Uploaded August 2024 | Updated September 2026, 2 weeks ago

---

In this video we take a look at 3 very simple ways to optimize your HLSL code.

1. Avoiding Branches/If Statements
2. Constant Variables
3. Low Precision Variables

Technical Summary:
If/Statements can introduce branching whereas step will avoid branching, in this simple example its not going to make a difference really.

The main point was just to show how you could use a step to avoid using a If/Statement if there was significant branching/divergence.

GPUs use multiple threads to process data, and these threads are grouped into clusters (Warps/Wavefronts). If one of those clusters hits a if statement that has varying results, sometimes passing the condition, other times not. This introduces divergence, where the whole group of threads has to wait while all possible branches are processed.

Whereas when you use a step function, it will avoid branching. It just performs the operation across the threads.

Of course there may be some situations where if there's no significant branching/divergence it won't matter much. Optimizations can also be dependent on gpu architecture or the compiler optimizations. Generally the best way is to benchmark the performance with profiling tools.

HLSL Tutorial Series:
youtube.com/playlist?list=PLoHLpVCC9RmMMmW5eP1aAyJrTjxd46rx_

Patreon:
patreon.com/renderBucket

Discord:
discord.gg/BxsersBPQc