Mercurial > mplayer.hg
comparison libmpcodecs/vf_spp.c @ 11277:fffd200d81a7
simple alternative postprocessing filter
author | michael |
---|---|
date | Sun, 26 Oct 2003 23:07:21 +0000 |
parents | |
children | b677102fa650 |
comparison
equal
deleted
inserted
replaced
11276:ec45f2106ec6 | 11277:fffd200d81a7 |
---|---|
1 /* | |
2 Copyright (C) 2003 Michael Niedermayer <michaelni@gmx.at> | |
3 | |
4 This program is free software; you can redistribute it and/or modify | |
5 it under the terms of the GNU General Public License as published by | |
6 the Free Software Foundation; either version 2 of the License, or | |
7 (at your option) any later version. | |
8 | |
9 This program is distributed in the hope that it will be useful, | |
10 but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 GNU General Public License for more details. | |
13 | |
14 You should have received a copy of the GNU General Public License | |
15 along with this program; if not, write to the Free Software | |
16 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
17 */ | |
18 | |
19 /* | |
20 * This implementation is based on an algorithm described in | |
21 * "Aria Nosratinia Embedded Post-Processing for | |
22 * Enhancement of Compressed Images (1999)" | |
23 * (http://citeseer.nj.nec.com/nosratinia99embedded.html) | |
24 */ | |
25 | |
26 #include <stdio.h> | |
27 #include <stdlib.h> | |
28 #include <string.h> | |
29 #include <inttypes.h> | |
30 #include <math.h> | |
31 | |
32 #include "../config.h" | |
33 #include "../mp_msg.h" | |
34 #include "../cpudetect.h" | |
35 #include "../libavcodec/avcodec.h" | |
36 #include "../libavcodec/dsputil.h" | |
37 | |
38 #ifdef HAVE_MALLOC_H | |
39 #include <malloc.h> | |
40 #endif | |
41 | |
42 #include "img_format.h" | |
43 #include "mp_image.h" | |
44 #include "vf.h" | |
45 #include "../libvo/fastmemcpy.h" | |
46 | |
47 #define XMIN(a,b) ((a) < (b) ? (a) : (b)) | |
48 | |
49 //===========================================================================// | |
50 const uint8_t __attribute__((aligned(8))) dither[8][8]={ | |
51 { 0, 48, 12, 60, 3, 51, 15, 63, }, | |
52 { 32, 16, 44, 28, 35, 19, 47, 31, }, | |
53 { 8, 56, 4, 52, 11, 59, 7, 55, }, | |
54 { 40, 24, 36, 20, 43, 27, 39, 23, }, | |
55 { 2, 50, 14, 62, 1, 49, 13, 61, }, | |
56 { 34, 18, 46, 30, 33, 17, 45, 29, }, | |
57 { 10, 58, 6, 54, 9, 57, 5, 53, }, | |
58 { 42, 26, 38, 22, 41, 25, 37, 21, }, | |
59 }; | |
60 | |
61 const uint8_t offset[64][2]= { | |
62 {0,0}, {4,4}, {0,4}, {4,0}, {2,2}, {6,6}, {2,6}, {6,2}, | |
63 {0,2}, {4,6}, {0,6}, {4,2}, {2,0}, {6,4}, {2,4}, {6,0}, | |
64 {1,1}, {5,5}, {1,5}, {5,1}, {3,3}, {7,7}, {3,7}, {7,3}, | |
65 {1,3}, {5,7}, {1,7}, {5,3}, {3,1}, {7,5}, {3,5}, {7,1}, | |
66 {0,1}, {4,5}, {0,5}, {4,1}, {2,3}, {6,7}, {2,7}, {6,3}, | |
67 {0,3}, {4,7}, {0,7}, {4,3}, {2,1}, {6,5}, {2,5}, {6,1}, | |
68 {1,0}, {5,4}, {1,4}, {5,0}, {3,2}, {7,6}, {3,6}, {7,2}, | |
