archived-dots

[viral sonnet]

for example

[devout prairie]

so i guess really it depends how much data you're storing in the buffer ( ie whether it's faster/slower than a blob )

[hot basin]

you can just attach buffer element to the chunk

🤔

[viral sonnet]

yes, it's all about keeping your data as local and small as possible

and i think the true strength of chunk buffers lies in simd.

and if it's not simd, at least that it's linear data

[devout prairie]

i think a blob should at least provide fast linear and native memory access right

[viral sonnet]

can't say that for most blob data. at least not in my case. i save spell data in blobs and that's quite a lot of data alltogether

[devout prairie]

in my case i'm using a blob to store for example 1500 float3's

so i don't think db would work there, in fact i don't think there's any real alternative

[hot basin]

sounds like amination data 😄

[devout prairie]

exactly what it is

[viral sonnet]

blobs are linear, yes the layout is like [blob struct data][ptrs/length][...][arraydata1][arraydata2]...

[devout prairie]

i'm not optimizing for the moment but yeah, just curious about that

[viral sonnet]

seems fine to me, the first access is random but then it's linear

[devout prairie]

atm it's just a flat array of float3's, all bone pos/rots packed into a single array

[viral sonnet]

i'd not save this any other way. you are not capped by any 16k limit

[hot basin]

can I have second opinion on that? 😄

[viral sonnet]

hm, i dunno, try it. should be a fast test 🙂

i'd say it doesn't work but who knows 😄

[hot basin]

it doesn't

[rotund token]

Pretty sure I would not have said that

It definitely does not

[robust scaffold]

Inheritance is a naughty word here in dots land

[balmy thistle]

word

[hot basin]

I'm pretty sure someone showed me that interfaces has some clever applications

and I don't think it was about generic jobs

nvm memory is a funny thing

[drowsy pagoda]

Is this what you remembered reading about? SceneManager.MoveGameObjectToScene(GameObject go, Scene scene) ?
Tried it, works perfect. Before this I tried what you mentioned Open scene, instantiate, and then go back to previous scene. T'was a disaster, buggy.

[rustic rain]

Maybe. My research aimed for smth else, kek

Great it works

It'll probably a good feature for designer tools

[devout prairie]

Anyone recognize this error from doing a build:

not using subscenes, wondering if my gameobject which is building a blob during Convert ( just in a mono / IConvertGameObjectToEntity ) has anything to do with it

last time i built was prior to building any blobs and it worked ok.. only other thing is, i recently created a new blank project and copied my assets/scripts in, so there's maybe possibly an outside chance it could be a package problem, but i doubt it

[rotund token]

Is this just a compile error?

I've seen this before when setting up a new project
What .net are you targeting

And do you have netcode in your project?

[devout prairie]

well oddly, it seems to have gone after messing around with it 😐

i'd deleted the sample scene which comes with a blank template URP project, which was showing as 'deleted' in the build scene list, so i removed that, enabled dev build, clicked my heels together three times and rubbed my nose, and error seems to be gone

spoke to soon it came back

[devout prairie]

no netcode no, targeting .net 4.x but flipping back to 2.0 just to try and get this to go away

[rotund token]

yeah

that's the .net 4.x error i'm pretty sure

[devout prairie]

weird i never had this before

and weird it worked once then didn't

[rotund token]

entities only supports standard

[devout prairie]

i'm 99% sure my previous version of this project was 4.x

[rotund token]

was that before 0.50 or 0.51?

[devout prairie]

now i'm getting some staging error

[devout prairie]

no, same 0.51

[rotund token]

did you add local packages recently

[devout prairie]

no

just entities/hybrid/tmp

going to load my previous project and compare/doublecheck

weird error, i'll try a restart, maybe as i'd changed over to net 2.0

[rotund token]

not sure sorry but i can tell you that there are at least 2 studios who had this error targeting .net framework after a 0.51 upgrade and fixed it switching to standard 2.1

[rotund token]

the error generally occurs anytime a package change is made

including any use of asmref

(the original, no idea about that new weird one)

[devout prairie]

rather than importing from git url, i manually changed manifest which was probably a mistake

when i was setting up the project

[rotund token]

dont think that should make any difference

[devout prairie]

i had a couple of odd issues during that

[rotund token]

anyone who pulls repo has to setup via manifest

[devout prairie]

mm

it's also locked this folder even after closing unity, inside my build dir

just having lots of fun

[rotund token]

just restart your pc on that

if you check task manager you will probably see some process not closed

unity.exe

or some compiler

[devout prairie]

'just' hehe

[rotund token]

easier/faster to just restart

[devout prairie]

yeah

[viral sonnet]

is it possible to setup performancetest data via subscenes? the editor entities are there so shoudkd be possible, right?

[rotund token]

Never tried but I don't see why not

Just load the subscene into a custom world in the test

Just need to add the 2 or 3 scene systems that are required for loading then tick it until it's loaded

[viral sonnet]

hm, gonna try it. i need some better testing ground 😄

now i've time to get back to the stats reading. that last coding session really annoyed me. i tried 3 different ways and every way was slower. i came to the conclusion. as i'm reading a few stats in the first job the must be already in cache and it's faster to then build the stats for the spell calculation. otherwise, i don't know how to explain that the first job needs 1.5ms for building and another job needs 2+ms

[rotund token]

Only concern would be maybe the requirement for editor tick to load assets but in pretty sure there's a way for unit tests to support this

[viral sonnet]

hm, i guess i don't need a full subscene. just a prefab that has to be converted

[rotund token]

Probably an easier alternative

Generally everything in my subscenes are prefabs anyway so I don't need to open to edit them

[drowsy pagoda]

Getting some strange behavior, can someone help me understand why this is happening? Ok. So I have static Game.Save(), which writes to disk data gathered by an ECS job like so var data = await _system.GenerateSnapshot();
_system is a SystemBase class. Here is the async GenerateSnapshot() method:

public async Task<WorldSnapshot> GenerateSnapshot()
        {
            // gather all unit positions and rotations
            var unitSnapshots = new NativeArray<UnitSnapshot>(_unitQuery.CalculateEntityCount(), Allocator.TempJob,
                NativeArrayOptions.UninitializedMemory);
            
            var unitSnapshotJob = new UnitDataExtractionJob
            {
                unitGuids = _unitQuery.ToComponentDataArrayAsync<UnitGuid>(Allocator.TempJob, out var guidsJobHandle),
                transforms = _unitQuery.ToComponentDataArrayAsync<LocalToWorld>(Allocator.TempJob, out var transformsJobHandle),
                unitSnapshots = unitSnapshots
            };
            
            var dependencies = JobHandle.CombineDependencies(guidsJobHandle, transformsJobHandle);
            var unitSnapshotJobHandle = unitSnapshotJob.ScheduleParallel(_unitQuery, dependencies);
            Dependency = JobHandle.CombineDependencies(unitSnapshotJobHandle, Dependency);

            while (!unitSnapshotJobHandle.IsCompleted)
                await Task.Yield();

            var snapshot = new WorldSnapshot(unitSnapshots);
            unitSnapshots.Dispose();
            return snapshot;
        }

