src/libnr/nr-compose-transform.cpp

   1 #define __NR_COMPOSE_TRANSFORM_C__
   2
   3 /*
   4  * Pixel buffer rendering library
   5  *
   6  * Authors:
   7  *   Lauris Kaplinski <lauris@kaplinski.com>
   8  *
   9  * This code is in public domain
  10  */
  11
  12 #ifdef HAVE_CONFIG_H
  13 # include "config.h"
  14 #endif
  15
  16 #include "nr-pixops.h"
  17 #include "nr-matrix.h"
  18
  19
  20 #ifdef WITH_MMX
  21 #ifdef __cplusplus
  22 extern "C" {
  23 #endif /* __cplusplus */
  24 /* fixme: */
  25 int nr_have_mmx (void);
  26 void nr_mmx_R8G8B8A8_P_R8G8B8A8_P_R8G8B8A8_N_TRANSFORM_0 (unsigned char *px, int w, int h, int rs,
  27                                                           const unsigned char *spx, int sw, int sh, int srs,
  28                                                           const long *FFd2s, unsigned int alpha);
  29 void nr_mmx_R8G8B8A8_P_R8G8B8A8_P_R8G8B8A8_N_TRANSFORM_n (unsigned char *px, int w, int h, int rs,
  30                                                           const unsigned char *spx, int sw, int sh, int srs,
  31                                                           const long *FFd2s, const long *FF_S, unsigned int alpha, int dbits);
  32 #define NR_PIXOPS_MMX (1 && nr_have_mmx ())
  33 #ifdef __cplusplus
  34 }
  35 #endif //__cplusplus
  36 #endif
  37
  38 /* fixme: Implement missing (Lauris) */
  39 /* fixme: PREMUL colors before calculating average (Lauris) */
  40
  41 /* Fixed point precision */
  42 #define FBITS 12
  43 #define FBITS_HP 18 // In some places we need a higher precision
  44
  45 void nr_R8G8B8A8_N_EMPTY_R8G8B8A8_N_TRANSFORM (unsigned char *px, int w, int h, int rs,
  46                                                const unsigned char *spx, int sw, int sh, int srs,
  47                                                const NR::Matrix &d2s, unsigned int alpha, int xd, int yd);
  48 void nr_R8G8B8A8_N_EMPTY_R8G8B8A8_P_TRANSFORM (unsigned char *px, int w, int h, int rs,
  49                                                const unsigned char *spx, int sw, int sh, int srs,
  50                                                const NR::Matrix &d2s, unsigned int alpha, int xd, int yd);
  51 void nr_R8G8B8A8_P_EMPTY_R8G8B8A8_N_TRANSFORM (unsigned char *px, int w, int h, int rs,
  52                                                const unsigned char *spx, int sw, int sh, int srs,
  53                                                const NR::Matrix &d2s, unsigned int alpha, int xd, int yd);
  54 void nr_R8G8B8A8_P_EMPTY_R8G8B8A8_P_TRANSFORM (unsigned char *px, int w, int h, int rs,
  55                                                const unsigned char *spx, int sw, int sh, int srs,
  56                                                const NR::Matrix &d2s, unsigned int alpha, int xd, int yd);
  57
  58 void
  59 nr_R8G8B8A8_N_R8G8B8A8_N_R8G8B8A8_N_TRANSFORM (unsigned char *px, int w, int h, int rs,
  60                                                const unsigned char *spx, int sw, int sh, int srs,
  61                                                const NR::Matrix &d2s, unsigned int alpha, int xd, int yd)
  62 {
  63         int xsize, ysize, size, dbits;
  64         long FFs_x_x, FFs_x_y, FFs_y_x, FFs_y_y, FFs__x, FFs__y;
  65         long FFs_x_x_S, FFs_x_y_S, FFs_y_x_S, FFs_y_y_S;
  66         /* Subpixel positions */
  67         int FF_sx_S[256];
  68         int FF_sy_S[256];
  69         unsigned char *d0;
  70         int FFsx0, FFsy0;
  71         int x, y;
  72
  73         if (alpha == 0) return;
  74
  75         xsize = (1 << xd);
  76         ysize = (1 << yd);
  77         size = xsize * ysize;
  78         dbits = xd + yd;
  79
  80         /* Set up fixed point matrix */
  81         FFs_x_x = (long) (d2s[0] * (1 << FBITS) + 0.5);
  82         FFs_x_y = (long) (d2s[1] * (1 << FBITS) + 0.5);
  83         FFs_y_x = (long) (d2s[2] * (1 << FBITS) + 0.5);
  84         FFs_y_y = (long) (d2s[3] * (1 << FBITS) + 0.5);
  85         FFs__x = (long) (d2s[4] * (1 << FBITS) + 0.5);
  86         FFs__y = (long) (d2s[5] * (1 << FBITS) + 0.5);
  87
  88         FFs_x_x_S = FFs_x_x >> xd;
  89         FFs_x_y_S = FFs_x_y >> xd;
  90         FFs_y_x_S = FFs_y_x >> yd;
  91         FFs_y_y_S = FFs_y_y >> yd;
  92
  93         /* Set up subpixel matrix */
  94         /* fixme: We can calculate that in floating point (Lauris) */
  95         for (y = 0; y < ysize; y++) {
  96                 for (x = 0; x < xsize; x++) {
  97                         FF_sx_S[y * xsize + x] = FFs_x_x_S * x + FFs_y_x_S * y;
