As seen at: https://learnopengl.com/Getting-started/Hello-Triangle

Why GPU Shaders Must Be Compiled at Runtime

The Core Problem: Lack of Cross-Vendor Compatibility

Unlike CPUs where you can run the same x86-64 binary on both AMD and Intel chips due to standardized instruction sets, GPUs do not have this cross-vendor compatibility. Each GPU vendor (NVIDIA, AMD, Intel) has:

• Different instruction sets: Each vendor uses proprietary GPU assembly instructions
• Different architectures: Shader cores, memory hierarchies, and execution units are organized differently
• Different optimization strategies: Each vendor’s driver optimizes code differently for their specific hardware

The OpenGL Compilation Process

When using OpenGL:

1. Shader objects are created using glCreateShader and referenced by an ID (stored as unsigned int)
2. Runtime compilation translates your high-level GLSL code into vendor-specific GPU assembly
3. Driver optimization occurs for the specific detected GPU architecture
4. Final machine code is generated that’s optimized for that particular hardware

What Runtime Compilation Achieves

• Cross-vendor portability: Same GLSL source works on NVIDIA, AMD, and Intel GPUs
• Hardware-specific optimization: Code is optimized for the exact GPU being used
• Driver improvements: Benefits from constantly updated driver optimizations

Alternative Approaches in Modern APIs

Some newer graphics APIs reduce compilation overhead:

• Vulkan: Uses SPIR-V bytecode for pre-compilation (though final optimization still occurs at runtime)
• DirectX: Supports shader bytecode to reduce compilation time
• Metal: Allows pre-compiled shaders

Even with these approaches, some driver-level compilation typically still happens to generate final GPU-specific machine code.

Performance Considerations

Runtime compilation adds startup overhead, so applications often:

• Cache compiled shaders between runs
• Pre-compile during loading screens
• Use background compilation threads