The reason why I am not using Source generation is because it gave me weird errors since I was trying to execute this outside the system's OnUpdate method. But old school worked 😉
Alright here is the strange behavior part. On my in-game UI, you click Save Data button, and it will just forward call to Game.Save(). The data gets saved without error.
But in my inspector, I have a component that has a button Save Game In Editor and it too does Game.Save(), but it gives me this error and I have no way of knowing how to plug this hole! Can someone help me understand why in one place it executes fine, but called from another it doesn't?
Here is the exception: InvalidOperationException: The previously scheduled job GatherComponentDataJob reads from the Unity.Entities.EntityTypeHandle GatherComponentDataJob.safety. You must call JobHandle.Complete() on the job GatherComponentDataJob, before you can deallocate the Unity.Entities.EntityTypeHandle safely.
As you know GatherComponentDataJob is a Unity job and not one I wrote...what's going on? If you need to see other code, let me know I'll paste it.

[rotund token]

What you're doing is not safe and the editor button is just pointing this out

It's not safe to use ecs access via public methods like that

A system only updates its dependencies in before on update

Just use em, cache the data and run your save

Or do it if the same system as caller

[drowsy pagoda]

So it's a matter of timing? Is there a way to "wait" for that moment and send the signal?

[drowsy pagoda]

Meaning, move the methods and logic to that SystemBase class?

Also both are being executed during runtime.

[rotund token]

Yeah pretty much, you can just use an Entity event (what I do) to trigger system or just use entity manager instead to sync data before and do it directly on the ui

[drowsy pagoda]

Could you elaborate on that Entity event?

And that system that "GenerateSnapshot" has "AlwaysUpdateSystem" attribute

[rotund token]

Also 1 note is you're not writing your dependency back either so no other system can wait on it

[rotund token]

I simply have a component called Save

And if I want to save game I create an entity with that component

And that triggers my system to run

[drowsy pagoda]

I had a feeling I was doing something wacky with deps. How do I write it back?

[drowsy pagoda]

Oh that's a good idea. I think I'll fallback to that approach if this dependency issue doesn't solve it.

Could you help me understand how to write the dep back so other systems can wait on it?

[rotund token]

You can't doing it that way

That's the problem

You're not in the right system

[drowsy pagoda]

Meaning the system is not in the stack trace initiating the call to that method? Because that method is in the system class.

[devout prairie]

build working fine after switch to net 2.0 and restart

[rotund token]

Arriving at gym back in like 90min

[drowsy pagoda]

Ok, cool. I'll mess around and hopefully find a solution before then 😉

[rotund token]

Just to confirm

[drowsy pagoda]

Ok, right after I wait for handle.IsCompleted to be true, I do CompleteDependency() and the operation actually works. It still throws that exception, but it actually does what I need it.
When I remove that await handle.IsCompleted and just do CompleteDependency() it works without throwing.

[rotund token]

Where are you calling generate snapshot from

[drowsy pagoda]

Game.Save()

Game is a static class that doesn't inherit

[rotund token]

So it's not called from the system the method is in

[drowsy pagoda]

No. Game.Save() has var data = await _system.GenerateSnapshot();

that's where that injection takes place. _system is a SystemBase class.

[rotund token]

Class starting afk again

[drowsy pagoda]

ok

I can upload all the relative code for you somewhere if you want to look at it later and give me some insight.

[drowsy pagoda]

Alright I just did Game.Save(bool forceMainThread = false). And when editor calls this, it does Game.Save(true); This then bypasses the handle.IsCompleted awaiter and just forces CompleteDependency() and it all works 🙂

[drowsy pagoda]

how do you clear/rebuild entity cache via script?

[rotund token]

Ideally you shouldn't have to

You can register asset dependencies in subscenes and any change to these assets will rebuild relative subscenes

And you can version conversion scripts

That said the version approach is a pain tbh

[drowsy pagoda]

I agree. But my editor script that loads worlds into a sub scene, and when that does, it spawns remains of previous data even though game objects have been destroyed.

So I need to clear cache when I clear scene before loading

[drowsy pagoda]

Can I just delete the SceneDependencyCache folder and that should force rebuild right?

[rotund token]

In the dots menu

There's I button that opens a window

That has a button that clears the cache for you

I can't remember what it's called, on a tram so can't check

If you look at the code of that window though you could see how to trigger it from code yourself

[drowsy pagoda]

All it does it look for GlobalEntitiesDependency scriptable object, and generate a new GUID for it. Unfortunately that class is internal. Quick question, can I execute internal class methods via Reflections? If not, then it looks like I just need to delete that asset....right?

[rotund token]

you can

[drowsy pagoda]

Oh! Can you give me a quick boiler plate? So this is the magic formula: EntitiesCacheUtility.UpdateEntitySceneGlobalDependency();, but EntitiesCacheUtility accessor is internal and sits in the Unity.Scenes.Editor assembly. How can I trigger this via Reflections?

[rotund token]

imo easier to just give yourself internal access using asmref

but if you want to do it via reflection you need to get access to type using something like Type.GetType()

then just call GetMethod using something like GetMethod("UpdateEntitySceneGlobalDependency", BindingFlags.Static | BindingFlags.NonPublic)

[drowsy pagoda]

Yeah I know how to use reflections. I'm just having a hard time EntitiesCacheUtility.GetType() because EntitiesCacheUtility is not allowed to be used since it's internal lol. Anyways, I just did EditorSceneManager.MarkSceneDirty(scene); and it seemed to do the trick, I'm no longer experiencing those issues. So for now, keep my fingers crossed cause this wasted way too much time.

[rotund token]

easiest way imo is just find something else in the assembly then just do a GetType

typeof(SceneImporterData).Assembly.GetType("Unity.Scenes.EntitiesCacheUtility")

[drowsy pagoda]

SceneImporterData was simply just something you picked at random that exists in Unity.Scenes.Editor assy right?