69 {1,2}, {5,6}, {1,6}, {5,2}, {3,0}, {7,4}, {3,4}, {7,0}, | |
70 }; | |
71 | |
72 struct vf_priv_s { | |
73 int log2_count; | |
74 int qp; | |
75 int mpeg2; | |
76 unsigned int outfmt; | |
77 int temp_stride; | |
78 uint8_t *src; | |
79 int16_t *temp; | |
80 AVCodecContext *avctx; | |
81 DSPContext dsp; | |
82 }; | |
83 | |
84 #define SHIFT 22 | |
85 | |
86 static inline void requantize(DCTELEM dst[64], DCTELEM src[64], int qp, uint8_t *permutation){ | |
87 int i; | |
88 const int qmul= qp<<1; | |
89 const int qadd= (qp-1)|1; | |
90 const int qinv= ((1<<(SHIFT-3)) + qmul/2)/ qmul; | |
91 int bias= 0; //FIXME | |
92 unsigned int threshold1, threshold2; | |
93 | |
94 threshold1= (1<<SHIFT) - bias - 1; | |
95 threshold2= (threshold1<<1); | |
96 | |
97 memset(dst, 0, 64*sizeof(DCTELEM)); | |
98 dst[0]= (src[0] + 4)>>3;; | |
99 | |
100 for(i=1; i<64; i++){ | |
101 int level= qinv*src[i]; | |
102 | |
103 if(((unsigned)(level+threshold1))>threshold2){ | |
104 const int j= permutation[i]; | |
105 if(level>0){ | |
106 level= (bias + level)>>SHIFT; | |
107 dst[j]= level*qmul + qadd; | |
108 }else{ | |
109 level= (bias - level)>>SHIFT; | |
110 dst[j]= -level*qmul - qadd; | |
111 } | |
112 } | |
113 } | |
114 } | |
115 | |
116 static inline void add_block(int16_t *dst, int stride, DCTELEM block[64]){ | |
117 int x,y; | |
118 | |
119 for(y=0; y<8; y++){ | |
120 for(x=0; x<8; x++){ | |
121 dst[x + y*stride]+= block[x + y*8]; | |
122 } | |
123 } | |
124 } | |
125 | |
126 static void filter(struct vf_priv_s *p, uint8_t *dst, uint8_t *src, int dst_stride, int src_stride, int width, int height, uint8_t *qp_store, int qp_stride, int is_luma){ | |
127 int x, y, i; | |
128 const int count= 1<<p->log2_count; | |
129 const int log2_scale= 6-p->log2_count; | |
130 const int stride= p->temp_stride; | |
131 uint64_t block_align[32]; | |
132 DCTELEM *block = (DCTELEM *)block_align; | |
133 DCTELEM *block2= (DCTELEM *)(block_align+16); | |
134 | |
135 for(y=0; y<height; y++){ | |
136 memcpy(p->src + 8 + 8*stride + y*stride, src + y*src_stride, width); | |
137 memset(p->temp + 8*stride + y*stride, 0, stride*sizeof(int16_t)); | |
138 for(x=0; x<8; x++){ | |
139 int index= 8 + 8*stride + y*stride; | |
140 p->src[index - x - 1]= p->src[index + x ]; | |
141 p->src[index + width + x ]= p->src[index + width - x - 1]; | |
142 } | |
143 } | |
144 for(y=0; y<8; y++){ | |
145 memcpy(p->src + ( 7-y)*stride, p->src + ( y+8)*stride, stride); | |
146 memcpy(p->src + (width+8+y)*stride, p->src + (width-y+7)*stride, stride); | |
147 } | |
148 //FIXME (try edge emu) | |
149 | |
150 for(y=0; y<height+8; y+=8){ | |
151 for(x=0; x<width+8; x+=8){ | |
152 const int qps= 3 + is_luma; | |
153 int qp; | |
154 | |
155 if(p->qp) | |
156 qp= p->qp; | |
157 else{ | |
158 qp= qp_store[ (XMIN(x, width-1)>>qps) + (XMIN(y, height-1)>>qps) * qp_stride]; | |
159 if(p->mpeg2) qp>>=1; | |
