spudec.c: minor improvements
[mplayer/glamo.git] / libmpcodecs / vf_gradfun.c
blob813b9ecb3f690590f8eaa612c3d0b4d6224d68f2
1 /*
2 * Copyright (C) 2009 Loren Merritt <lorenm@u.washignton.edu>
4 * This file is part of MPlayer.
6 * MPlayer is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * MPlayer 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
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License along
17 * with MPlayer; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
22 * Debanding algorithm (from gradfun2db by prunedtree):
23 * Boxblur.
24 * Foreach pixel, if it's within threshold of the blurred value, make it closer.
25 * So now we have a smoothed and higher bitdepth version of all the shallow
26 * gradients, while leaving detailed areas untouched.
27 * Dither it back to 8bit.
30 #include <stdio.h>
31 #include <stdlib.h>
32 #include <string.h>
33 #include <inttypes.h>
35 #include "config.h"
36 #include "cpudetect.h"
37 #include "img_format.h"
38 #include "mp_image.h"
39 #include "vf.h"
40 #include "libvo/fastmemcpy.h"
41 #include "libavutil/avutil.h"
42 #include "ffmpeg_files/x86_cpu.h"
44 struct vf_priv_s {
45 int thresh;
46 int radius;
47 uint16_t *buf;
48 void (*filter_line)(uint8_t *dst, uint8_t *src, uint16_t *dc,
49 int width, int thresh, const uint16_t *dithers);
50 void (*blur_line)(uint16_t *dc, uint16_t *buf, uint16_t *buf1,
51 uint8_t *src, int sstride, int width);
54 static const uint16_t __attribute__((aligned(16))) pw_7f[8] = {127,127,127,127,127,127,127,127};
55 static const uint16_t __attribute__((aligned(16))) pw_ff[8] = {255,255,255,255,255,255,255,255};
56 static const uint16_t __attribute__((aligned(16))) dither[8][8] = {
57 { 0, 96, 24,120, 6,102, 30,126 },
58 { 64, 32, 88, 56, 70, 38, 94, 62 },
59 { 16,112, 8,104, 22,118, 14,110 },
60 { 80, 48, 72, 40, 86, 54, 78, 46 },
61 { 4,100, 28,124, 2, 98, 26,122 },
62 { 68, 36, 92, 60, 66, 34, 90, 58 },
63 { 20,116, 12,108, 18,114, 10,106 },
64 { 84, 52, 76, 44, 82, 50, 74, 42 },
67 static void filter_line_c(uint8_t *dst, uint8_t *src, uint16_t *dc,
68 int width, int thresh, const uint16_t *dithers)
70 int x;
71 for (x=0; x<width; x++, dc+=x&1) {
72 int pix = src[x]<<7;
73 int delta = dc[0] - pix;
74 int m = abs(delta) * thresh >> 16;
75 m = FFMAX(0, 127-m);
76 m = m*m*delta >> 14;
77 pix += m + dithers[x&7];
78 dst[x] = av_clip_uint8(pix>>7);
82 static void blur_line_c(uint16_t *dc, uint16_t *buf, uint16_t *buf1,
83 uint8_t *src, int sstride, int width)
85 int x, v, old;
86 for (x=0; x<width; x++) {
87 v = buf1[x] + src[2*x] + src[2*x+1] + src[2*x+sstride] + src[2*x+1+sstride];
88 old = buf[x];
89 buf[x] = v;
90 dc[x] = v - old;
94 #if HAVE_MMX2
95 static void filter_line_mmx2(uint8_t *dst, uint8_t *src, uint16_t *dc,
96 int width, int thresh, const uint16_t *dithers)
98 intptr_t x;
99 if (width&3) {
100 x = width&~3;