[rotund token]

yep

just the first thing i saw in that assembly that was public

most of it's internal

alternatively you could use a full qualified name with Type.GetType(namespace, type, assembly, version etc)

[drowsy pagoda]

smart thinking. That's exactly what push I was looking for. Thx!

[rotund token]

but that's a bit more painful to figure out exactly

i always forget the format

[drowsy pagoda]

lol

[late mural]

had this warning pop up randomly, is this something i should try fixing, or is it just a random one off thing? I have not touched any of what it was saying in the warning, as far as i know.

[robust scaffold]

What package version are you using and what version of unity?

[late mural]

the latest 2021 long term release, and ill quickly check the package version brb

[robust scaffold]

HR or Entities?

[late mural]

i dont know what HR is lol, so i was gonna check entities, is that right?

[robust scaffold]

Hybrid Renderer vs the Entities package.

[late mural]

oh ok, should i check both?

[robust scaffold]

You havent been putting [BurstDiscard] anywhere have you?

https://docs.unity3d.com/Packages/com.unity.burst@1.7/manual/docs/Warnings.html#bc1371

[late mural]

i dont think so

[robust scaffold]

Check your entire solution for any of those tags on methods. See if it's the same situation as in that documentation page. If you know for certain it isnt your code, just ignore it.

[late mural]

ok ill check around, cause it only popped up after i was doing some dll import stuff, could that cause it?

ok ye no burst discard anywhere

[robust scaffold]

Hrm. I dont think so. Unless you have burst code in those DLLs. This error is due to C# compiler optimizing and unintentionally disabling burst.

[late mural]

weird, i doubt it would have burst code in the dll, it a c++ dll,

[robust scaffold]

If you aren't using it anywhere. Restart the editor and see if it happens again. It might just be an old burst cache.

[late mural]

ok ill try that and report back, thanks!

[late mural]

ok editor restarted, and yup it is gone, thanks so much!

[rotund token]

anyone know much about these artifacts for unity physics?

enabling it definitely seems to show a more stable simulation

[rustic rain]

Sir, do you happen to know whether it's possible to swap burst compiled function pointer with another?
Or whether it's even presented this way internally?
It's regarding modding

[rotund token]

sure why not?

new FunctionPointer<TryFunction>(ptr)
function ptr just takes an int ptr

you can pass in anything you want

as long as signature is correct it'll work fine

[rustic rain]

yeah, but what I mean is swapping for existing burst compiled code

are those pointers stored somewhere available to user code?

[rotund token]

you can pass whatever function pointer you want into a job conditionally

[rustic rain]

but what I mean is modding from outside

after you built game

you launch it, load your assembly and you want to swap pointers

meanwhile original implementation is vanilla burst in Entities.ForEach loop

[rotund token]

there's really nothing special about function pointers as far as i'm aware

[rustic rain]

waaaait a second

[rotund token]

        {
            get
            {
                CheckIsCreated();
                return Marshal.GetDelegateForFunctionPointer<T>(_ptr);
            }
        }```
by default (outside of burst) all it does is marshal pointer

[rustic rain]

let's say there's 3 bursted functions with same signature

can you put them all in one, that will be executed one by one?

[rotund token]

ahh what

[rustic rain]

prefix,postfix

[rotund token]

a function pointer maps 1 : 1 to a method

[rustic rain]

hmmm

[rotund token]

just think of it as a delegate

[rustic rain]

yeah

that's what I mean

() => {
/// code
}

instead of this

have

()=>{
ptr.Invoke();
ptr1.Invoke();
ptr2.Invoke();
}

for bursted function pointers

so that you can swap original one with that one

[rotund token]

no because you can't do () => in burst

[rustic rain]

and pass all parameters between them

that's just example

for the sake of passing my thought process

[rotund token]

if you have a job with 3 function ptr fields

you can pass whatever 3 functions you want to those methods

they can be different each frame

[rustic rain]

no, I mean is modify existing job without any function pointers

so instead of only original code that was compiled AOT

you could run additionally loaded more bursted functions

[rotund token]

probably going to have a much easier time just writing a call to a native library

and have that run extra code

i guess the only way i could think of

is if function pointer of function pointers work

i dont know about this though

[rustic rain]

eh, I just hope Unity will simply make a JIT compilation on demand

for burst and modification of code

[rotund token]

you're probably going to have to pay for an engine license if you want that anytime remotely soon

in which case you couldn't share it anyway

modification license*

[rustic rain]

well, at least something to be able to modify execution of bursted code

allthough

considering ECS

I have a feeling

that simply creating a System with constraints UpdateBefore/After can be enough

literally will work like prefix/postfix

[rustic rain]

ahem

FixedString serialization doesn't work?

[rustic rain]

oof

I really need to think of a way to have prefab system

But I can't figure any solution

I need a way to keep prefab reference of existing entity

so that I can easily "clone" entity based of it's initial prefab

not based of it's existing state

[open condor]

In Dots NetCode is it possible to define a Component that will NOT be synced through network? Because everytime i ADD a component on serverside it seems to result in a network update for this entity, even though the component is not actually being synced

[rustic rain]

        public void Convert(Entity entity, EntityManager dstManager, GameObjectConversionSystem conversionSystem)
        {
            var builder = new BlobBuilder(Allocator.Temp);
            ref var def = ref builder.ConstructRoot<Def>();
            def.defType = new FixedString128Bytes(defType);
            def.defName = new FixedString128Bytes(defName);

            var reference = builder.CreateBlobAssetReference<Def>(Allocator.Persistent);
            conversionSystem.BlobAssetStore.AddUniqueBlobAsset(ref reference);

            dstManager.AddComponentData(entity, new DefReference { def = reference });
        }

Is that the correct way to create blob assets in conversion?

[viral sonnet]

hm, i think as long as a comp is on the ghost prefab it will be created on client side. it won't be synced though. that would require the sync tag afaik

[viral sonnet]

i have GOs in a subscene. those are converted to prefabs automatically and on runtime start i create a hashmap based on comps. would that also work for you?

[rustic rain]

that's not only about having database

it's also about having link on entity

to that prefab

rn I figured

I could store blob asset reference on every entity

[viral sonnet]

so the instantiated entity needs a reference to its original prefab?

