babl: add lut for some u8 conversions

The fish path now marks created fishes that shall be monitored for
usage and have full 24bit LUTs generated on demand when limits are
exceeded. This 3-4x increases the throughput.

diff --git a/babl/babl-fish-path.c b/babl/babl-fish-path.c
index f005a86241eb82b08032debff2b8d091e1bd990e..99b8380ec493e68c7287bd427cfc1e0f3346988f 100644 (file)
--- a/babl/babl-fish-path.c
+++ b/babl/babl-fish-path.c
@@ -552,6 +552,7 @@ babl_fish_path2 (const Babl *source,
     static const Babl *run_once[512]={NULL};
     int i;
     int done = 0;
+
     for (i = 0; run_once[i]; i++)
     {
       if (run_once[i] == source->format.space)
@@ -582,7 +583,6 @@ babl_fish_path2 (const Babl *source,
     {
       babl_conversion_class_for_each (show_item, (void*)source->format.space);
     }
-
   }
 
   babl = babl_calloc (1, sizeof (BablFishPath) +
@@ -601,6 +601,7 @@ babl_fish_path2 (const Babl *source,
   babl->fish_path.conversion_list = babl_list_init_with_size (BABL_HARD_MAX_PATH_LENGTH);
 
 
+
   {
     PathContext pc;
     int start_depth = max_path_length ();
@@ -652,6 +653,14 @@ babl_fish_path2 (const Babl *source,
     }
 
   _babl_fish_prepare_bpp (babl);
+  if (source->format.space != destination->format.space &&
+      babl->fish_path.source_bpp == babl->fish_path.dest_bpp &&
+      babl->fish_path.source_bpp == 4)
+  {
+     if ((source->format.model->flags & BABL_MODEL_FLAG_ASSOCIATED)==0)
+       babl->fish_path.is_u8_color_conv = 1;
+  }
+
   _babl_fish_rig_dispatch (babl);
   /* Since there is not an already registered instance by the required
    * name, inserting newly created class into database.
@@ -696,7 +705,6 @@ babl_fish_path (const Babl *source,
   return babl_fish_path2 (source, destination, 0.0);
 }
 
-
 static void
 babl_fish_path_process (const Babl *babl,
                         const char *source,
@@ -704,6 +712,37 @@ babl_fish_path_process (const Babl *babl,
                         long        n,
                         void       *data)
 {
+  if (babl->fish_path.is_u8_color_conv)
+  {
+     uint32_t *lut = (uint32_t*)babl->fish_path.u8_lut;
+     ((Babl*)babl)->fish.pixels += n;
+     if (!lut && babl->fish.pixels > 256 * 256)
+     {
+       ((Babl*)babl)->fish_path.u8_lut = malloc (256 * 256 * 256 * 4);
+       lut = (uint32_t*)babl->fish_path.u8_lut;
+       for (int o = 0; o < 256 * 256 * 256; o++)
+         lut[o] = o | 0xff000000;
+       process_conversion_path (babl->fish_path.conversion_list,
+                                lut,
+                                babl->fish_path.source_bpp,
+                                lut,
+                                babl->fish_path.dest_bpp,
+                                256*256*256);
+     }
+     if (lut)
+     {
+        uint32_t *src = (uint32_t*)source;
+        uint32_t *dst = (uint32_t*)destination;
+        lut = (uint32_t*)babl->fish_path.u8_lut;
+        while (n--)
+        {
+           uint32_t col = *src++;
+           uint32_t alpha = col & 0xff000000;
+           *dst++ = lut[col & 0xffffff] | alpha;
+        }
+        return;
+     }
+  }
   process_conversion_path (babl->fish_path.conversion_list,
                            source,
                            babl->fish_path.source_bpp,

diff --git a/babl/babl-fish.h b/babl/babl-fish.h
index 35382f03f36a9732d0520812e6db5c6f832a0e4d..966fa4e32e7798a662b8e22ec0772252311c9886 100644 (file)
--- a/babl/babl-fish.h
+++ b/babl/babl-fish.h
@@ -69,6 +69,8 @@ typedef struct
   double     cost;   /* number of  ticks *10 + chain_length */
   int        source_bpp;
   int        dest_bpp;
+  unsigned int is_u8_color_conv:1; // keep track of count, and make 
+  uint8_t   *u8_lut;
   BablList  *conversion_list;
 } BablFishPath;