Running on Linux!

Ok, three biggies for this week's update: Linux port, performance and results of first art commission.

Linux port

I've wanted for a while to put Linux on one of my home PCs, so after a recent upgrade to my laptop (better/bigger SSD) I decided to add Ubuntu. I must confess the process of setting up dual-booting with Windows pre-installed has become far better since I last tried it (ahem, a while ago). Since Godot exports for Linux and my Native plugin doesn't use anything windows-specific, why not make the game work on Linux? That's how that mini-adventure began. The C++ code took a little bit, due to OS-branching CMake preprocessor defines, GCC take-no-prisoners approach to compilation, some missing OS-specific functions for bit-handling etc. But in the end I got a shared library! .NET was thankfully uneventful. The C++/C# glue code needed a bit of an adjustment due to the way dynamic libraries are loaded in Linux. Then I had some "fun" with compute shaders, as windows drivers (both NVIDIA and AMD) were "forgiving" with some shader layout mismatch in my code (declaring an image to have a 8-bit red channel in the shader, but C++ code says it's RGBA8) so I kept getting no data even though the shaders had no errors. After a bit of headscratching and debugging, found the issue and fixed it. Phew. And now the code works on linux! (disclaimer: I just run it in the editor, haven't exported it yet, but I assume it will be fine ... -ish). This means that I now have grand plans for releasing on Windows, Linux and Steam Deck eventually, whenever that happens. If anybody has a list of gotchas for publishing on Linux/Steam Deck, I'd be happy to hear any tales from the trenches.

Performance work

My laptop is from 2018, and it wasn't high-end back then either, so it's a bit slow. But, I'll shortly only have the laptop to work with for 3 weeks, so I can't have the game crawling. After noticing that the linux version was very very slow in certain cases, I started investigation. The culprit was the fog-of-war ubershader, which:

  • applies funky wispy fog of war
  • applies heat haze distortion
  • displays lightning strikes
  • recolors the visible areas based on time-of-day lighting
  • applies lighting modification based on light sources
  • applies darkvision effect based on light sources and darkvision skill
  • recolors explored-but-not-visible areas with a sepia color
  • applies fog "clouds" under appropriate conditions

I think that's it. So, what was the cause of the performance dip? Apparently, it was mainly the number octaves used for the fBm-based fog cloud. Note to self: try using an fBm noise texture instead of evaluating the whole thing! Anyway, the very heavy performance hit got lifted, but the game still felt laggy. Ok, let's look for shader optimisation tools out there! From a quick search, apparently there's NSight Graphics and GPUOpen tools these days, for GLSL/Vulkan stuff. None of them have a simple profiler, so in the future I need to learn how to use them really. Also, to avoid only looking at shader disassembly, I need to specially compile shaders into SPIR-V with a special flag (/Zi), but Godot doesn't seem to support that. So, future work looks like it's going to be utilizing an external compiler to build the SPIR-V bytecode (or at least a .PDB) so that NSight has something better to work with. But for now, that looks way too much like a rabbit hole. So, what to do instead?

To avoid spending too much time on it (I'm in the middle of controller support and GUI, remember? Then there's item refactor and a bunch of other tangents to the main deliverable), I did the only sensible thing: I created a different shader version which is much simpler but contains the essential visual information (black blocky tiles for fog of war, no weather). And, tada, it's a LOT faster on my laptop too, so I classify this as grand success while silently biding my time, waiting to optimise the hell out of that shader... One day...

In the meantime, (un)funnily enough Godot's profiler didn't show my performance problem at all, which I wasn't happy about. So, like any sensible person, I decided to take matters in my own hands. Godot thankfully provides GPU timestamp support via its RenderingDevice, so I wrote a bit of code to be able to identify performance cost of different "render passes". This works, as in it showed that my fog-of-war shader was terribly slow compared to everything else. If I'm to trust it, currently my entire rendering requires 1ms on the desktop and 10ms on the laptop. In Godot's source code, the vulkan driver function that gets the timestamp has some documentation starting with "This sucks" complaining about NVIDIA's parameters and continuing with magic math to deal with some GPU device magic values, so I don't quite trust it.


Commissioned image! Must use it as much as possible

New artwork

Because of reasons (explained in 1-2 weeks), I needed some game-related art commission, and I went with Fiverr. I am very happy with the results (layers and all), so besides my super-secret reason, I decided to reuse this art as much as possible, and of course the first target is my Steam page! I had a horribly coloured image before, which didn't quite communicate the game's style. So, with a bit of crop and zoom, I now got a new placeholder which I think is much better. Not perfect for a capsule image, but a lot better, IMO.

A little anecdote on this... March 2022, I noticed some beautiful piece of art, probably on Twitter or /r/pixelart. I found an ArtStation profile, made an account and followed the artist. But me being me, I forgot to check ArtStation ever again, and I don't check that email often. Fast forward two weeks ago, I was looking for this commission on Fiverr. I sent a request to ~5 artists, all were happy to do it and Karina sent me an example ruins-in-the-jungle scene for reference, that I thought was incredible. Without hesitation, I asked her to make the piece. Obvious finale: I went to browse Karina's ArtStation, and something sparked in that sleepy web of synapses. I log in using the email that I would have used, and ... yep, it was Karina I was following. Moral of the story: trust your gut, and check your emails!

And that's all for this week!