[babl] fast-float: use higher accuracy LUTs and rework some alpha handling

[Øyvind Kolås <ok src gnome org>]

commit 054c9e4aaafff5a79b64cd036d960c712f27adce
Author: Øyvind Kolås <pippin gimp org>
Date:   Wed Nov 16 22:39:17 2016 +0100

    fast-float: use higher accuracy LUTs and rework some alpha handling

 extensions/fast-float.c |   37 ++++++++++++++++++++++++++++---------
 extensions/float.c      |   42 +++++++++++++++++++++---------------------
 2 files changed, 49 insertions(+), 30 deletions(-)
---
diff --git a/extensions/fast-float.c b/extensions/fast-float.c
index 9103bc9..7393645 100644
--- a/extensions/fast-float.c
+++ b/extensions/fast-float.c
@@ -222,7 +222,7 @@ static BablLookup *fast_pow = NULL;
 
 static inline float core_lookup (float val, void *userdata)
 {
-  return linear_to_gamma_2_2 (val);
+  return babl_linear_to_gamma_2_2f (val);
 }
 
 static float
@@ -236,7 +236,7 @@ static BablLookup *fast_rpow = NULL;
 
 static inline float core_rlookup (float val, void *userdata)
 {
-  return gamma_2_2_to_linear (val);
+  return babl_gamma_2_2_to_linearf (val);
 }
 
 static float
@@ -265,10 +265,29 @@ conv_rgbaF_linear_rgbAF_gamma (unsigned char *src,
    while (n--)
      {
        float alpha = fsrc[3];
-       *fdst++ = linear_to_gamma_2_2_lut (*fsrc++) * alpha;
-       *fdst++ = linear_to_gamma_2_2_lut (*fsrc++) * alpha;
-       *fdst++ = linear_to_gamma_2_2_lut (*fsrc++) * alpha;
-       *fdst++ = *fsrc++;
+       if (alpha == 1.0)
+       {
+         *fdst++ = linear_to_gamma_2_2_lut (*fsrc++);
+         *fdst++ = linear_to_gamma_2_2_lut (*fsrc++);
+         *fdst++ = linear_to_gamma_2_2_lut (*fsrc++);
+         *fdst++ = *fsrc++;
+       }
+       else if (alpha == 0.0)
+       {
+         *fdst++ = 0.0;
+         *fdst++ = 0.0;
+         *fdst++ = 0.0;
+         *fdst++ = 0.0;
+         fsrc+=4;
+       }
+       else
+       {
+         *fdst++ = linear_to_gamma_2_2_lut (*fsrc++) * alpha;
+         *fdst++ = linear_to_gamma_2_2_lut (*fsrc++) * alpha;
+         *fdst++ = linear_to_gamma_2_2_lut (*fsrc++) * alpha;
+         *fdst++ = alpha;
+         fsrc++;
+       }
      }
   return samples;
 }
@@ -446,10 +465,10 @@ init (void)
     float f;
     float a;
 
-    fast_pow = babl_lookup_new (core_lookup, NULL, 0.0, 1.0,   0.0001);
-    fast_rpow = babl_lookup_new (core_rlookup, NULL, 0.0, 1.0, 0.0001);
+    fast_pow = babl_lookup_new (core_lookup, NULL, 0.0, 1.0,   0.00005);
+    fast_rpow = babl_lookup_new (core_rlookup, NULL, 0.0, 1.0, 0.00005);
 
