frame is deinterlaced
325 mlt_properties_set_int( MLT_FRAME_PROPERTIES( this ), "progressive", 1 );
326 }
327
328 return error;
329 }
330
331 /** Deinterlace filter processing - this should be lazy evaluation here...
332 */
333
334 static mlt_frame deinterlace_process( mlt_filter this, mlt_frame frame )
335 {
336 // Push the get_image method on to the stack
337 mlt_frame_push_get_image( frame, filter_get_image );
338
339 return frame;
340 }
341
342 /** Constructor for the filter.
343 */
344
345 mlt_filter filter_avdeinterlace_init( void *arg )
346 {
347 mlt_filter this = mlt_filter_new( );
348 if ( this != NULL )
349 this->process = deinterlace_process;
350 return this;
351 }
352