libmpcodecs/vf_gradfun.c

   1 /*
   2  * Copyright (C) 2009 Loren Merritt <lorenm@u.washignton.edu>
   3  *
   4  * This file is part of MPlayer.
   5  *
   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.
  10  *
  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.
  15  *
  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.
  19  */
  20
  21 /*
  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.
  28  */
  29
  30 #include <stdio.h>
  31 #include <stdlib.h>
  32 #include <string.h>
  33 #include <inttypes.h>
  34
  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 "libavutil/x86_cpu.h"
  43
  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);
  52 };
  53
  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 },
  65 };
  66
  67 static void filter_line_c(uint8_t *dst, uint8_t *src, uint16_t *dc,
  68                           int width, int thresh, const uint16_t *dithers)
  69 {
  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);
  79     }
  80 }
  81
  82 static void blur_line_c(uint16_t *dc, uint16_t *buf, uint16_t *buf1,
  83                         uint8_t *src, int sstride, int width)
  84 {
  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;
  91     }
  92 }
  93
  94 #if HAVE_SSSE3
  95 static void filter_line_mmx2(uint8_t *dst, uint8_t *src, uint16_t *dc,
  96                              int width, int thresh, const uint16_t *dithers)
  97 {
  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;
 103     }
 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"
 139     );
 140 }
 141
 142 static void filter_line_ssse3(uint8_t *dst, uint8_t *src, uint16_t *dc,
 143                               int width, int thresh, const uint16_t *dithers)
 144 {
 145     intptr_t x;
 146     if (width&7) {
 147         // could be 10% faster if I somehow eliminated this
 148         x = width&~7;
 149         filter_line_c(dst+x, src+x, dc+x/2, width-x, thresh, dithers);
 150         width = x;
 151     }
 152     x = -width;
 153     asm volatile(
 154         "movd           %4, %%xmm5 \n"
 155         "pxor       %%xmm7, %%xmm7 \n"
 156         "pshuflw $0,%%xmm5, %%xmm5 \n"
 157         "movdqa         %6, %%xmm6 \n"
 158         "punpcklqdq %%xmm5, %%xmm5 \n"
 159         "movdqa         %5, %%xmm4 \n"
 160         "1: \n"
 161         "movq      (%2,%0), %%xmm0 \n"
 162         "movq      (%3,%0), %%xmm1 \n"
 163         "punpcklbw  %%xmm7, %%xmm0 \n"
 164         "punpcklwd  %%xmm1, %%xmm1 \n"
 165         "psllw          $7, %%xmm0 \n"
 166         "psubw      %%xmm0, %%xmm1 \n" // delta = dc - pix
 167         "pabsw      %%xmm1, %%xmm2 \n"
 168         "pmulhuw    %%xmm5, %%xmm2 \n" // m = abs(delta) * thresh >> 16
 169         "psubw      %%xmm6, %%xmm2 \n"
 170         "pminsw     %%xmm7, %%xmm2 \n" // m = -max(0, 127-m)
 171         "pmullw     %%xmm2, %%xmm2 \n"
 172         "psllw          $1, %%xmm2 \n"
 173         "paddw      %%xmm4, %%xmm0 \n" // pix += dither
 174         "pmulhrsw   %%xmm2, %%xmm1 \n" // m = m*m*delta >> 14
 175         "paddw      %%xmm1, %%xmm0 \n" // pix += m
 176         "psraw          $7, %%xmm0 \n"
 177         "packuswb   %%xmm0, %%xmm0 \n"
 178         "movq       %%xmm0, (%1,%0) \n" // dst = clip(pix>>7)
 179         "add            $8, %0 \n"
 180         "jl 1b \n"
 181         :"+&r"(x)
 182         :"r"(dst+width), "r"(src+width), "r"(dc+width/2),
 183          "rm"(thresh), "m"(*dithers), "m"(*pw_7f)
 184         :"memory"
 185     );
 186 }
 187
 188 #define BLURV(load)\
 189     intptr_t x = -2*width;\
 190     asm volatile(\
 191         "movdqa %6, %%xmm7 \n"\
 192         "1: \n"\
 193         load"   (%4,%0), %%xmm0 \n"\
 194         load"   (%5,%0), %%xmm1 \n"\
 195         "movdqa  %%xmm0, %%xmm2 \n"\