  98                         FF_sy_S[y * xsize + x] = FFs_x_y_S * x + FFs_y_y_S * y;
  99                 }
 100         }
 101
 102         d0 = px;
 103         FFsx0 = FFs__x;
 104         FFsy0 = FFs__y;
 105
 106         for (y = 0; y < h; y++) {
 107                 unsigned char *d;
 108                 long FFsx, FFsy;
 109                 d = d0;
 110                 FFsx = FFsx0;
 111                 FFsy = FFsy0;
 112                 for (x = 0; x < w; x++) {
 113                         unsigned int r, g, b, a;
 114                         long sx, sy;
 115                         int i;
 116                         r = g = b = a = 0;
 117                         for (i = 0; i < size; i++) {
 118                                 sx = (FFsx + FF_sx_S[i]) >> FBITS;
 119                                 if ((sx >= 0) && (sx < sw)) {
 120                                         sy = (FFsy + FF_sy_S[i]) >> FBITS;
 121                                         if ((sy >= 0) && (sy < sh)) {
 122                                                 const unsigned char *s;
 123                                                 unsigned int ca;
 124                                                 s = spx + sy * srs + sx * 4;
 125                                                 ca = NR_PREMUL_112 (s[3], alpha);
 126                                                 r += NR_PREMUL_121 (s[0], ca);
 127                                                 g += NR_PREMUL_121 (s[1], ca);
 128                                                 b += NR_PREMUL_121 (s[2], ca);
 129                                                 a += NR_NORMALIZE_21(ca);
 130                                         }
 131                                 }
 132                         }
 133                         a >>= dbits;
 134                         if (a != 0) {
 135                                 r = r >> dbits;
 136                                 g = g >> dbits;
 137                                 b = b >> dbits;
 138                                 if (a == 255) {
 139                                         /* Transparent BG, premul src */
 140                                         d[0] = r;
 141                                         d[1] = g;
 142                                         d[2] = b;
 143                                         d[3] = a;
 144                                 } else {
 145                                         unsigned int ca;
 146                                         /* Full composition */
 147                                         ca = NR_COMPOSEA_112(a, d[3]);
 148                                         d[0] = NR_COMPOSENNN_111121 (r, a, d[0], d[3], ca);
 149                                         d[1] = NR_COMPOSENNN_111121 (g, a, d[1], d[3], ca);
 150                                         d[2] = NR_COMPOSENNN_111121 (b, a, d[2], d[3], ca);
 151                                         d[3] = NR_NORMALIZE_21(ca);
 152                                 }
 153                         }
 154                         /* Advance pointers */
 155                         FFsx += FFs_x_x;
 156                         FFsy += FFs_x_y;
 157                         d += 4;
 158                 }
 159                 FFsx0 += FFs_y_x;
 160                 FFsy0 += FFs_y_y;
 161                 d0 += rs;
 162         }
 163 }
 164
 165 void nr_R8G8B8A8_N_R8G8B8A8_N_R8G8B8A8_P_TRANSFORM (unsigned char *px, int w, int h, int rs,
 166                                                     const unsigned char *spx, int sw, int sh, int srs,
 167                                                     const NR::Matrix &d2s, unsigned int alpha, int xd, int yd);
 168
 169 static void
 170 nr_R8G8B8A8_P_R8G8B8A8_P_R8G8B8A8_N_TRANSFORM_0 (unsigned char *px, int w, int h, int rs,
 171                                                  const unsigned char *spx, int sw, int sh, int srs,
 172                                                  const long long *FFd2s, unsigned int alpha)
 173 {
 174     unsigned char *d0;
 175         long long FFsx0, FFsy0;
 176         int x, y;
 177
 178         d0 = px;
 179         FFsx0 = FFd2s[4];
 180         FFsy0 = FFd2s[5];
 181
 182         for (y = 0; y < h; y++) {
 183                 unsigned char *d;
 184                 long long FFsx, FFsy;
 185                 d = d0;
 186                 FFsx = FFsx0;
 187                 FFsy = FFsy0;
 188                 for (x = 0; x < w; x++) {
 189                         long sx, sy;