101 filter_line_c(dst+x, src+x, dc+x/2, width-x, thresh, dithers);
102 width = x;
104 x = -width;
105 __asm__ volatile(
106 "movd %4, %%mm5 \n"
107 "pxor %%mm7, %%mm7 \n"
108 "pshufw $0, %%mm5, %%mm5 \n"
109 "movq %6, %%mm6 \n"
110 "movq %5, %%mm4 \n"
111 "1: \n"
112 "movd (%2,%0), %%mm0 \n"
113 "movd (%3,%0), %%mm1 \n"
114 "punpcklbw %%mm7, %%mm0 \n"
115 "punpcklwd %%mm1, %%mm1 \n"
116 "psllw $7, %%mm0 \n"
117 "pxor %%mm2, %%mm2 \n"
118 "psubw %%mm0, %%mm1 \n" // delta = dc - pix
119 "psubw %%mm1, %%mm2 \n"
120 "pmaxsw %%mm1, %%mm2 \n"
121 "pmulhuw %%mm5, %%mm2 \n" // m = abs(delta) * thresh >> 16
122 "psubw %%mm6, %%mm2 \n"
123 "pminsw %%mm7, %%mm2 \n" // m = -max(0, 127-m)
124 "pmullw %%mm2, %%mm2 \n"
125 "paddw %%mm4, %%mm0 \n" // pix += dither
126 "pmulhw %%mm2, %%mm1 \n"
127 "psllw $2, %%mm1 \n" // m = m*m*delta >> 14
128 "paddw %%mm1, %%mm0 \n" // pix += m
129 "psraw $7, %%mm0 \n"
130 "packuswb %%mm0, %%mm0 \n"
131 "movd %%mm0, (%1,%0) \n" // dst = clip(pix>>7)
132 "add $4, %0 \n"
133 "jl 1b \n"
134 "emms \n"
135 :"+r"(x)
136 :"r"(dst+width), "r"(src+width), "r"(dc+width/2),
137 "rm"(thresh), "m"(*dithers), "m"(*pw_7f)
138 :"memory"
141 #endif
143 #if HAVE_SSSE3
144 static void filter_line_ssse3(uint8_t *dst, uint8_t *src, uint16_t *dc,
145 int width, int thresh, const uint16_t *dithers)
147 intptr_t x;
148 if (width&7) {
149 // could be 10% faster if I somehow eliminated this
150 x = width&~7;
151 filter_line_c(dst+x, src+x, dc+x/2, width-x, thresh, dithers);
152 width = x;
154 x = -width;
155 __asm__ volatile(
156 "movd %4, %%xmm5 \n"
157 "pxor %%xmm7, %%xmm7 \n"
158 "pshuflw $0,%%xmm5, %%xmm5 \n"
159 "movdqa %6, %%xmm6 \n"
160 "punpcklqdq %%xmm5, %%xmm5 \n"
161 "movdqa %5, %%xmm4 \n"
162 "1: \n"
163 "movq (%2,%0), %%xmm0 \n"
164 "movq (%3,%0), %%xmm1 \n"
165 "punpcklbw %%xmm7, %%xmm0 \n"
166 "punpcklwd %%xmm1, %%xmm1 \n"
167 "psllw $7, %%xmm0 \n"
168 "psubw %%xmm0, %%xmm1 \n" // delta = dc - pix
169 "pabsw %%xmm1, %%xmm2 \n"
170 "pmulhuw %%xmm5, %%xmm2 \n" // m = abs(delta) * thresh >> 16
171 "psubw %%xmm6, %%xmm2 \n"
172 "pminsw %%xmm7, %%xmm2 \n" // m = -max(0, 127-m)
173 "pmullw %%xmm2, %%xmm2 \n"
174 "psllw $1, %%xmm2 \n"
175 "paddw %%xmm4, %%xmm0 \n" // pix += dither
176 "pmulhrsw %%xmm2, %%xmm1 \n" // m = m*m*delta >> 14
177 "paddw %%xmm1, %%xmm0 \n" // pix += m
178 "psraw $7, %%xmm0 \n"
179 "packuswb %%xmm0, %%xmm0 \n"
180 "movq %%xmm0, (%1,%0) \n" // dst = clip(pix>>7)
181 "add $8, %0 \n"
182 "jl 1b \n"
183 :"+&r"(x)
184 :"r"(dst+width), "r"(src+width), "r"(dc+width/2),
185 "rm"(thresh), "m"(*dithers), "m"(*pw_7f)
186 :"memory"
189 #endif // HAVE_SSSE3
191 #if HAVE_SSE2 && HAVE_6REGS
192 #define BLURV(load)\
193 intptr_t x = -2*width;\
194 __asm__ volatile(\
195 "movdqa %6, %%xmm7 \n"\
196 "1: \n"\
197 load" (%4,%0), %%xmm0 \n"\
198 load" (%5,%0), %%xmm1 \n"\
199 "movdqa %%xmm0, %%xmm2 \n"\
200 "movdqa %%xmm1, %%xmm3 \n"\