-    for (f = 0.0; f < 1.0; f+= 0.000012)
+    for (f = 0.0; f < 1.0; f+= 0.0000001)
       {
         a = linear_to_gamma_2_2_lut (f);
         a = gamma_2_2_to_linear_lut (f);
diff --git a/extensions/float.c b/extensions/float.c
index 5cbbeb6..9460804 100644
--- a/extensions/float.c
+++ b/extensions/float.c
@@ -41,9 +41,9 @@ conv_rgbaF_linear_rgbAF_gamma (unsigned char *src,
    while (n--)
      {
        float alpha = fsrc[3];
-       *fdst++ = babl_linear_to_gamma_2_2 (*fsrc++) * alpha;
-       *fdst++ = babl_linear_to_gamma_2_2 (*fsrc++) * alpha;
-       *fdst++ = babl_linear_to_gamma_2_2 (*fsrc++) * alpha;
+       *fdst++ = babl_linear_to_gamma_2_2f (*fsrc++) * alpha;
+       *fdst++ = babl_linear_to_gamma_2_2f (*fsrc++) * alpha;
+       *fdst++ = babl_linear_to_gamma_2_2f (*fsrc++) * alpha;
        *fdst++ = *fsrc++;
      }
   return samples;
@@ -71,17 +71,17 @@ conv_rgbAF_linear_rgbAF_gamma (unsigned char *src,
          }
        else if (alpha >= 1.0)
          {
-           *fdst++ = babl_linear_to_gamma_2_2 (*fsrc++);
-           *fdst++ = babl_linear_to_gamma_2_2 (*fsrc++);
-           *fdst++ = babl_linear_to_gamma_2_2 (*fsrc++);
+           *fdst++ = babl_linear_to_gamma_2_2f (*fsrc++);
+           *fdst++ = babl_linear_to_gamma_2_2f (*fsrc++);
+           *fdst++ = babl_linear_to_gamma_2_2f (*fsrc++);
            *fdst++ = *fsrc++;
          }
        else
          {
            float alpha_recip = 1.0 / alpha;
-           *fdst++ = babl_linear_to_gamma_2_2 (*fsrc++ * alpha_recip) * alpha;
-           *fdst++ = babl_linear_to_gamma_2_2 (*fsrc++ * alpha_recip) * alpha;
-           *fdst++ = babl_linear_to_gamma_2_2 (*fsrc++ * alpha_recip) * alpha;
+           *fdst++ = babl_linear_to_gamma_2_2f (*fsrc++ * alpha_recip) * alpha;
+           *fdst++ = babl_linear_to_gamma_2_2f (*fsrc++ * alpha_recip) * alpha;
+           *fdst++ = babl_linear_to_gamma_2_2f (*fsrc++ * alpha_recip) * alpha;
            *fdst++ = *fsrc++;
          }
      }
@@ -99,9 +99,9 @@ conv_rgbaF_linear_rgbaF_gamma (unsigned char *src,
 
    while (n--)
      {
-       *fdst++ = babl_linear_to_gamma_2_2 (*fsrc++);
-       *fdst++ = babl_linear_to_gamma_2_2 (*fsrc++);
-       *fdst++ = babl_linear_to_gamma_2_2 (*fsrc++);
+       *fdst++ = babl_linear_to_gamma_2_2f (*fsrc++);
+       *fdst++ = babl_linear_to_gamma_2_2f (*fsrc++);
+       *fdst++ = babl_linear_to_gamma_2_2f (*fsrc++);
        *fdst++ = *fsrc++;
      }
   return samples;
@@ -118,9 +118,9 @@ conv_rgbF_linear_rgbF_gamma (unsigned char *src,
 
    while (n--)
      {
-       *fdst++ = babl_linear_to_gamma_2_2 (*fsrc++);
-       *fdst++ = babl_linear_to_gamma_2_2 (*fsrc++);
-       *fdst++ = babl_linear_to_gamma_2_2 (*fsrc++);
+       *fdst++ = babl_linear_to_gamma_2_2f (*fsrc++);
+       *fdst++ = babl_linear_to_gamma_2_2f (*fsrc++);
+       *fdst++ = babl_linear_to_gamma_2_2f (*fsrc++);
      }
   return samples;
 }
@@ -137,9 +137,9 @@ conv_rgbaF_gamma_rgbaF_linear (unsigned char *src,
 
    while (n--)
      {
-       *fdst++ = babl_gamma_2_2_to_linear (*fsrc++);
-       *fdst++ = babl_gamma_2_2_to_linear (*fsrc++);
-       *fdst++ = babl_gamma_2_2_to_linear (*fsrc++);
+       *fdst++ = babl_gamma_2_2_to_linearf (*fsrc++);
+       *fdst++ = babl_gamma_2_2_to_linearf (*fsrc++);
+       *fdst++ = babl_gamma_2_2_to_linearf (*fsrc++);
        *fdst++ = *fsrc++;
      }
   return samples;
@@ -156,9 +156,9 @@ conv_rgbF_gamma_rgbF_linear (unsigned char *src,
 
    while (n--)
      {
-       *fdst++ = babl_gamma_2_2_to_linear (*fsrc++);
-       *fdst++ = babl_gamma_2_2_to_linear (*fsrc++);
-       *fdst++ = babl_gamma_2_2_to_linear (*fsrc++);
+       *fdst++ = babl_gamma_2_2_to_linearf (*fsrc++);
+       *fdst++ = babl_gamma_2_2_to_linearf (*fsrc++);
+       *fdst++ = babl_gamma_2_2_to_linearf (*fsrc++);
      }
   return samples;
 }