[rustic rain]

yes

[viral sonnet]

i also had that at one point. you can patch it via ecb

[rustic rain]

this for A: saving system
B: gameplay interaction

nah, I want to do it one time in conversion

I decided

that I just limit myself from having prespawned entities

and instead do all spawning through code

[viral sonnet]

would also work when the prefab references itself

[rustic rain]

unless it's debugging

yeah, prefab will reference itself in this case

[viral sonnet]

i had true runtime data so when i was doing it i had to use ecb patching

[rustic rain]

but actually

it's not Entity reference

it's string reference

I figured I'll use

defType and defName strings

[viral sonnet]

some stable hash or id makes sense when it's for a savesystem too

[rustic rain]

to define unique combinations for any sort of defnitiion I want

currently I have this weird error

and counter makes me worry a bit 😅

    [Serializable]
    public struct Def
    {
        public FixedString128Bytes defType;
        public FixedString128Bytes defName;
    }

    public struct DefReference : IComponentData
    {
        public BlobAssetReference<Def> def;
    }

not sure if I did anything wrong

[viral sonnet]

it's correct. try to remove the serializable

[rustic rain]

nope

[viral sonnet]

hm, that's odd. i don't get any errors with FixedString in my blob.

[rustic rain]

I am very confused

I even changed name for type

and it still says same error with old name

[viral sonnet]

just copy/pasted your code. no errors

maybe restart unity 😄

[rustic rain]

yeah, that helped only on second attempt

bruh, now I get this odd error which pauses editor on start

tf is wrong with blobs

hmm

[viral sonnet]

doesn't look like anything from blobs

[rustic rain]

    public struct Definition
    {
        public FixedString128Bytes defType;
        public FixedString128Bytes defName;
    }

    public struct DefReference
    {
        public BlobAssetReference<Definition> def;
    }

    public struct Def : IComponentData
    {
        public DefReference def;
    }

oh god...

do I really need this monstrocity?

[viral sonnet]

no 🙂

[rustic rain]

that's the only way error dissappears

[viral sonnet]

does it work with a second struct?

[rustic rain]

hehe

[viral sonnet]

lol wtf

[rustic rain]

ok, I'll try again

with 1 wrapper

nope

errors

[assembly: RegisterGenericComponentType(typeof(BlobAssetReference<Definition>))]
hmmm

what if

nah

double wrapper lezzgo

wait a second

turns out

this error appeared

if you had query debug window opened

huh

damn, it was hidden so well behind other windows 😅

[rustic rain]

@rotund token sir. I remember that you made custom inspector for some pointer based component. Could you tip me how were you able to do that?

[rotund token]

Just implement Inspecter<>

[rustic rain]

uuugh, no idea what is that

at least

can't locate any useful reference in manual

[rotund token]

I am the manual

It's probably Inspector can't remember

Yeah Inspector

Just override the build method

    {
        public override VisualElement Build()
        {
            var list = this.Target.Targets;
            var label = new Label($"Target Chunks: {(list.IsCreated ? list.Count() : 0)}");
            return label;
        }
    }

[rustic rain]

ah yes, this is so good

thank you sir

[devout prairie]

math has this little helper math.all(bool2 x) which will return if both bools are true

any idea if there's something like a math.none?

ie if both bools are false

[rustic rain]

hm

just do NOT

on result?

actually

[devout prairie]

mm no because, if one is true and one is false, it'll return false

[rustic rain]

you need

math.any

and then invert

[devout prairie]

ahhhhh

any

nice

[rustic rain]

no idea if that one exists, kek

but would be weird if it's not

[devout prairie]

it does haha

useful little helpers

[rustic rain]

smid friendly little helpers hehe

[devout prairie]

yeahh

tbf, it's just this:

but cleans things up and as you say any math stuff should be simd friendly afaik

[stiff valve]

Does anyone know how to fix this problem: my cube disappears when I "close" the scene

I have this

[rotund token]

I assume you have hybrid renderer installed?

[stiff valve]

Yes

[rotund token]

What render pipeline?

[rotund token]

You don't seem to have urp or hdrp

It does not work with built in pipeline

[stiff valve]

what do you mean

[rustic rain]

https://docs.unity3d.com/Packages/com.unity.rendering.hybrid@0.51/manual/creating-a-new-hybrid-renderer-project.html

For the Universal Render Pipeline (URP), use the Universal Render Pipeline template.
I'd say start with that

[stiff valve]

Thanks

[robust scaffold]

Burrrrrrst. Why is IJobEntity not vectorizing any of the code. PAIN. Dont make me write out the whole IJobEntityBatch by hand

UGGGGHHHHHHHHHH. I'm gonna have to atomize these components if I want actual vectorization arn't I.

PAIN

[rotund token]

yes

generally a requirement if you want to vectorize your components

[robust scaffold]

Now I see why aspects are gonna change the world. Because this is pain

Holy shit unity. VECTORIZE IT PLZ

[rotund token]

IJobEntityBatch 4 life

[robust scaffold]

I cant believe i have to resort to this nonsense

What was the entire purpose of 0.50 if I have to sully myself with this manual code writing?

come on burrrrst

Wait no, that's the fallback. Here we go

This is so sad. 0.50 did absolutely nothing to the standard workflow

If I have to manually list the component type handles in order to get some vectorization, this was a complete waste.

At this point, I'm using a slightly fixed up version of 0.17. Unity has failed me.

I feel like I regressed in some way. After seeing the light that is code-generated jobs, this is just so disappointing.

I guess the upside is that now I can see the burst compiled code really easily with the no inlining tag.

[viral sonnet]

you can rewrite that with Reinterpret instead of the float*

[robust scaffold]

Ehh, do you remember the burst loop unrolling tests we did a year or two back? How [NoAlias] tagged pointers resulted in unrolled floor loops slightly more often than similarly tagged ref floats or native array inputs?

[viral sonnet]

i had some good results with burst 1.7.3

[robust scaffold]

In this case, I can use Reinterpret as it did unroll it successfully but I would like to keep my code style standardized.

[viral sonnet]

auto 2 loops for counts that are % 4 > 0

[robust scaffold]

I guess, would make things easier

But it would balloon the size of the method parameter list

[viral sonnet]

afaik it's not even burst but LLVM

don't make a method. i stopped doing that

should be quick enough for you to test out the results

[robust scaffold]

I'm using it for easy burst inspection. I use aggressive inlining tag when I'm happy with the inspector.

Otherwise it's buried in a mountain of chunk code.