201 "psrlw $8, %%xmm0 \n"\
202 "psrlw $8, %%xmm1 \n"\
203 "pand %%xmm7, %%xmm2 \n"\
204 "pand %%xmm7, %%xmm3 \n"\
205 "paddw %%xmm1, %%xmm0 \n"\
206 "paddw %%xmm3, %%xmm2 \n"\
207 "paddw %%xmm2, %%xmm0 \n"\
208 "paddw (%2,%0), %%xmm0 \n"\
209 "movdqa (%1,%0), %%xmm1 \n"\
210 "movdqa %%xmm0, (%1,%0) \n"\
211 "psubw %%xmm1, %%xmm0 \n"\
212 "movdqa %%xmm0, (%3,%0) \n"\
213 "add $16, %0 \n"\
214 "jl 1b \n"\
215 :"+&r"(x)\
216 :"r"(buf+width),\
217 "r"(buf1+width),\
218 "r"(dc+width),\
219 "r"(src+width*2),\
220 "r"(src+width*2+sstride),\
221 "m"(*pw_ff)\
222 :"memory"\
225 static void blur_line_sse2(uint16_t *dc, uint16_t *buf, uint16_t *buf1,
226 uint8_t *src, int sstride, int width)
228 if (((intptr_t)src|sstride)&15) {
229 BLURV("movdqu");
230 } else {
231 BLURV("movdqa");
234 #endif // HAVE_6REGS && HAVE_SSE2
236 static void filter(struct vf_priv_s *ctx, uint8_t *dst, uint8_t *src,
237 int width, int height, int dstride, int sstride, int r)
239 int bstride = ((width+15)&~15)/2;
240 int y;
241 uint32_t dc_factor = (1<<21)/(r*r);
242 uint16_t *dc = ctx->buf+16;
243 uint16_t *buf = ctx->buf+bstride+32;
244 int thresh = ctx->thresh;
246 memset(dc, 0, (bstride+16)*sizeof(*buf));
247 for (y=0; y<r; y++)
248 ctx->blur_line(dc, buf+y*bstride, buf+(y-1)*bstride, src+2*y*sstride, sstride, width/2);
249 for (;;) {
250 if (y < height-r) {
251 int mod = ((y+r)/2)%r;
252 uint16_t *buf0 = buf+mod*bstride;
253 uint16_t *buf1 = buf+(mod?mod-1:r-1)*bstride;
254 int x, v;
255 ctx->blur_line(dc, buf0, buf1, src+(y+r)*sstride, sstride, width/2);
256 for (x=v=0; x<r; x++)
257 v += dc[x];
258 for (; x<width/2; x++) {
259 v += dc[x] - dc[x-r];
260 dc[x-r] = v * dc_factor >> 16;
262 for (; x<(width+r+1)/2; x++)
263 dc[x-r] = v * dc_factor >> 16;
264 for (x=-r/2; x<0; x++)
265 dc[x] = dc[0];
267 if (y == r) {
268 for (y=0; y<r; y++)
269 ctx->filter_line(dst+y*dstride, src+y*sstride, dc-r/2, width, thresh, dither[y&7]);
271 ctx->filter_line(dst+y*dstride, src+y*sstride, dc-r/2, width, thresh, dither[y&7]);
272 if (++y >= height) break;
273 ctx->filter_line(dst+y*dstride, src+y*sstride, dc-r/2, width, thresh, dither[y&7]);
274 if (++y >= height) break;
278 static void get_image(struct vf_instance *vf, mp_image_t *mpi)
280 if (mpi->flags&MP_IMGFLAG_PRESERVE) return; // don't change
281 // ok, we can do pp in-place:
282 vf->dmpi = vf_get_image(vf->next, mpi->imgfmt,
283 mpi->type, mpi->flags, mpi->width, mpi->height);
284 mpi->planes[0] = vf->dmpi->planes[0];
285 mpi->stride[0] = vf->dmpi->stride[0];
286 mpi->width = vf->dmpi->width;
287 if (mpi->flags&MP_IMGFLAG_PLANAR){
288 mpi->planes[1] = vf->dmpi->planes[1];
289 mpi->planes[2] = vf->dmpi->planes[2];
290 mpi->stride[1] = vf->dmpi->stride[1];
291 mpi->stride[2] = vf->dmpi->stride[2];
293 mpi->flags |= MP_IMGFLAG_DIRECT;
296 static int put_image(struct vf_instance *vf, mp_image_t *mpi, double pts)
