modules/video_filter/gradfun.h

   1 /*
   2  * Copyright (c) 2010 Nolan Lum <nol888@gmail.com>
   3  * Copyright (c) 2009 Loren Merritt <lorenm@u.washignton.edu>
   4  *
   5  * This file is part of Libav.
   6  *
   7  * Libav is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * Libav is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with Libav; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 /*
  23  * Debanding algorithm (from gradfun2db by prunedtree):
  24  * Boxblur.
  25  * Foreach pixel, if it's within threshold of the blurred value, make it closer.
  26  * So now we have a smoothed and higher bitdepth version of all the shallow
  27  * gradients, while leaving detailed areas untouched.
  28  * Dither it back to 8bit.
  29  */
  30
  31 struct vf_priv_s {
  32     int thresh;
  33     int radius;
  34     uint16_t *buf;
  35     void (*filter_line)(uint8_t *dst, uint8_t *src, uint16_t *dc,
  36                         int width, int thresh, const uint16_t *dithers);
  37     void (*blur_line)(uint16_t *dc, uint16_t *buf, uint16_t *buf1,
  38                       uint8_t *src, int sstride, int width);
  39 };
  40
  41 static alignas (16) const uint16_t pw_7f[8] = {127,127,127,127,127,127,127,127};
  42 static alignas (16) const uint16_t pw_ff[8] = {255,255,255,255,255,255,255,255};
  43 static alignas (16) const uint16_t dither[8][8] = {
  44     {  0, 96, 24,120,  6,102, 30,126 },
  45     { 64, 32, 88, 56, 70, 38, 94, 62 },
  46     { 16,112,  8,104, 22,118, 14,110 },
  47     { 80, 48, 72, 40, 86, 54, 78, 46 },
  48     {  4,100, 28,124,  2, 98, 26,122 },
  49     { 68, 36, 92, 60, 66, 34, 90, 58 },
  50     { 20,116, 12,108, 18,114, 10,106 },
  51     { 84, 52, 76, 44, 82, 50, 74, 42 },
  52 };
  53
  54 static void filter_line_c(uint8_t *dst, uint8_t *src, uint16_t *dc,
  55                           int width, int thresh, const uint16_t *dithers)
  56 {
  57     for( int x = 0; x < width; x++, dc += x&1 ) {
  58         int pix = src[x]<<7;
  59         int delta = dc[0] - pix;
  60         int m = abs(delta) * thresh >> 16;
  61         m = FFMAX(0, 127-m);
  62         m = m*m*delta >> 14;
  63         pix += m + dithers[x&7];
  64         dst[x] = av_clip_uint8(pix>>7);
  65     }
  66 }
  67
  68 static void blur_line_c(uint16_t *dc, uint16_t *buf, uint16_t *buf1,
  69                         uint8_t *src, int sstride, int width)
  70 {
  71     for( int x = 0; x < width; x++ ) {
  72         int v = buf1[x] + src[2*x] + src[2*x+1] + src[2*x+sstride] +
  73                 src[2*x+1+sstride];
  74         int old = buf[x];
  75         buf[x] = v;
  76         dc[x] = v - old;
  77     }
  78 }
  79
  80 #if HAVE_MMX2
  81 VLC_MMX
  82 static void filter_line_mmx2(uint8_t *dst, uint8_t *src, uint16_t *dc,
  83                              int width, int thresh, const uint16_t *dithers)
  84 {
  85     intptr_t x;
  86     if (width&3) {
  87         x = width&~3;
  88         filter_line_c(dst+x, src+x, dc+x/2, width-x, thresh, dithers);
  89         width = x;
  90     }
  91     x = -width;
  92     __asm__ volatile(
  93         "movd          %4, %%mm5 \n"
  94         "pxor       %%mm7, %%mm7 \n"
  95         "pshufw $0, %%mm5, %%mm5 \n"
  96         "movq          %6, %%mm6 \n"
  97         "movq          %5, %%mm4 \n"
  98         "1: \n"