 196         "movdqa  %%xmm1, %%xmm3 \n"\
 197         "psrlw       $8, %%xmm0 \n"\
 198         "psrlw       $8, %%xmm1 \n"\
 199         "pand    %%xmm7, %%xmm2 \n"\
 200         "pand    %%xmm7, %%xmm3 \n"\
 201         "paddw   %%xmm1, %%xmm0 \n"\
 202         "paddw   %%xmm3, %%xmm2 \n"\
 203         "paddw   %%xmm2, %%xmm0 \n"\
 204         "paddw  (%2,%0), %%xmm0 \n"\
 205         "movdqa (%1,%0), %%xmm1 \n"\
 206         "movdqa  %%xmm0, (%1,%0) \n"\
 207         "psubw   %%xmm1, %%xmm0 \n"\
 208         "movdqa  %%xmm0, (%3,%0) \n"\
 209         "add        $16, %0 \n"\
 210         "jl 1b \n"\
 211         :"+&r"(x)\
 212         :"r"(buf+width),\
 213          "r"(buf1+width),\
 214          "r"(dc+width),\
 215          "r"(src+width*2),\
 216          "r"(src+width*2+sstride),\
 217          "m"(*pw_ff)\
 218         :"memory"\
 219     );
 220
 221 #if HAVE_6REGS
 222 static void blur_line_sse2(uint16_t *dc, uint16_t *buf, uint16_t *buf1,
 223                            uint8_t *src, int sstride, int width)
 224 {
 225     if (((intptr_t)src|sstride)&15) {
 226         BLURV("movdqu");
 227     } else {
 228         BLURV("movdqa");
 229     }
 230 }
 231 #endif // HAVE_6REGS
 232 #endif // HAVE_SSSE3
 233
 234 static void filter(struct vf_priv_s *ctx, uint8_t *dst, uint8_t *src,
 235                    int width, int height, int dstride, int sstride, int r)
 236 {
 237     int bstride = ((width+15)&~15)/2;
 238     int y;
 239     uint32_t dc_factor = (1<<21)/(r*r);
 240     uint16_t *dc = ctx->buf+16;
 241     uint16_t *buf = ctx->buf+bstride+32;
 242     int thresh = ctx->thresh;
 243
 244     memset(dc, 0, (bstride+16)*sizeof(*buf));
 245     for (y=0; y<r; y++)
 246         ctx->blur_line(dc, buf+y*bstride, buf+(y-1)*bstride, src+2*y*sstride, sstride, width/2);
 247     for (;;) {
 248         if (y < height-r) {
 249             int mod = ((y+r)/2)%r;
 250             uint16_t *buf0 = buf+mod*bstride;
 251             uint16_t *buf1 = buf+(mod?mod-1:r-1)*bstride;
 252             int x, v;
 253             ctx->blur_line(dc, buf0, buf1, src+(y+r)*sstride, sstride, width/2);
 254             for (x=v=0; x<r; x++)
 255                 v += dc[x];
 256             for (; x<width/2; x++) {
 257                 v += dc[x] - dc[x-r];
 258                 dc[x-r] = v * dc_factor >> 16;
 259             }
 260             for (; x<(width+r+1)/2; x++)
 261                 dc[x-r] = v * dc_factor >> 16;
 262             for (x=-r/2; x<0; x++)
 263                 dc[x] = dc[0];
 264         }
 265         if (y == r) {
 266             for (y=0; y<r; y++)
 267                 ctx->filter_line(dst+y*dstride, src+y*sstride, dc-r/2, width, thresh, dither[y&7]);
 268         }
 269         ctx->filter_line(dst+y*dstride, src+y*sstride, dc-r/2, width, thresh, dither[y&7]);
 270         if (++y >= height) break;
 271         ctx->filter_line(dst+y*dstride, src+y*sstride, dc-r/2, width, thresh, dither[y&7]);
 272         if (++y >= height) break;
 273     }
 274 }
 275
 276 static void get_image(struct vf_instance_s* vf, mp_image_t *mpi)
 277 {
 278     if (mpi->flags&MP_IMGFLAG_PRESERVE) return; // don't change
 279     // ok, we can do pp in-place:
 280     vf->dmpi = vf_get_image(vf->next, mpi->imgfmt,
 281                             mpi->type, mpi->flags, mpi->width, mpi->height);
 282     mpi->planes[0] = vf->dmpi->planes[0];
 283     mpi->stride[0] = vf->dmpi->stride[0];
 284     mpi->width = vf->dmpi->width;
 285     if (mpi->flags&MP_IMGFLAG_PLANAR){
 286         mpi->planes[1] = vf->dmpi->planes[1];
 287         mpi->planes[2] = vf->dmpi->planes[2];
 288         mpi->stride[1] = vf->dmpi->stride[1];
 289         mpi->stride[2] = vf->dmpi->stride[2];
 290     }
 291     mpi->flags |= MP_IMGFLAG_DIRECT;
