Increase LUT_ELEM to 512

Subtly improved precision for lookup_gamut function due to larger table.
Also, as we use LUT_ELEM to be as 2^n we can redefine lookup_gamut
- removing the restriction of hue input angle
- improved performance

data/kernels/colorspace.h

@@ -965,24 +965,18 @@ static inline float soft_clip(const float x, const float soft_threshold, const f
 
 static inline float lookup_gamut(global const float *gamut_lut, const float x)
 {
-  // WARNING : x should be between [-pi ; pi ], which is the default output of atan2 anyway
-
+  // Linearly interpolate the value of the gamut LUT at the hue angle in radians.
   // convert in LUT coordinate
   const float x_test = (LUT_ELEM - 1) * (x + M_PI_F) / (2.f * M_PI_F);
 
   // find the 2 closest integer coordinates (next/previous)
-  float x_prev = floor(x_test);
-  float x_next = ceil(x_test);
+  const float x_prev = floor(x_test);
+  const float x_next = ceil(x_test);
 
   // get the 2 closest LUT elements at integer coordinates
   // cycle on the hue ring if out of bounds
-  int xi = (int)x_prev;
-  if(xi < 0) xi = LUT_ELEM - 1;
-  else if(xi > LUT_ELEM - 1) xi = 0;
-
-  int xii = (int)x_next;
-  if(xii < 0) xii = LUT_ELEM - 1;
-  else if(xii > LUT_ELEM - 1) xii = 0;
+  const int xi = ((int)x_prev) & (LUT_ELEM - 1);
+  const int xii = ((int)x_next) & (LUT_ELEM - 1);
 
   // fetch the corresponding y values
   const float y_prev = gamut_lut[xi];

data/kernels/common.h

@@ -32,7 +32,7 @@ constant sampler_t samplerA = CLK_NORMALIZED_COORDS_FALSE | CLK_ADDRESS_NONE
 #define M_PI_F           3.14159265358979323846  // should be defined by the OpenCL compiler acc. to standard
 #endif
 
-#define LUT_ELEM 360 // gamut LUT number of elements:
+#define LUT_ELEM 512 // gamut LUT number of elements:
 
 #define RED 0
 #define GREEN 1

src/common/darktable_ucs_22_helpers.h

@@ -134,21 +134,14 @@ static inline float lookup_gamut(const float gamut_lut[LUT_ELEM], const float hu
   // convert hue in LUT index coordinate
   const float x_test = (float)LUT_ELEM * (hue + M_PI_F) / (2.f * M_PI_F);
 
-  // find the 2 closest integer coordinates (next/previous)
-  float x_prev = floorf(x_test);
-  float x_next = ceilf(x_test);
+  const float x_prev = floor(x_test);
+  const float x_next = ceil(x_test);
 
   // get the 2 closest LUT elements at integer coordinates
   // cycle on the hue ring if out of bounds
-  int xi = (int)x_prev;
-  if(xi < 0) xi = LUT_ELEM - 1;
-  else if(xi > LUT_ELEM - 1) xi = 0;
-
-  int xii = (int)x_next;
-  if(xii < 0) xii = LUT_ELEM - 1;
-  else if(xii > LUT_ELEM - 1) xii = 0;
-
-  // fetch the corresponding y values
+  const int xi = (int)x_prev & (LUT_ELEM - 1);
+  const int xii = (int)x_next & (LUT_ELEM - 1);
+ // fetch the corresponding y values
   const float y_prev = gamut_lut[xi];
 
   // return y_prev if we are on the same integer LUT element or do linear interpolation

src/common/math.h

@@ -23,7 +23,7 @@
 #include <stdint.h>
 #include "common/sse.h"		// also includes darktable.h
 
-#define LUT_ELEM 360 // gamut LUT number of elements:
+#define LUT_ELEM 512 // gamut LUT number of elements:
 
 #define NORM_MIN 1.52587890625e-05f // norm can't be < to 2^(-16)