[viral sonnet]

yeah makes sense. i wish there was some focus on the codefile to get rid of all the other nonsense

[robust scaffold]

During the early days of 0.50 announcement when Unity felt brave enough to talk on the forums, the Burst team did say that was a proposal they were looking into. Not implementing, just testing out things.

[viral sonnet]

or a simple keyword search ... 😄

i have a 42k loc assembly

[robust scaffold]

Yea, it unrolled. Hopefully this works for everything.

I dont get why Unity's code gen version doesnt. It's literally the same code.

Literally.

[viral sonnet]

wait, what doesn't unroll?

[robust scaffold]

IJobEntity. The code gen'ed job struct

[viral sonnet]

wdym?

how does it help?

[robust scaffold]

Atomized component structure.

Aspects will group together component structs without actually merging them together into one struct.

You want them in separate structs to pack the native array for burst. But an aspect will simplify querying the components together

So you dont need to type in a quarter million ComponentTypeHandle<> for all the components now separated into their own unique struct/component

[viral sonnet]

ah i see, less compHandles and GetNativeArrays, right?

yeah ok

[robust scaffold]

Yea, hopefully

[viral sonnet]

that will be nice. haven't thought of that

[robust scaffold]

But seeing how the current code gen isnt able to handle vectorizing a simple multiplication, i'm not holding my breath

[viral sonnet]

i think it will still need some reinterprets though. not sure the compiler gets it

[robust scaffold]

Hrm

This allows for a single loop vectorized multiplication

However, not unrolled.

This will allow for the unrolling of the angular multiplication but not the second. Both loops are vectorized though but not merged into 1 by burst.

This does not vectorize.

[viral sonnet]

interesting. does this one unroll with a tmp stack value of float dampening = dampening[i];?

[robust scaffold]

shouldnt matter, modern compilers will simplify to the code I have now but I'll try

Yea, doesnt matter

[viral sonnet]

k ... yeah auto vec is reaaally basic

expect more than 1 line to fail 😄

[robust scaffold]

What NativeArray<> will not allow is this.

[viral sonnet]

needs a ref afaik

which would also mean pointer reinterpret to float3

[robust scaffold]

how?

[viral sonnet]

ElementAt returns a ref

don't think that's wise though. just cast the pointer.

[rotund token]

(elementat on a native array is my extension, it only exists on buffer by default)

[viral sonnet]

but a ref float3 should be able to just write back to xy

[robust scaffold]

This also works, I think

[viral sonnet]

oh lol, right i just have it as extension. sorry 😄

[rotund token]

i'm curious if it'd be different if you used element at

[viral sonnet]

anyway, that's the method if you need it public static unsafe ref T ElementAt<T>(this NativeArray<T> array, int index) where T : struct { return ref UnsafeUtility.ArrayElementAsRef<T>(array.GetUnsafePtr(), index); }

[robust scaffold]

Using a pointer also removes several commands.

[viral sonnet]

pointers are my jam 😄

[robust scaffold]

Well that just gets the pointer but hides it behind a extension method

If I just use the pointer directly, there's no need for custom extension methods

[rotund token]

hmm

why does this show up in the mix

is that just the indexer

doesnt look right

[viral sonnet]

yeah, i think it is

[robust scaffold]

Kinda a mess

I dont think it's the indexer

Seems like if I were to use .xy swizzler, it moves the result of the addition as a single operation instead of packed

Dont know if the cost of an addition insert and blend operation, even packed vectorized, is worth having the move be vectorized as well

Of course, if I assume everything is a float4, that simplifies everything

That however means 4 floats being wasted. Hrm, I might need to make translation a float2

Rolling my own translation simplifies the commands slightly. But that would mean completely re-implementing the transform system.

[rotund token]

out of interest why the noinlining

[robust scaffold]

Burst inspection. Moves the code to the top of the inspector

I change it to aggressive inlining when I'm done staring at purple colors.

[rotund token]

because without that

mine generates nicer code

burst generated different operations

[robust scaffold]

No difference

[rotund token]

interesting

[robust scaffold]

Also fails to vectorize when merging the for loops together. Commented out the [NoInlining]

And copy pasting it into the Execute() method doesnt change the code. Just disperses it across the entire output

Im gonna stick with [NoInlining]. Makes debugging small changes a lot easier to find.

[rotund token]

left inline, right noinline
removes the vbroadcastss instruction

[robust scaffold]

huh. hrm

[rotund token]

🤷‍♂️

[robust scaffold]

Can not replicate

[rotund token]

is your damping marked readonly?

[rotund token]

no idea

quite odd

nothing special i don't think

[robust scaffold]

Can you paste that as text?

[rotund token]

        private struct TestInline : IJob
        {
            public float FixedDt;

            [ReadOnly]
            public NativeArray<float> Damping;
            public NativeArray<float> Angular;
            public NativeArray<float2> Velocity;
            public NativeArray<float3> Translation;

            public void Execute()
            {
                this.Vectorized(this.Damping, this.Angular, this.Velocity, this.Translation);
            }

            private void Vectorized(NativeArray<float> damping, NativeArray<float> angular, NativeArray<float2> velocity, NativeArray<float3> translation)
            {
                for (var i = 0; i < angular.Length; i++)
                {
                    angular[i] *= damping[i];
                }

                for (var i = 0; i < translation.Length; i++)
                {
                    velocity[i] *= damping[i];
                    translation.ElementAt(i).xy += velocity[i] * this.FixedDt;
                }
            }
        }```

thought i'd just throw it in a performance test for interest

[robust scaffold]

hrm, thats pretty significant

[rotund token]

10 mill elements

[robust scaffold]

Yea, broadcast is gone

what is going on?

[rotund token]

the only thing i can think of is native array vs ptrs

but i dont see why burst couldnt figure that out

alternatively can you try to use another value for count

[robust scaffold]

Not pointers

[viral sonnet]

oh tertle did the test with elementat. seems fine, right?

[rotund token]

oh mine was using withelement

same as the ptr

[robust scaffold]

I tested both. I expanded the elementat to the full unsafeutil as I dont have the extension

[rotund token]

yeah i saw

[rotund token]

i got no idea, can't break it changing anything

[robust scaffold]

The broadcast is for FixedDt.

Commenting it out and turning it back to no inlining removes the broadcast

[rotund token]

ah

[rotund token]

if you pass it into the method does it produce a similar thing

or maybe cache it at start of loop

odd

[robust scaffold]

Hrm, yea. Alright, so make Vectorized static.