  99         "movd     (%2,%0), %%mm0 \n"
 100         "movd     (%3,%0), %%mm1 \n"
 101         "punpcklbw  %%mm7, %%mm0 \n"
 102         "punpcklwd  %%mm1, %%mm1 \n"
 103         "psllw         $7, %%mm0 \n"
 104         "pxor       %%mm2, %%mm2 \n"
 105         "psubw      %%mm0, %%mm1 \n" // delta = dc - pix
 106         "psubw      %%mm1, %%mm2 \n"
 107         "pmaxsw     %%mm1, %%mm2 \n"
 108         "pmulhuw    %%mm5, %%mm2 \n" // m = abs(delta) * thresh >> 16
 109         "psubw      %%mm6, %%mm2 \n"
 110         "pminsw     %%mm7, %%mm2 \n" // m = -max(0, 127-m)
 111         "pmullw     %%mm2, %%mm2 \n"
 112         "paddw      %%mm4, %%mm0 \n" // pix += dither
 113         "pmulhw     %%mm2, %%mm1 \n"
 114         "psllw         $2, %%mm1 \n" // m = m*m*delta >> 14
 115         "paddw      %%mm1, %%mm0 \n" // pix += m
 116         "psraw         $7, %%mm0 \n"
 117         "packuswb   %%mm0, %%mm0 \n"
 118         "movd       %%mm0, (%1,%0) \n" // dst = clip(pix>>7)
 119         "add           $4, %0 \n"
 120         "jl 1b \n"
 121         "emms \n"
 122         :"+r"(x)
 123         :"r"(dst+width), "r"(src+width), "r"(dc+width/2),
 124          "rm"(thresh), "m"(*dithers), "m"(*pw_7f)
 125         :"mm0", "mm1", "mm2", "mm4", "mm5", "mm6", "memory"
 126     );
 127 }
 128 #endif
 129
 130 #if HAVE_SSSE3
 131 VLC_SSE
 132 static void filter_line_ssse3(uint8_t *dst, uint8_t *src, uint16_t *dc,
 133                               int width, int thresh, const uint16_t *dithers)
 134 {
 135     intptr_t x;
 136     if (width&7) {
 137         // could be 10% faster if I somehow eliminated this
 138         x = width&~7;
 139         filter_line_c(dst+x, src+x, dc+x/2, width-x, thresh, dithers);
 140         width = x;
 141     }
 142     x = -width;
 143     __asm__ volatile(
 144         "movd           %4, %%xmm5 \n"
 145         "pxor       %%xmm7, %%xmm7 \n"
 146         "pshuflw $0,%%xmm5, %%xmm5 \n"
 147         "movdqa         %6, %%xmm6 \n"
 148         "punpcklqdq %%xmm5, %%xmm5 \n"
 149         "movdqa         %5, %%xmm4 \n"
 150         "1: \n"
 151         "movq      (%2,%0), %%xmm0 \n"
 152         "movq      (%3,%0), %%xmm1 \n"
 153         "punpcklbw  %%xmm7, %%xmm0 \n"
 154         "punpcklwd  %%xmm1, %%xmm1 \n"
 155         "psllw          $7, %%xmm0 \n"
 156         "psubw      %%xmm0, %%xmm1 \n" // delta = dc - pix
 157         "pabsw      %%xmm1, %%xmm2 \n"
 158         "pmulhuw    %%xmm5, %%xmm2 \n" // m = abs(delta) * thresh >> 16
 159         "psubw      %%xmm6, %%xmm2 \n"
 160         "pminsw     %%xmm7, %%xmm2 \n" // m = -max(0, 127-m)
 161         "pmullw     %%xmm2, %%xmm2 \n"
 162         "psllw          $1, %%xmm2 \n"
 163         "paddw      %%xmm4, %%xmm0 \n" // pix += dither
 164         "pmulhrsw   %%xmm2, %%xmm1 \n" // m = m*m*delta >> 14
 165         "paddw      %%xmm1, %%xmm0 \n" // pix += m
 166         "psraw          $7, %%xmm0 \n"
 167         "packuswb   %%xmm0, %%xmm0 \n"
 168         "movq       %%xmm0, (%1,%0) \n" // dst = clip(pix>>7)
 169         "add            $8, %0 \n"
 170         "jl 1b \n"
 171         :"+&r"(x)
 172         :"r"(dst+width), "r"(src+width), "r"(dc+width/2),
 173          "rm"(thresh), "m"(*dithers), "m"(*pw_7f)
 174         :"xmm0", "xmm1", "xmm2", "xmm4", "xmm5", "xmm6", "xmm7", "memory"
 175     );
 176 }
 177 #endif // HAVE_SSSE3
 178