160 } | |
161 for(i=0; i<count; i++){ | |
162 const int x1= x + offset[i][0]; | |
163 const int y1= y + offset[i][1]; | |
164 const int index= x1 + y1*stride; | |
165 | |
166 p->dsp.get_pixels(block, p->src + index, stride); | |
167 p->dsp.fdct(block); | |
168 requantize(block2, block, qp, p->dsp.idct_permutation); | |
169 p->dsp.idct(block2); | |
170 add_block(p->temp + index, stride, block2); | |
171 } | |
172 } | |
173 } | |
174 | |
175 for(y=0; y<height; y++){ | |
176 uint8_t *d= dither[y&7]; | |
177 for(x=0; x<width; x+=8){ | |
178 const int index= 8 + 8*stride + x + y*stride; | |
179 dst[x + y*src_stride + 0]= ((p->temp[index + 0]<<log2_scale) + d[0])>>6; | |
180 dst[x + y*src_stride + 1]= ((p->temp[index + 1]<<log2_scale) + d[1])>>6; | |
181 dst[x + y*src_stride + 2]= ((p->temp[index + 2]<<log2_scale) + d[2])>>6; | |
182 dst[x + y*src_stride + 3]= ((p->temp[index + 3]<<log2_scale) + d[3])>>6; | |
183 dst[x + y*src_stride + 4]= ((p->temp[index + 4]<<log2_scale) + d[4])>>6; | |
184 dst[x + y*src_stride + 5]= ((p->temp[index + 5]<<log2_scale) + d[5])>>6; | |
185 dst[x + y*src_stride + 6]= ((p->temp[index + 6]<<log2_scale) + d[6])>>6; | |
186 dst[x + y*src_stride + 7]= ((p->temp[index + 7]<<log2_scale) + d[7])>>6; | |
187 } | |
188 } | |
189 //FIXME reorder for better caching | |
190 } | |
191 | |
192 static int config(struct vf_instance_s* vf, | |
193 int width, int height, int d_width, int d_height, | |
194 unsigned int flags, unsigned int outfmt){ | |
195 | |
196 vf->priv->temp_stride= (width+16+15)&(~15); | |
197 vf->priv->temp= malloc(vf->priv->temp_stride*(height+16)*sizeof(int16_t)); | |
198 vf->priv->src = malloc(vf->priv->temp_stride*(height+16)*sizeof(uint8_t)); | |
199 | |
200 return vf_next_config(vf,width,height,d_width,d_height,flags,outfmt); | |
201 } | |
202 | |
203 static void get_image(struct vf_instance_s* vf, mp_image_t *mpi){ | |
204 if(mpi->flags&MP_IMGFLAG_PRESERVE) return; // don't change | |
205 if(mpi->imgfmt!=vf->priv->outfmt) return; // colorspace differ | |
206 // ok, we can do pp in-place (or pp disabled): | |
207 vf->dmpi=vf_get_image(vf->next,mpi->imgfmt, | |
208 mpi->type, mpi->flags, mpi->w, mpi->h); | |
209 mpi->planes[0]=vf->dmpi->planes[0]; | |
210 mpi->stride[0]=vf->dmpi->stride[0]; | |
211 mpi->width=vf->dmpi->width; | |
212 if(mpi->flags&MP_IMGFLAG_PLANAR){ | |
213 mpi->planes[1]=vf->dmpi->planes[1]; | |
214 mpi->planes[2]=vf->dmpi->planes[2]; | |
215 mpi->stride[1]=vf->dmpi->stride[1]; | |
216 mpi->stride[2]=vf->dmpi->stride[2]; | |
217 } | |
218 mpi->flags|=MP_IMGFLAG_DIRECT; | |
219 } | |
220 | |
221 static int put_image(struct vf_instance_s* vf, mp_image_t *mpi){ | |
222 mp_image_t *dmpi; | |
223 | |
224 if(!(mpi->flags&MP_IMGFLAG_DIRECT)){ | |
225 // no DR, so get a new image! hope we'll get DR buffer: | |
226 vf->dmpi=vf_get_image(vf->next,vf->priv->outfmt, | |