Yep, this is for the actual IJobEntityBatch method.

[robust scaffold]

[BurstCompile]
private unsafe struct TestInline : IJob
{
    public float FixedDt;

    [ReadOnly]
    public NativeArray<float> Damping;
    public NativeArray<float> Angular;
    public NativeArray<float2> Velocity;
    public NativeArray<float3> Translation;

    public void Execute()
    {
        Vectorized(Damping.Length, FixedDt, (float*) Damping.GetUnsafeReadOnlyPtr(), (float*) Angular.GetUnsafePtr(),
            (float2*) Velocity.GetUnsafePtr(), (float3*) Translation.GetUnsafePtr());
    }

    [MethodImpl(MethodImplOptions.NoInlining)]
    private static void Vectorized([AssumeRange(1, int.MaxValue)] int length, float dt, float* damping, float* angular,
        float2* velocity, float3* translation)
    {
        for (var i = 0; i < length; i++)
        {
            angular[i] *= damping[i];
        }

        for (var i = 0; i < length; i++)
        {
            velocity[i] *= damping[i];
            translation[i].xy += velocity[i] * dt;
        }
    }
}```
Can you performance test this?

The only difference now is inlining. And with [MethodImpl(MethodImplOptions.NoInlining)] -> [MethodImpl(MethodImplOptions.AggressiveInlining)]

[rotund token]

let me undo my deletes then sure

actually

take that second one

i think the first was flawed

[robust scaffold]

Hrm, relatively similar?

Wait, no inlining is faster?

by nanoseconds

eh, it falls within 1SD

Is the inline using aggressive?

[rotund token]

i think they're exactly the same but the results vary 1ms between tests

[rotund token]

yes

[robust scaffold]

how about without the tag at all.

[rotund token]

actually id say on average no inline does tend to edge ahead

but maybe its because its running second 🤷‍♂️

[robust scaffold]

Yea, with a difference in the 10 ns, it might as well be equal.

Thats what, 4 operations?

[rotund token]

run it in reverse order

and it swaps around

but yeah results are quite varied

sometimes randomly have a 17 vs 21ms fights

alternating

[robust scaffold]

Im gonna run with yea, no difference.

[rotund token]

i dont think we'd expect there to be

[robust scaffold]

Thanks for that. I still need to figure out how to make that performance tester work

[rotund token]

static method in a method with nothing else

good distraction anyway

time to back to debugging something that's causing me issues =\

[robust scaffold]

Huh, using floating point deterministic mode does actually change the burst output

They did say it was in very early testing phase. No announcement but if you do every single math operation to floats in burst, it's largely deterministic

[rotund token]

on the modern hardware within the same environment we've found it to be completely determinstic (as expected)

any modern intel/amd chip seems to be fine

it fails across platforms though

i.e. consoles

[robust scaffold]

ah, about to say cross platform

[robust scaffold]

with the deterministic tag or without?

[rotund token]

without

[robust scaffold]

How about fast float operations, MADD?

[rotund token]

havent tested

[robust scaffold]

hrm, interesting. I guess deterministic will be for cross platforms and for 100% guarantee determinism by removing any optimization

[rotund token]

we don't rely on determinism we just use it for debugging seeds that people report as having issues

[robust scaffold]

Im very slowly rolling my own 2D deterministic stateless physics engine

[rotund token]

so it's nice to be able to generate the same environment if an issue is reported

[robust scaffold]

works across desktop platforms? Linux - Windows?

[rotund token]

we haven't released linux/osx

so can't say

[robust scaffold]

hrm, good to know

[rotund token]

only tested consoles + windows

dont quote me but i believe scarlett might be fine with desktop windows

i haven't tested enough to be certain though

[robust scaffold]

It's unannounced feature anyways. I'll develop with the assumption it'll work someday

Unity has advertised the 3D physics as deterministic so burst will have to be deterministic one day

[rotund token]

burst team have mostly worked on usability, bug fixes and performance for last like ~year

which i appreciate

but i feel like it's at a good point now

so we might start seeing new features sometime

[robust scaffold]

The inspector itself has been making leaps and bounds in usability.

[rotund token]

yeah the one thing htat would make it amazing

is the ability to hide job boilerplate code in it

basically, an only show user code option

[robust scaffold]

Yea, as I said above, the Burst team has stated on the forums they're looking into it. Compiler magic if they figure it out.

[rotund token]

from a purist coding perspective it sounds a bit gross

[robust scaffold]

Until then, I'll stick with no-inlining

[rotund token]

like the compiler having knowledge about other libraries that may or may not exist

[robust scaffold]

might just be a tag, something like "folder this method in the inspector"

and have unity tag everything they do to the job under it

[rotund token]

but yeah definitely my #2 feature i'd like to see

#1 being my request for an always/force burst compile option for specific jobs

[robust scaffold]

Make the [BurstCompile] tag optional?

[rotund token]

there are certain jobs that are just not usable without burst on

[robust scaffold]

Assume everything is burst compiled, make the burst discard the new conditional?

[rotund token]

an example which i think may expedite my request for them is the dots hierarchy

accidently start the game without burst and spawn like 1mill entities with that window open

and your editor just locks up for ages

[robust scaffold]

Oh yea, without burst it simply doesnt function with large amount of entities

[rotund token]

yeah hence that window should be force burst compiled always regardless if burst compile is enabled!

[robust scaffold]

I have to kill unity in task manager just to close and reopen

Why does the editor pull all entities? Shouldn't it pull just enough to populate the current hierarchy view. Dont know if thats possible though

[rotund token]

i've done testing with 100mill entities recently

and the window handle it fine

[robust scaffold]

With burst, anything is possible

[rotund token]

well i'm more thinking about ui toolkit

and how it's holding up

imagine if that was drawn with imgui

[robust scaffold]

They were planning on something like a DOTS hook. But last I heard of DOTS integration with UIE was back when it was UIE and not named UIT.

I've gotta check the code for the hierarchy to see how they're passing data to UIT. The current method of per element is very slow.

ugh, burst why

[rotund token]

why don't you actually memset this

[robust scaffold]

Burst recognizes it as a mem set. This is easier.

[rotund token]

does it? interesting

it's always done some simd instruction to set it for me

instead of using an intrinsic

tbf i haven't tested in ages

[rotund token]

well thats nice of it

think i'll still keep using my memset so some other dev doesn't come in and add something to the loop

and just break everything

but good to know

[robust scaffold]

If it's grouped with other operations in the same loop, yea it breaks the mem set

which is why there are 3 for loops and not 2.