 179 #if HAVE_SSE2 && HAVE_6REGS
 180 #define BLURV(load)\
 181     intptr_t x = -2*width;\
 182     __asm__ volatile(\
 183         "movdqa %6, %%xmm7 \n"\
 184         "1: \n"\
 185         load"   (%4,%0), %%xmm0 \n"\
 186         load"   (%5,%0), %%xmm1 \n"\
 187         "movdqa  %%xmm0, %%xmm2 \n"\
 188         "movdqa  %%xmm1, %%xmm3 \n"\
 189         "psrlw       $8, %%xmm0 \n"\
 190         "psrlw       $8, %%xmm1 \n"\
 191         "pand    %%xmm7, %%xmm2 \n"\
 192         "pand    %%xmm7, %%xmm3 \n"\
 193         "paddw   %%xmm1, %%xmm0 \n"\
 194         "paddw   %%xmm3, %%xmm2 \n"\
 195         "paddw   %%xmm2, %%xmm0 \n"\
 196         "paddw  (%2,%0), %%xmm0 \n"\
 197         "movdqa (%1,%0), %%xmm1 \n"\
 198         "movdqa  %%xmm0, (%1,%0) \n"\
 199         "psubw   %%xmm1, %%xmm0 \n"\
 200         "movdqa  %%xmm0, (%3,%0) \n"\
 201         "add        $16, %0 \n"\
 202         "jl 1b \n"\
 203         :"+&r"(x)\
 204         :"r"(buf+width),\
 205          "r"(buf1+width),\
 206          "r"(dc+width),\
 207          "r"(src+width*2),\
 208          "r"(src+width*2+sstride),\
 209          "m"(*pw_ff)\
 210         :"xmm0", "xmm1", "xmm2", "xmm3", "xmm7", "memory"\
 211     );
 212
 213 VLC_SSE
 214 static void blur_line_sse2(uint16_t *dc, uint16_t *buf, uint16_t *buf1,
 215                            uint8_t *src, int sstride, int width)
 216 {
 217     if (((intptr_t)src|sstride)&15) {
 218         BLURV("movdqu");
 219     } else {
 220         BLURV("movdqa");
 221     }
 222 }
 223 #endif // HAVE_6REGS && HAVE_SSE2
 224
 225 static void filter_plane(struct vf_priv_s *ctx, uint8_t *dst, uint8_t *src,
 226                          int width, int height, int dstride, int sstride, int r)
 227 {
 228     int bstride = ((width+15)&~15)/2;
 229     int y;
 230     uint32_t dc_factor = (1<<21)/(r*r);
 231     uint16_t *dc = ctx->buf+16;
 232     uint16_t *buf = ctx->buf+bstride+32;
 233     int thresh = ctx->thresh;
 234
 235     memset(dc, 0, (bstride+16)*sizeof(*buf));
 236     for (y=0; y<r; y++)
 237         ctx->blur_line(dc, buf+y*bstride, buf+(y-1)*bstride, src+2*y*sstride, sstride, width/2);
 238     for (;;) {
 239         if (y < height-r) {
 240             int mod = ((y+r)/2)%r;
 241             uint16_t *buf0 = buf+mod*bstride;
 242             uint16_t *buf1 = buf+(mod?mod-1:r-1)*bstride;
 243             int x, v;
 244             ctx->blur_line(dc, buf0, buf1, src+(y+r)*sstride, sstride, width/2);
 245             for (x=v=0; x<r; x++)
 246                 v += dc[x];
 247             for (; x<width/2; x++) {
 248                 v += dc[x] - dc[x-r];
 249                 dc[x-r] = v * dc_factor >> 16;
 250             }
 251             for (; x<(width+r+1)/2; x++)
 252                 dc[x-r] = v * dc_factor >> 16;
 253             for (x=-r/2; x<0; x++)
 254                 dc[x] = dc[0];
 255         }
 256         if (y == r) {
 257             for (y=0; y<r; y++)
 258                 ctx->filter_line(dst+y*dstride, src+y*sstride, dc-r/2, width, thresh, dither[y&7]);
 259         }
 260         ctx->filter_line(dst+y*dstride, src+y*sstride, dc-r/2, width, thresh, dither[y&7]);
 261         if (++y >= height) break;
 262         ctx->filter_line(dst+y*dstride, src+y*sstride, dc-r/2, width, thresh, dither[y&7]);
 263         if (++y >= height) break;
 264     }
 265 }
 266