Almost every node in the PCGEx inherit from the same point processor, and as such they have some shared capabilities.
PCGEx has a focus on performance and multithreading – very few nodes are actively computing anything on the main thread, and instead the bulk of the tasks is handled asynchronously; and parallelized whenever possible.
This helps keeping the editor relatively smooth when performing heavy tasks.
Common Tweaks
Performance
Side note that most PCGEx nodes are greedy, and contrary to vanilla nodes, they process all inputs in parallel instead of one after another. Most of the time it’s a non-issue, but may be noticeable if you have a lot of very large datasets plugged into the same node.
Do Async Processing
Checked by default, you can toggle it off to force synchronous/unparallelized execution of the code.
This is a very legacy option, best leave it to true.
Chunk Size
The chunk size usually represents the number of point a node will process in a single parallel batch. There is no ideal value: too small and you loose the gain of parallelization, too high and you’re just hogging thread ressources. Ultimately, it depends on your specific setup.
A value of -1 fallbacks to that specific’ node default value under the hood.
Unreal PCG plugin recommend a minimum batch size of 256, which is the default value I’m using for most of the nodes. Heavier operations can go as low as 32.
Cache Result
Under the hood, all PCG node come with the ability to cache their result; but the system is designed so it’s a compile-time choice, not an editor-time one. I exposed the ability to cache on-demand at the price of some harmless asserts, because once you’re done iterating on certain settings, it’s worth caching the results.
Be aware that the cache is easily corrupted, and sometime leads to missing points or data; it’s still a small price to pay when you’re working iteratively with hundreds of thousands points.
Flatten Output
Flatten the output of this node. On 5.3 this is a very expensive operation, it’s better in 5.4 and should be even faster in 5.5. Flattening ensure all inherited attribute values are copied to the output, and metadata parenting/inheritance is forfeited in the process. This is a required step to ensure attribute values are properly saved to PCG Data Assets!
Input Pruning & De-duping
De-duplication
Datasets are de-duplicated by design in PCGEx – this means that if you plug the exact same data (as in same memory pointer) it will be only processed once – even if it has different tags.
Pruning
Datasets that are empty and contains no points will be ignored, and won’t be forwarded to the ouput pins.