 292 }
 293
 294 static int put_image(struct vf_instance_s* vf, mp_image_t *mpi, double pts)
 295 {
 296     mp_image_t *dmpi = vf->dmpi;
 297     int p;
 298
 299     if (!(mpi->flags&MP_IMGFLAG_DIRECT)) {
 300         // no DR, so get a new image. hope we'll get DR buffer:
 301         dmpi = vf_get_image(vf->next,mpi->imgfmt, MP_IMGTYPE_TEMP,
 302                             MP_IMGFLAG_ACCEPT_STRIDE|MP_IMGFLAG_PREFER_ALIGNED_STRIDE,
 303                             mpi->w, mpi->h);
 304     }
 305     vf_clone_mpi_attributes(dmpi, mpi);
 306
 307     for (p=0; p<mpi->num_planes; p++) {
 308         int w = mpi->w;
 309         int h = mpi->h;
 310         int r = vf->priv->radius;
 311         if (p) {
 312             w >>= mpi->chroma_x_shift;
 313             h >>= mpi->chroma_y_shift;
 314             r = ((r>>mpi->chroma_x_shift) + (r>>mpi->chroma_y_shift)) / 2;
 315             r = av_clip((r+1)&~1,4,32);
 316         }
 317         if (FFMIN(w,h) > 2*r)
 318             filter(vf->priv, dmpi->planes[p], mpi->planes[p], w, h,
 319                              dmpi->stride[p], mpi->stride[p], r);
 320         else if (dmpi->planes[p] != mpi->planes[p])
 321             memcpy_pic(dmpi->planes[p], mpi->planes[p], w, h,
 322                        dmpi->stride[p], mpi->stride[p]);
 323     }
 324
 325     return vf_next_put_image(vf, dmpi, pts);
 326 }
 327
 328 static int query_format(struct vf_instance_s* vf, unsigned int fmt)
 329 {
 330     switch (fmt){
 331     case IMGFMT_YVU9:
 332     case IMGFMT_IF09:
 333     case IMGFMT_YV12:
 334     case IMGFMT_I420:
 335     case IMGFMT_IYUV:
 336     case IMGFMT_CLPL:
 337     case IMGFMT_Y800:
 338     case IMGFMT_Y8:
 339     case IMGFMT_NV12:
 340     case IMGFMT_NV21:
 341     case IMGFMT_444P:
 342     case IMGFMT_422P:
 343     case IMGFMT_411P:
 344     case IMGFMT_HM12:
 345         return vf_next_query_format(vf,fmt);
 346     }
 347     return 0;
 348 }
 349
 350 static int config(struct vf_instance_s* vf,
 351                   int width, int height, int d_width, int d_height,
 352                   unsigned int flags, unsigned int outfmt)
 353 {
 354     free(vf->priv->buf);
 355     vf->priv->buf = av_mallocz((((width+15)&~15)*(vf->priv->radius+1)/2+32)*sizeof(uint16_t));
 356     return vf_next_config(vf,width,height,d_width,d_height,flags,outfmt);
 357 }
 358
 359 static void uninit(struct vf_instance_s* vf)
 360 {
 361     if (!vf->priv) return;
 362     av_free(vf->priv->buf);
 363     free(vf->priv);
 364     vf->priv = NULL;
 365 }
 366
 367 static int open(vf_instance_t *vf, char* args)
 368 {
 369     float thresh = 1.2;
 370     int radius = 16;
 371
 372     vf->get_image=get_image;
 373     vf->put_image=put_image;
 374     vf->query_format=query_format;
 375     vf->config=config;
 376     vf->uninit=uninit;
 377     vf->priv=malloc(sizeof(struct vf_priv_s));
 378     memset(vf->priv, 0, sizeof(struct vf_priv_s));
 379
 380     if (args) sscanf(args, "%f:%d", &thresh, &radius);
 381     vf->priv->thresh = (1<<15)/av_clipf(thresh,0.51,255);
 382     vf->priv->radius = av_clip((radius+1)&~1,4,32);
 383
 384     vf->priv->blur_line = blur_line_c;
 385     vf->priv->filter_line = filter_line_c;
 386 #if HAVE_SSSE3
 387 #if HAVE_6REGS
 388     if (gCpuCaps.hasSSE2)
 389         vf->priv->blur_line = blur_line_sse2;
 390 #endif
 391     if (gCpuCaps.hasMMX2)
 392         vf->priv->filter_line = filter_line_mmx2;
 393     if (gCpuCaps.hasSSSE3)
 394         vf->priv->filter_line = filter_line_ssse3;
 395 #endif
 396
 397     return 1;
 398 }
 399
 400 const vf_info_t vf_info_gradfun = {
 401     "gradient deband",
 402     "gradfun",
 403     "Loren Merritt",
 404     "",
 405     open,
 406     NULL
 407 };