 190                         sx = long(FFsx >> FBITS_HP);
 191                         if ((sx >= 0) && (sx < sw)) {
 192                                 sy = long(FFsy >> FBITS_HP);
 193                                 if ((sy >= 0) && (sy < sh)) {
 194                                         const unsigned char *s;
 195                                         unsigned int a;
 196                                         s = spx + sy * srs + sx * 4;
 197                                         a = NR_PREMUL_112 (s[3], alpha);
 198                                         if (a != 0) {
 199                                                 if ((a == 255*255) || (d[3] == 0)) {
 200                                                         /* Transparent BG, premul src */
 201                                                         d[0] = NR_PREMUL_121 (s[0], a);
 202                                                         d[1] = NR_PREMUL_121 (s[1], a);
 203                                                         d[2] = NR_PREMUL_121 (s[2], a);
 204                                                         d[3] = NR_NORMALIZE_21(a);
 205                                                 } else {
 206                                                         d[0] = NR_COMPOSENPP_1211 (s[0], a, d[0]);
 207                                                         d[1] = NR_COMPOSENPP_1211 (s[1], a, d[1]);
 208                                                         d[2] = NR_COMPOSENPP_1211 (s[2], a, d[2]);
 209                                                         d[3] = NR_COMPOSEA_211(a, d[3]);
 210                                                 }
 211                                         }
 212                                 }
 213                         }
 214                         /* Advance pointers */
 215                         FFsx += FFd2s[0];
 216                         FFsy += FFd2s[1];
 217                         d += 4;
 218                 }
 219                 FFsx0 += FFd2s[2];
 220                 FFsy0 += FFd2s[3];
 221                 d0 += rs;
 222         }
 223 }
 224
 225 static void
 226 nr_R8G8B8A8_P_R8G8B8A8_P_R8G8B8A8_N_TRANSFORM_n (unsigned char *px, int w, int h, int rs,
 227                                                  const unsigned char *spx, int sw, int sh, int srs,
 228                                                  const long long *FFd2s, const long *FF_S, unsigned int alpha, int dbits)
 229 {
 230         int size;
 231         unsigned char *d0;
 232         long long FFsx0, FFsy0;
 233         int x, y;
 234
 235         size = (1 << dbits);
 236     unsigned alpha_rounding_fix = size * 255;
 237     unsigned rgb_rounding_fix = size * (255 * 256);
 238     if (alpha > 127) ++alpha;
 239
 240         d0 = px;
 241         FFsx0 = FFd2s[4];
 242         FFsy0 = FFd2s[5];
 243
 244         for (y = 0; y < h; y++) {
 245                 unsigned char *d;
 246                 long long FFsx, FFsy;
 247                 d = d0;
 248                 FFsx = FFsx0;
 249                 FFsy = FFsy0;
 250                 for (x = 0; x < w; x++) {
 251                         unsigned int r, g, b, a;
 252                         int i;
 253                         r = g = b = a = 0;
 254                         for (i = 0; i < size; i++) {
 255                                 long sx, sy;
 256                                 sx = (FFsx >> FBITS_HP) + (FF_S[2 * i] >> FBITS);
 257                                 if ((sx >= 0) && (sx < sw)) {
 258                                         sy = (FFsy >> FBITS_HP) + (FF_S[2 * i + 1] >> FBITS);
 259                                         if ((sy >= 0) && (sy < sh)) {
 260                                                 const unsigned char *s;
 261                                                 unsigned int ca;
 262                                                 s = spx + sy * srs + sx * 4;
 263                                                 ca = NR_PREMUL_112(s[3], alpha);
 264                                                 r += NR_PREMUL_123(s[0], ca);
 265                                                 g += NR_PREMUL_123(s[1], ca);
 266                                                 b += NR_PREMUL_123(s[2], ca);
 267                                                 a += ca;
 268                                         }
 269                                 }
 270                         }
 271                         a = (a + alpha_rounding_fix) >> (8 + dbits);