227 MP_IMGTYPE_TEMP, MP_IMGFLAG_ACCEPT_STRIDE, | |
228 mpi->w,mpi->h); | |
229 } | |
230 | |
231 dmpi= vf->dmpi; | |
232 | |
233 vf->priv->mpeg2= mpi->qscale_type; | |
234 | |
235 filter(vf->priv, dmpi->planes[0], mpi->planes[0], dmpi->stride[0], mpi->stride[0], mpi->w, mpi->h, mpi->qscale, mpi->qstride, 1); | |
236 filter(vf->priv, dmpi->planes[1], mpi->planes[1], dmpi->stride[1], mpi->stride[1], mpi->w>>1, mpi->h>>1, mpi->qscale, mpi->qstride, 0); | |
237 filter(vf->priv, dmpi->planes[2], mpi->planes[2], dmpi->stride[2], mpi->stride[2], mpi->w>>1, mpi->h>>1, mpi->qscale, mpi->qstride, 0); | |
238 | |
239 vf_clone_mpi_attributes(dmpi, mpi); | |
240 | |
241 #ifdef HAVE_MMX | |
242 if(gCpuCaps.hasMMX) asm volatile ("emms\n\t"); | |
243 #endif | |
244 #ifdef HAVE_MMX2 | |
245 if(gCpuCaps.hasMMX2) asm volatile ("sfence\n\t"); | |
246 #endif | |
247 | |
248 return vf_next_put_image(vf,dmpi); | |
249 } | |
250 | |
251 static void uninit(struct vf_instance_s* vf){ | |
252 if(!vf->priv) return; | |
253 | |
254 if(vf->priv->temp) free(vf->priv->temp); | |
255 vf->priv->temp= NULL; | |
256 if(vf->priv->src) free(vf->priv->src); | |
257 vf->priv->src= NULL; | |
258 if(vf->priv->avctx) free(vf->priv->avctx); | |
259 vf->priv->avctx= NULL; | |
260 | |
261 free(vf->priv); | |
262 vf->priv=NULL; | |
263 } | |
264 | |
265 //===========================================================================// | |
266 | |
267 static int query_format(struct vf_instance_s* vf, unsigned int fmt){ | |
268 switch(fmt) | |
269 { | |
270 case IMGFMT_YV12: | |
271 case IMGFMT_I420: | |
272 case IMGFMT_IYUV: | |
273 return vf_next_query_format(vf,vf->priv->outfmt); | |
274 } | |
275 return 0; | |
276 } | |
277 | |
278 static unsigned int fmt_list[]={ | |
279 IMGFMT_YV12, | |
280 IMGFMT_I420, | |
281 IMGFMT_IYUV, | |
282 0 | |
283 }; | |
284 | |
285 static int open(vf_instance_t *vf, char* args){ | |
286 vf->config=config; | |
287 vf->put_image=put_image; | |
288 vf->get_image=get_image; | |
289 vf->query_format=query_format; | |
290 vf->uninit=uninit; | |
291 vf->priv=malloc(sizeof(struct vf_priv_s)); | |
292 memset(vf->priv, 0, sizeof(struct vf_priv_s)); | |
293 vf->priv->avctx= avcodec_alloc_context(); | |
294 dsputil_init(&vf->priv->dsp, vf->priv->avctx); | |
295 | |
296 vf->priv->log2_count= 6; | |
297 | |
298 if (args) sscanf(args, "%d:%d", &vf->priv->log2_count, &vf->priv->qp); | |
299 | |
300 // check csp: | |
301 vf->priv->outfmt=vf_match_csp(&vf->next,fmt_list,IMGFMT_YV12); | |
302 if(!vf->priv->outfmt) | |
303 { | |
304 uninit(vf); | |
305 return 0; // no csp match :( | |
306 } | |
307 | |
308 return 1; | |
309 } | |
310 | |
311 vf_info_t vf_info_spp = { | |
312 "simple postprocess", | |
313 "spp", | |
314 "Michael Niedermayer", | |
315 "", | |
316 open, | |
317 NULL | |
318 }; | |
319 | |
320 //===========================================================================// |