Use the new shader include system to avoid having to upload 256 uniform floats to two different programs on the GPU

Author: kovidgoyal

.gitattributes

@@ -5,7 +5,7 @@ kitty/key_encoding.py linguist-generated=true
 kitty/unicode-data.c linguist-generated=true
 kitty/rowcolumn-diacritics.c linguist-generated=true
 kitty/rgb.py linguist-generated=true
-kitty/srgb_gamma.c linguist-generated=true
+kitty/srgb_gamma.* linguist-generated=true
 kitty/gl-wrapper.* linguist-generated=true
 kitty/glfw-wrapper.* linguist-generated=true
 kitty/parse-graphics-command.h linguist-generated=true

gen-srgb-lut.py

@@ -2,6 +2,7 @@
 # vim:fileencoding=utf-8
 
 import os
+from functools import lru_cache
 from typing import List
 
 
@@ -12,7 +13,8 @@ def to_linear(a: float) -> float:
         return float(pow((a + 0.055) / 1.055, 2.4))
 
 
-def generate_srgb_lut(line_prefix: str = '') -> List[str]:
+@lru_cache
+def generate_srgb_lut(line_prefix: str = '    ') -> List[str]:
     values: List[str] = []
     lines: List[str] = []
 
@@ -22,26 +24,30 @@ def generate_srgb_lut(line_prefix: str = '') -> List[str]:
     for i in range(16):
         lines.append(line_prefix + ', '.join(values[i * 16:(i + 1) * 16]) + ',')
 
+    lines[-1] = lines[-1].rstrip(',')
     return lines
 
 
-def generate_srgb_gamma_c() -> str:
+def generate_srgb_gamma(declaration: str = 'static const GLfloat srgb_lut[256] = {', close: str = '};') -> str:
     lines: List[str] = []
+    a = lines.append
 
-    lines.append('// Generated by gen-srgb-lut.py DO NOT edit')
-    lines.append('#include "srgb_gamma.h"')
-    lines.append('')
-    lines.append('const GLfloat srgb_lut[256] = {')
-    lines += generate_srgb_lut('    ')
-    lines.append('};')
+    a('// Generated by gen-srgb-lut.py DO NOT edit')
+    a('')
+    a(declaration)
+    lines += generate_srgb_lut()
+    a(close)
 
     return "\n".join(lines)
 
 
 def main() -> None:
-    c = generate_srgb_gamma_c()
-    with open(os.path.join('kitty', 'srgb_gamma.c'), 'w') as f:
+    c = generate_srgb_gamma()
+    with open(os.path.join('kitty', 'srgb_gamma.h'), 'w') as f:
         f.write(f'{c}\n')
+    g = generate_srgb_gamma('const float gamma_lut[256] = float[256](', ');')
+    with open(os.path.join('kitty', 'srgb_gamma.glsl'), 'w') as f:
+        f.write(f'{g}\n')
 
 
 if __name__ == '__main__':

kitty/border_vertex.glsl

@@ -1,8 +1,8 @@
+#pragma kitty_include_shader <srgb_gamma.glsl>
 uniform uvec2 viewport;
 uniform uint colors[9];
 uniform float background_opacity;
 uniform float tint_opacity, tint_premult;
-uniform float gamma_lut[256];
 in vec4 rect;  // left, top, right, bottom
 in uint rect_color;
 out vec4 color;

kitty/cell_vertex.glsl

@@ -1,4 +1,5 @@
 #extension GL_ARB_explicit_attrib_location : require
+#pragma kitty_include_shader <srgb_gamma.glsl>
 
 #define {WHICH_PROGRAM}
 #define NOT_TRANSPARENT
@@ -30,7 +31,6 @@ uniform uint draw_bg_bitfield;
 layout(location=0) in uvec3 colors;
 layout(location=1) in uvec4 sprite_coords;
 layout(location=2) in uint is_selected;
-uniform float gamma_lut[256];
 
 
 const int fg_index_map[] = int[3](0, 1, 0);

kitty/shaders.c

@@ -593,7 +593,6 @@ set_cell_uniforms(float current_inactive_text_alpha, bool force) {
         }
         for (int i = CELL_PROGRAM; i <= CELL_FG_PROGRAM; i++) {
             bind_program(i); const CellUniforms *cu = &cell_program_layouts[i].uniforms;
-            glUniform1fv(cu->gamma_lut, arraysz(srgb_lut), srgb_lut);
             switch(i) {
                 case CELL_PROGRAM: case CELL_FG_PROGRAM:
                     glUniform1i(cu->sprites, SPRITE_MAP_UNIT);
@@ -1040,7 +1039,6 @@ draw_borders(ssize_t vao_idx, unsigned int num_border_rects, BorderRect *rect_bu
         glUniform1f(border_program_layout.uniforms.tint_opacity, tint_opacity);
         glUniform1f(border_program_layout.uniforms.tint_premult, tint_premult);
         glUniform2ui(border_program_layout.uniforms.viewport, viewport_width, viewport_height);
-        glUniform1fv(border_program_layout.uniforms.gamma_lut, 256, srgb_lut);
         if (has_bgimage(w)) {
             if (w->is_semi_transparent) { BLEND_PREMULT; }
             else { BLEND_ONTO_OPAQUE_WITH_OPAQUE_OUTPUT; }