 272                         if (a != 0) {
 273                                 r = (r + rgb_rounding_fix) >> (16 + dbits);
 274                                 g = (g + rgb_rounding_fix) >> (16 + dbits);
 275                                 b = (b + rgb_rounding_fix) >> (16 + dbits);
 276                                 if ((a == 255) || (d[3] == 0)) {
 277                                         /* Transparent BG, premul src */
 278                                         d[0] = r;
 279                                         d[1] = g;
 280                                         d[2] = b;
 281                                         d[3] = a;
 282                                 } else {
 283                                         d[0] = NR_COMPOSEPPP_1111 (r, a, d[0]);
 284                                         d[1] = NR_COMPOSEPPP_1111 (g, a, d[1]);
 285                                         d[2] = NR_COMPOSEPPP_1111 (b, a, d[2]);
 286                                         d[3] = NR_COMPOSEA_111(a, d[3]);
 287                                 }
 288                         }
 289                         /* Advance pointers */
 290                         FFsx += FFd2s[0];
 291                         FFsy += FFd2s[1];
 292                         d += 4;
 293                 }
 294                 FFsx0 += FFd2s[2];
 295                 FFsy0 += FFd2s[3];
 296                 d0 += rs;
 297         }
 298 }
 299
 300 void nr_R8G8B8A8_P_R8G8B8A8_P_R8G8B8A8_N_TRANSFORM (unsigned char *px, int w, int h, int rs,
 301                                                     const unsigned char *spx, int sw, int sh, int srs,
 302                                                     const NR::Matrix &d2s, unsigned int alpha, int xd, int yd)
 303 {
 304         int dbits;
 305         long FFd2s[6];
 306         long long FFd2s_HP[6]; // with higher precision
 307         int i;
 308
 309         if (alpha == 0) return;
 310
 311         dbits = xd + yd;
 312
 313         for (i = 0; i < 6; i++) {
 314                 FFd2s[i] = (long) (d2s[i] * (1 << FBITS) + 0.5);
 315                 FFd2s_HP[i] = (long long) (d2s[i] * (1 << FBITS_HP) + 0.5);;
 316         }
 317
 318         if (dbits == 0) {
 319 #ifdef WITH_MMX
 320                 if (NR_PIXOPS_MMX) {
 321                         /* WARNING: MMX composer REQUIRES w > 0 and h > 0 */
 322                         nr_mmx_R8G8B8A8_P_R8G8B8A8_P_R8G8B8A8_N_TRANSFORM_0 (px, w, h, rs, spx, sw, sh, srs, FFd2s, alpha);
 323                         return;
 324                 }
 325 #endif
 326                 nr_R8G8B8A8_P_R8G8B8A8_P_R8G8B8A8_N_TRANSFORM_0 (px, w, h, rs, spx, sw, sh, srs, FFd2s_HP, alpha);
 327         } else {
 328                 int xsize, ysize;
 329                 long FFs_x_x_S, FFs_x_y_S, FFs_y_x_S, FFs_y_y_S;
 330                 long FF_S[2 * 256];
 331                 int x, y;
 332
 333                 xsize = (1 << xd);
 334                 ysize = (1 << yd);
 335
 336                 FFs_x_x_S = FFd2s[0] >> xd;
 337                 FFs_x_y_S = FFd2s[1] >> xd;
 338                 FFs_y_x_S = FFd2s[2] >> yd;
 339                 FFs_y_y_S = FFd2s[3] >> yd;
 340
 341                 /* Set up subpixel matrix */
 342                 /* fixme: We can calculate that in floating point (Lauris) */
 343                 for (y = 0; y < ysize; y++) {
 344                         for (x = 0; x < xsize; x++) {
 345                                 FF_S[2 * (y * xsize + x)] = FFs_x_x_S * x + FFs_y_x_S * y;
 346                                 FF_S[2 * (y * xsize + x) + 1] = FFs_x_y_S * x + FFs_y_y_S * y;
 347                         }
 348                 }
 349
 350 #ifdef WITH_MMX
 351                 if (NR_PIXOPS_MMX) {
 352                         /* WARNING: MMX composer REQUIRES w > 0 and h > 0 */
 353                         nr_mmx_R8G8B8A8_P_R8G8B8A8_P_R8G8B8A8_N_TRANSFORM_n (px, w, h, rs, spx, sw, sh, srs, FFd2s, FF_S, alpha, dbits);
 354                         return;
 355                 }
 356 #endif
 357                 nr_R8G8B8A8_P_R8G8B8A8_P_R8G8B8A8_N_TRANSFORM_n (px, w, h, rs, spx, sw, sh, srs, FFd2s_HP, FF_S, alpha, dbits);
 358         }
 359 }
 360
 361 void nr_R8G8B8A8_P_R8G8B8A8_P_R8G8B8A8_P_TRANSFORM (unsigned char *px, int w, int h, int rs,
 362                                                     const unsigned char *spx, int sw, int sh, int srs,
 363                                                     const NR::Matrix &d2s, unsigned int alpha, int xd, int yd);