[rotund token]

sure it wouldn't perform better if the top 2 loops were combined?

[robust scaffold]

Burst will not vectorize the first for loop for some reason. Switches to scalar single operations.

If it's merged with the second.

Isolating it results in proper PS operations and even unrolling. A similar drop in efficiency occurs when I merge the quaternion.RotateZ(angle[i]) in the second for loop. But it's just a single mov operation added to the loop so I determined it, without testing mind you, worth the cost of not having a 4th loop.

[rotund token]

oh

it's the quaternion

yeah

[robust scaffold]

no no, removing the quat and merging it has the same issue

it's the float2 of velocity and float3 of translation thats the problem

i have screenshots of the burst output above with and without merged, this was before I added the quat line.

[rotund token]

the single loop seems to reliably perform just marginally faster in a test for me

[robust scaffold]

Without the unrolling and vectorization?

[rotund token]

but within SD

burst shows it's much more scalar but doesn't really matter

[robust scaffold]

Scroll down a bit and it should produce this:

for 2 loops

[rotund token]

yeah i get that

but it just performs worse than combined

[robust scaffold]

alright, so merge it together seems to win. Wow

I really gotta figure out how to run performance tests. That is quite surprising

[rotund token]

again though, it's too close to really say

[robust scaffold]

Easier to read if it's merged. And I dont think I need to run a test to know that mem-set is always faster than a vectorized set.

[rotund token]

1 sec

might have been a minor flaw in my test let me rerun

burst above was fine but because i have safety off for running these missed something

ok your double loop actually helps

measurably

👍

[robust scaffold]

Alright, thats much more logical.

[rotund token]

starts performing worse once you get to 5 loops 😄 (makes sense)

saw no difference with 4

anyway i really need to stop procrastinating

[robust scaffold]

Ha, imagine procrastinating (looks at unfinished thesis)

[mortal crow]

Can't find anything on a DOTS physics rope. Is there any asset or tutorial to have that?

[viral sonnet]

heavily testing burst and simd = procrastinating. got it, i'll continue blowing leafs and be productive

[rotund token]

not heavily testing my own burst and simd = procrastinating

i'm too easily distracted by interesting things others are doing

[viral sonnet]

i can relate. at least it's a learning experience and eventually useful 🙂

[viral sonnet]

well, that's not good. i've now a testcase for my job. pretty complicated one to setup. in game it's around 3.8ms, in performance test ~12ms.

[robust scaffold]

@rotund token There has to be a better way than this. Is there?

[viral sonnet]

afraid that's it

put a #region around this abomination. this is what i did so at least i don't have to look at it

[robust scaffold]

noooooooooooooooooooooooooo, God I wish Unity's code gen wasnt so bad.

[rotund token]

the only thing more ugly than the above is regions

[viral sonnet]

you take that back!

[rotund token]

my project is a safe space from regions

[viral sonnet]

ok i'll bite. what's so bad about regions? especially around ugly fire and forget code blocks?

[rotund token]

i mostly stick to stylecop ruleset to ensure consistency

and their reasoning
https://github.com/DotNetAnalyzers/StyleCopAnalyzers/blob/master/documentation/SA1124.md

In many editors, including Visual Studio, the region will appear collapsed by default, hiding the code within the region. It is generally a bad practice to hide code, as this can lead to bad decisions as the code is maintained over time.

i don't necessarily agree with all the rules in the set, but i like having a standard ruleset for consistency that i can enforce with analyzers

(though personally i somewhat agree with regions, i've never felt they improved any codebase I worked with)

[timber ivy]

is the navmesh api available in dots?

I have a large world that spans multiple scenes so I need to generate a navmesh 100x100 around the player. I dont plan on using ecs but i'd like to use the job system if possible

[rotund token]

kind of

they used it in the first demo way way back

[timber ivy]

so would it be possible to generate a navmesh at runtime in a job?

[rotund token]

oh there's no need

they already have runtime navmesh generation

(but yes)

[timber ivy]

yes but I imagine its very slow right?

[rotund token]

a job would not change anything

but no i find unitys navmesh generation extremely fast

i wrote my own wrapper for recast

and unless i drop down all settings

unitys is about 6x faster

if i drop down all settings is near identical

(it does produce a very similar navmesh)

except their height is better for some reason

[timber ivy]

What is the class name i would be looking for in the docs to do this?

[timber ivy]

what size maps were you doing when you found this out?

[rotund token]

i cant remember size

but it was a city

[timber ivy]

larger than 100 meters id imagine?

[rotund token]

with 8000+ meshes

[timber ivy]

ok then yeah

[rotund token]

it does it async in background

(optionally)

[timber ivy]

alright awesome what api did you use to do this?

[rotund token]

unity actually has a tutorial for this (never looked at it)

https://learn.unity.com/tutorial/runtime-navmesh-generation

https://docs.unity3d.com/Packages/com.unity.ai.navigation@1.1/manual/index.html

but it's just all from this package

[timber ivy]

Do you know if it supports 2d navmesh?

my world is 3d but its a top down on a flat level so I could in theory just make the navmesh 2d

[rotund token]

there's no specific option i don't think but if you define your components so only the ground is walkable and everything else blocks

it'll just be a single ground mesh

[timber ivy]

Alright thank you

So since I am just generating a surface around the player could I just attach a navmesh surface component to the camera or does it actually have to be attached to the geometry I am generating for?

[rotund token]

can be anything

CollectObjects use volume

[timber ivy]

ah yep thats the one thanks

[timber ivy]

how do I enable this? the NavmeshSurface.BuildNavmesh isn't done async

do i just use this when I want to regenerate?

/// <summary> Rebuilds parts of an existing NavMesh in the regions of the scene where the objects have changed. </summary>
        /// <remarks> This operation is executed asynchronously. </remarks>
        /// <param name="data"> The NavMesh to update according to the changes in the scene. </param>
        /// <returns> A reference to the asynchronous coroutine that builds the NavMesh. </returns>
        public AsyncOperation UpdateNavMesh(NavMeshData data)

[rotund token]

that is async

        {
            var sources = CollectSources();

            // Use unscaled bounds - this differs in behaviour from e.g. collider components.
            // But is similar to reflection probe - and since navmesh data has no scaling support - it is the right choice here.
            var surfaceBounds = new Bounds(m_Center, Abs(m_Size));
            if (m_CollectObjects == CollectObjects.All || m_CollectObjects == CollectObjects.Children)
                surfaceBounds = CalculateWorldBounds(sources);

            return NavMeshBuilder.UpdateNavMeshDataAsync(data, GetBuildSettings(), sources, surfaceBounds);
        }```

it returns an async handle for you

nav mesh surface just provides a convenient wrapper for you

the underlying actual api if you want to write your own implementation
NavMeshBuilder.BuildNavMeshData
NavMeshBuilder.UpdateNavMeshData
NavMeshBuilder.UpdateNavMeshDataAsync
etc

[robust scaffold]

@rotund token Is this the right setup for a performance test?

