Official Feedback wanted! Profiling tools: how do you use them and what do you need?

I posted these suggestions in the survey, but just adding it here as well:

The possibility of adding custom profilers to the built-in profiler is great. I wish it would also had some additional features:

• Adding any type of custom meta data per frame and custom details UI at the bottom. Currently, custom profilers are limited to numerical values. It would be useful to also log some additional meta info per frame, such as current settings, list of active objects etc. These could then be used to display a custom info panel on the bottom.
• Visualizing boolean values, i.e. indicating whether something was on or off in each frame. Currently, you can plot it as 1/0 values, but this is not very convenient or intuitive. Instead, I suggest using a background color that is used when the given flag is set and invisible otherwise. If multiple of these boolean values are enabled, split the view vertically and stack them on top of each other.
• Visualizing time ranges. Sometimes you need to know how long different processes took across multiple frames, e.g. a loading process. It would be nice to visualize this in a similar way as the per-frame visualization of the CPU profiler.
• Adding custom markers, like in the built-in UI Details profiler.

 

Thanks for the hint. This API turned out to be useful.

Yes. We have some async processes that run over many frames (loading and processing of complicated assets at runtime) and consist of a bunch of different steps, like loading and uncompressing from disk, uploading to GPU, running some compute shaders. Currently, we can nicely visualize the workload per frame that was done, but we cannot easily see where specific tasks started and ended.

 

I'd love for the standalone profiler to work! I haven't been able to use it once in a real project since it shipped.

It used to be that it didn't work because we had the UITK experimental package installed. That got fixed when the package went away, but right now it's incompatible with Cinemachine instead, apparently? This happens every time the standalone profiler loses and then gains focus:


Other than that, the Profiler team is as always doing a great job. I've been very happy with the development of profiling over the years - the ease of profiling has always been a strong point of the engine, and you keep improving it.

 

I just want to know how long rendering operations take on the CPU side (to submit) and on the GPU side (to render) excluding vsync/any sort of sync time.

Just 2 easy numbers that I can look at, compare, and try to minimize.

For Mobile! Support for Vulkan and OpenGL3+

 

I would like more actionable views of native memory in builds, as elaborated on in my recent post here.