From 025dccf182d7f9bfb4f45459284d8c0c9045e291 Mon Sep 17 00:00:00 2001
From: Kovid Goyal <kovid@kovidgoyal.net>
Date: Wed, 10 Dec 2025 08:49:02 +0530
Subject: [PATCH] Micro optimization: When sending cell uniform data dont even
 map the color table region in memory unless the color table is dirty

---
 kitty/gl.c      | 10 ++++++++--
 kitty/gl.h      |  1 +
 kitty/shaders.c |  6 ++++--
 3 files changed, 13 insertions(+), 4 deletions(-)

diff --git a/kitty/gl.c b/kitty/gl.c
index 7ced321f9..9e593449d 100644
--- a/kitty/gl.c
+++ b/kitty/gl.c
@@ -391,6 +391,12 @@ map_buffer(ssize_t idx, GLenum access) {
     return ans;
 }
 
+static void*
+map_buffer_range(ssize_t idx, GLbitfield access, int offset, unsigned size) {
+    return glMapBufferRange(buffers[idx].usage, offset, size, access);
+}
+
+
 static void
 unmap_buffer(ssize_t idx) {
     glUnmapBuffer(buffers[idx].usage);
@@ -500,10 +506,10 @@ alloc_vao_buffer(ssize_t vao_idx, GLsizeiptr size, size_t bufnum, GLenum usage)
 }
 
 void*
-map_vao_buffer(ssize_t vao_idx, size_t bufnum, GLenum access) {
+map_vao_buffer_for_write_only(ssize_t vao_idx, size_t bufnum, int offset, unsigned size) {
     ssize_t buf_idx = vaos[vao_idx].buffers[bufnum];
     bind_buffer(buf_idx);
-    return map_buffer(buf_idx, access);
+    return map_buffer_range(buf_idx, GL_MAP_WRITE_BIT | GL_MAP_INVALIDATE_RANGE_BIT, offset, size);
 }
 
 void*
diff --git a/kitty/gl.h b/kitty/gl.h
index 783e80964..99c2a00ef 100644
--- a/kitty/gl.h
+++ b/kitty/gl.h
@@ -55,6 +55,7 @@ ssize_t alloc_vao_buffer(ssize_t vao_idx, GLsizeiptr size, size_t bufnum, GLenum
 void* alloc_and_map_vao_buffer(ssize_t vao_idx, GLsizeiptr size, size_t bufnum, GLenum usage, GLenum access);
 void unmap_vao_buffer(ssize_t vao_idx, size_t bufnum);
 void* map_vao_buffer(ssize_t vao_idx, size_t bufnum, GLenum access);
+void* map_vao_buffer_for_write_only(ssize_t vao_idx, size_t bufnum, int offset, unsigned size);
 void bind_program(int program);
 void bind_vertex_array(ssize_t vao_idx);
 void bind_vao_uniform_buffer(ssize_t vao_idx, size_t bufnum, GLuint block_index);
diff --git a/kitty/shaders.c b/kitty/shaders.c
index d281d7ec4..b72f9fb9e 100644
--- a/kitty/shaders.c
+++ b/kitty/shaders.c
@@ -450,9 +450,11 @@ cell_update_uniform_block(ssize_t vao_idx, Screen *screen, int uniform_buffer, c
         GLfloat bg_opacities0, bg_opacities1, bg_opacities2, bg_opacities3, bg_opacities4, bg_opacities5, bg_opacities6, bg_opacities7;
     };
     // Send the uniform data
-    struct GPUCellRenderData *rd = (struct GPUCellRenderData*)map_vao_buffer(vao_idx, uniform_buffer, GL_WRITE_ONLY);
     ColorProfile *cp = screen->paused_rendering.expires_at ? &screen->paused_rendering.color_profile : screen->color_profile;
-    if (UNLIKELY(cp->dirty || screen->reload_all_gpu_data)) {
+    const bool color_table_needs_upload = cp->dirty || screen->reload_all_gpu_data;
+    const unsigned sz = color_table_needs_upload ? cell_program_layouts[CELL_PROGRAM].render_data.size : cell_program_layouts[CELL_PROGRAM].color_table.offset;
+    struct GPUCellRenderData *rd = (struct GPUCellRenderData*)map_vao_buffer_for_write_only(vao_idx, uniform_buffer, 0, sz);
+    if (color_table_needs_upload) {
         copy_color_table_to_buffer(cp, (GLuint*)rd, cell_program_layouts[CELL_PROGRAM].color_table.offset / sizeof(GLuint), cell_program_layouts[CELL_PROGRAM].color_table.stride / sizeof(GLuint));
     }
 #define COLOR(name) colorprofile_to_color(cp, cp->overridden.name, cp->configured.name).rgb