298 mp_image_t *dmpi = vf->dmpi;
299 int p;
301 if (!(mpi->flags&MP_IMGFLAG_DIRECT)) {
302 // no DR, so get a new image. hope we'll get DR buffer:
303 dmpi = vf_get_image(vf->next,mpi->imgfmt, MP_IMGTYPE_TEMP,
304 MP_IMGFLAG_ACCEPT_STRIDE|MP_IMGFLAG_PREFER_ALIGNED_STRIDE,
305 mpi->w, mpi->h);
307 vf_clone_mpi_attributes(dmpi, mpi);
309 for (p=0; p<mpi->num_planes; p++) {
310 int w = mpi->w;
311 int h = mpi->h;
312 int r = vf->priv->radius;
313 if (p) {
314 w >>= mpi->chroma_x_shift;
315 h >>= mpi->chroma_y_shift;
316 r = ((r>>mpi->chroma_x_shift) + (r>>mpi->chroma_y_shift)) / 2;
317 r = av_clip((r+1)&~1,4,32);
319 if (FFMIN(w,h) > 2*r)
320 filter(vf->priv, dmpi->planes[p], mpi->planes[p], w, h,
321 dmpi->stride[p], mpi->stride[p], r);
322 else if (dmpi->planes[p] != mpi->planes[p])
323 memcpy_pic(dmpi->planes[p], mpi->planes[p], w, h,
324 dmpi->stride[p], mpi->stride[p]);
327 return vf_next_put_image(vf, dmpi, pts);
330 static int query_format(struct vf_instance *vf, unsigned int fmt)
332 switch (fmt){
333 case IMGFMT_YVU9:
334 case IMGFMT_IF09:
335 case IMGFMT_YV12:
336 case IMGFMT_I420:
337 case IMGFMT_IYUV:
338 case IMGFMT_CLPL:
339 case IMGFMT_Y800:
340 case IMGFMT_Y8:
341 case IMGFMT_NV12:
342 case IMGFMT_NV21:
343 case IMGFMT_444P:
344 case IMGFMT_422P:
345 case IMGFMT_411P:
346 case IMGFMT_HM12:
347 return vf_next_query_format(vf,fmt);
349 return 0;
352 static int config(struct vf_instance *vf,
353 int width, int height, int d_width, int d_height,
354 unsigned int flags, unsigned int outfmt)
356 free(vf->priv->buf);
357 vf->priv->buf = av_mallocz((((width+15)&~15)*(vf->priv->radius+1)/2+32)*sizeof(uint16_t));
358 return vf_next_config(vf,width,height,d_width,d_height,flags,outfmt);
361 static void uninit(struct vf_instance *vf)
363 if (!vf->priv) return;
364 av_free(vf->priv->buf);
365 free(vf->priv);
366 vf->priv = NULL;
369 static int vf_open(vf_instance_t *vf, char *args)
371 float thresh = 1.2;
372 int radius = 16;
374 vf->get_image=get_image;
375 vf->put_image=put_image;
376 vf->query_format=query_format;
377 vf->config=config;
378 vf->uninit=uninit;
379 vf->priv=malloc(sizeof(struct vf_priv_s));
380 memset(vf->priv, 0, sizeof(struct vf_priv_s));
382 if (args) sscanf(args, "%f:%d", &thresh, &radius);
383 vf->priv->thresh = (1<<15)/av_clipf(thresh,0.51,255);
384 vf->priv->radius = av_clip((radius+1)&~1,4,32);
386 vf->priv->blur_line = blur_line_c;
387 vf->priv->filter_line = filter_line_c;
388 #if HAVE_SSE2 && HAVE_6REGS
389 if (gCpuCaps.hasSSE2)
390 vf->priv->blur_line = blur_line_sse2;
391 #endif
392 #if HAVE_MMX2
393 if (gCpuCaps.hasMMX2)
394 vf->priv->filter_line = filter_line_mmx2;
395 #endif
396 #if HAVE_SSSE3
397 if (gCpuCaps.hasSSSE3)
398 vf->priv->filter_line = filter_line_ssse3;
399 #endif
401 return 1;
404 const vf_info_t vf_info_gradfun = {
405 "gradient deband",
406 "gradfun",
407 "Loren Merritt",
409 vf_open,
410 NULL