smugdev nag-retweet
smugdev
1.3K posts
smugdev
@smug_dev
Professional #gamedev by day, Aspiring #indiedev by night. Programmer for @UnWarGame Interested in just about everything development related and games
/dev/random Sumali Temmuz 2019
236 Sinusundan131 Mga Tagasunod
@vadiks2003 @ccallac7 Your GPU runs your shaders in batches. Being coherent means all shader invocations inside a batch take the same path.
If you put a single bool variable into your shader that is the same for every invocation during a frame (e.g. isNight), then dynamic branching is basically free.
English
@vadiks2003 @ccallac7 It's more complicated than if/else=bad.
1. The code doesn't matter. What it compiles into does. Your compiler doesn't necessarily turn an if/else into a branch.
2. GPUs these days can handle dynamic branches quite well if the branches are coherent.
English
@DeltaWave0x It's the perfect storm, because they all happened around the same time.
English
@DeltaWave0x And in part due to "modern" (outdated by now) APIs (DX12/Vulkan) being designed in a way that also increases the number of permutations you need to compile.
English
You cannot make an engine that has to compile literally 5000+ shaders on startup, you just can't, there's fundamentally no way to fix it apart from a total fucking rewrite.
Why do you even have that many fucking shaders? Are they all permutations? Do you have one shader per thing?
What the actual fuck unreal engine, this literally the epitome of "Yeah this was never meant to be a generic engine, but for some fucking reason we decided to use it as such and now we can't go back"
English
@valigo @michaelkap19 I found it useful to think of it that way, because double buffering comes with tradeoffs like input lag/latency. You can address that with triple buffering.
In the case of few writes, many reads, I'd recommend checking out this approach: blog.s-schoener.com/2026-01-28-rwl…
English
@valigo @michaelkap19 Never heard of this name. I implemented that technique many times, but always call it double buffering, since it's exactly how simple, double-buffered vsync works. Two framebuffers: One to render the next frame into and one to display until the other is ready. Then you swap.
English
Very good talk by Jon Gjengset about concurrency - read-write locks, left-right data structures, false sharing, all that. It's not Rust-specific either, it's more about how CPUs work.
English
@Lanstrash There are two in-depth articles covering this kind of stuff:
dev.epicgames.com/community/lear…
English
@jnvcia @simplex_fx Always doing all X before doing all Y can creates guarantees that you can rely on to make your Y-logic simpler.
E.g. knowing that your pathfinding never has to handle edge-cases where the level is partially out-of-date, because you run all level modification prior to AI.
English
@jnvcia @simplex_fx This here makes perfect sense: x.com/Jonathan_Blow/…
Bucket entities by type and process each bucket one by one in a predefined order. It simplifies your dependencies, because you always know what happens when.
This then allows you to make simplifying assumptions in your logic.
Jonathan Blow@Jonathan_Blow
@anicic_filip @BOENSAW My game code has evolved toward "iterate over all entities of type X, do Y on them, at the place in the code where this needs to happen". The entity does not own the function, the game engine owns the function.
English
this shit is everywhere.
and when you point it out and try to fix it, they get mad at you
inigo quilez@iquilezles
Once I saw a project where everything was entities with a .tick(), plus a gigantic priority table intended to orchestrate the mess. Because reasoning the flow was still too hard, developers had added yield() for_rand_number_of_frames everywhere, "to give other entities some time to run their tick()s". Naturally all actions took multiple frames to execute, and they still had flow/synch bugs everywhere anyways. I advised them to burn thing down and move on to just an app loop with { do_thing_A(); do_thing_B(); ... do_thing_Z(); }. But they didn't. The project failed a year after under the weight of its own spaghetti design. You could say the problem was that particular team of coders, not the design per-se. But, I mean, I've seen this story repeat with different teams repeat 3 or 4 times by now...
English
@sarahb_paw @babadokia @fsfblake Specialization is a luxury many of them can't afford (esp. the smaller ones). You see a lot of indie devs filling multiple roles already.
If they have money to spare, they typically spend it on visuals. Understandable, but good audio is still underrated.
English
@babadokia @fsfblake Indie isn't really sustainable either as they often don't hire/contract out for most roles that you see in AA-AAA spaces. I rarely see any indies looking for audio pros... Just VAs and artists. Maybe occasionally a programmer
English
Like I understand why so many students are giving are up on their game dev dreams.
I wanna support and inspire but it truly does not matter if you ship a hit or a flop at a big studio. Your career will eventually be in the hands of some corporate person you've never met
English
@MrMPFR @SebAaltonen As much as I liked his proposal, it would be a workaround and not a fix. The actual fix would be for GPUs to have an actual ISA your shader compilers target.
You don't recompile CPU code all the time, because your compiler outputs x86/Arm, which your CPUs are build to execute.
English
@SebAaltonen Can this be fixed within DX12 and Vulkan or do we need a brand new API like you suggest months ago?
English
The PSO explosion has grown so much that today we need massive cloud servers to deliver the PSOs. Each vendor has their own solutions. We should fix the root cause instead of building this massive infrastructure.
tomshardware.com/pc-components/…
English
smugdev nag-retweet
@EyeOfScar @lukinu_u I've not done a deep-dive into how BPs work, but what I saw during that optimization session reminded me of seeing how Python works as opposed to something like JS.
I didn't see a JIT. Just a ton of function calls.
English
English
I highly do not recommend doing this, esp in UE. If possible, move all code to C++ as BPs have extreme limitations and some really terrible practices and issues. I guess for a turn based game where you don't do much, you could get away with it. Otherwise BPs can hinder perf.
Elan Ruskin@despair
English