Texture Pipeline Optimization (Immutable Storage)¶

1. Problem: Mutable Storage¶

The historical function glTexImage2D: 1. Allows resizing the texture at any time. 2. Allows changing the format (RGB -> RGBA) at any time. 3. Consequence: The driver must check the "completeness" of the texture at every Draw Call. 4. Overhead: Validation CPU cost.

2. Solution: Immutable Storage¶

Since OpenGL 4.2, glTexStorage2D: 1. Allocates all mipmap levels in one go. 2. Sizes and formats are frozen (Immutable). 3. Data is uploaded via glTexSubImage2D. 4. Benefit: The driver knows the texture is valid. No validation at Draw time.

3. Implementation in Engine¶

Code (texture.c)¶

// Old Method (BAD)
// glTexImage2D(GL_TEXTURE_2D, level, GL_RGBA8, w, h, 0, GL_RGBA, GL_UNSIGNED_BYTE, data);

// New Method (GOOD)
// 1. Determine number of levels
int levels = 1 + floor(log2(max(width, height)));

// 2. Allocate storage (VRAM)
glTexStorage2D(GL_TEXTURE_2D, levels, GL_RGBA8, width, height);

// 3. Upload data
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, width, height, GL_RGBA, GL_UNSIGNED_BYTE, data);

// 4. Generate Mipmaps (Hardware)
glGenerateMipmap(GL_TEXTURE_2D);

4. Alignment Constraint (Unpack Alignment)¶

When stb_image loads an image with 3 channels (RGB), the rows may not be aligned to 4 bytes (standard OpenGL default). * Symptom: Skewed or slanted image. * Fix:

if (channels == 3) glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
else               glPixelStorei(GL_UNPACK_ALIGNMENT, 4);

5. Memory Layout Comparison¶

6. Performance Gains¶

Significantly reduced driver overhead.
Better memory locality (Mipmaps are often packed together).
Eliminates "Incomplete Texture" errors.