[rotund token]

seems fine

i usually have my measures in separate methods

but whatever works

this might be nicer if it works

i am far from an expert on this package, there are a lot of tools in it that i am not familiar with that are probably really useful

[robust scaffold]

This is my first time writing unit tests so any knowledge is infinitely more than my knowledge

[robust scaffold]

Wow, performance test shows that PS operations is not always better: Another test shows 0 difference.

Ran it for twice as long. There is 0 difference

[drowsy pagoda]

Can someone give me a quick example on how to use ArchetypeChunk.GetComponentDataPtrRW()

[robust scaffold]

Seems self explanatory. Gets the pointer to the component data of entities within the chunk. Although there should be a component type for declaration for it.

Yea. It's basically GetNativeArray<>().GetUnsafePtr() you see in IJobEntityBatch.

[rotund token]

this seems like rotate0 is not used?

[robust scaffold]

in another job, I copied it there for a screenshot

Imagine if unity code gen worked:

And it's gonna get even longer. That's only a circle.

Imagine, if unity provided an "unsafe" IJobEntity code gen job type. Instead of handholding users with "safe" ref and in parameters that break burst vectorization, they pass the raw pointers and let the user deal with the safety. For example, my Vectorized() method should be the Execute() passed by the code gen. Everything else is boilerplate.

Well, code gen for IJobEntity doesnt work on ISystem anyways.

[robust scaffold]

TRS to Local to World. Slightly more readable with the one shipped with DOTS.

Fully vectorized. The WithoutRotation is actually unrolled by burst.

[rotund token]

why are you rewriting all of this 😅

thought i was going to be 2d or something at least

[rustic rain]

hmm, is there a way to register generic component type through attribute to that generic type, not to assembly?

[rustic rain]

Fatal error. Internal CLR error. (0x80131506)

bruh

don't mess with generic components

or else you get this

bruuuh

I have no idea what's up, it appeared after I added generic component

but now it won't be gone

even after full library folder removal

[rustic rain]

all right, I went fully rogue with New Input System and burst

kek

[rustic rain]

I assume RequireSingletonForUpdate ignores Disabled tagged entities?

[timber ivy]

so were you just awaiting it in a coroutine when you used it?

[rotund token]

All my stuff was in jobs at my renderers were render meshes and entities not game objects

But yeah I don't use this anymore as I use my own wrapper to recast as mentioned

[rotund token]

Yes

[timber ivy]

Any reason you prefer it over unity?

Doesn't appear UpdateNavMesh works with volume surfaces

Its just building a one piece mesh the same size/shap of the ground ignoring any buildings x)

[rotund token]

I have a lot more options

Also how I wrapped it and exposed it to unity still let me take individual methods from the native library and rewrite them in burst over time

[rotund token]

I'm not sure why it would be any different

You need to register what can be in your volume from memory

[rustic rain]

hmmm

having WAY too many systems is ok?

I talk about like 200-400

or more

With the way I decided to handle my Input

I'd need quite a lot of ISystems

to handle events

[rotund token]

we have about that many or more at work

main thread cost definitely adds up

strongly recommend making as many ISystem

imagine if they're only 0.1ms each

thats 40ms

or 25fps

[rustic rain]

hmm

they are quite faster

lemme actually check

[rotund token]

you say that now

but your project will get slower as you add more entities, systems and archetypes

[rustic rain]

actually they are not even running, kek

[rotund token]

usually 0.03ms is about as fast as SystemBase will run at for the most basic work

[rustic rain]

I forgot that they are reactive systems

that react on singleton events

[rustic rain]

so, what do you actually mean by that?

        [BurstCompatible]
        public static bool ToggleState<T>(this EntityManager em, in Entity stateEntity)
            where T : unmanaged, IComponentData
        {
            if (em.RemoveComponent<T>(stateEntity))
            {
                return false;
            }

            em.AddComponent<T>(stateEntity);
            return true;
        }

On average my reactive input handlers will just run smt hlike this in OnUpdate

and only once trigger fired

meaning like once per 10 seconds or smth

so the most of my concern - checks whether that system should run

    [UpdateInGroup(typeof(LateSimulationSystemGroup))]
    [BurstCompile]
    public struct PlaySimulationHandler : ISystem
    {
        private Entity _playSimulationState;

        public void OnCreate(ref SystemState state)
        {
            state.RequireSingletonForUpdate<GalaxyLoaded>();
            state.RequireSingletonForUpdate<Performed<TogglePlay>>();

            _playSimulationState = state.EntityManager.GetOrCreateSingletonEntity<Singleton<PlaySimulation>>();
        }

        public void OnDestroy(ref SystemState state) { }

        [BurstCompile]
        public void OnUpdate(ref SystemState state)
        {
            state.EntityManager.ToggleState<PlaySimulation>(_playSimulationState);
        }
    }

here one example of such system

[rotund token]

your system is still running

in the sense it has to look at a query

[rustic rain]

(this is for starting/pausing game)

[rustic rain]

yeah, so that kind of work will eat 0.03ms?

[rotund token]

simply checking if theres any entity in the query

[rustic rain]

hmm

[rotund token]

the Complete() that happens at the start of every system

[rustic rain]

I'd need to figure a way to group them

[rotund token]

actually nevermind on this

that doesn't happen anymore unless the system actually runs

but yeah if you're doing this in ISystem

it won't be that bad

[rustic rain]

are such checks bursted?

[rotund token]

i believe its in a function ptr

but its actually cached as well

theres been a lot of work on this

they are certainly trying to build entities to support this number

via ISystem

[rustic rain]

if it was me who did it I'd make a one bursted check that generates array of bools, so later it just does simple bool comparison

[rotund token]

really hope they have some improvements to ComponentSystemGroup in 1.0

[rustic rain]

whether system should update I mean

so, you mentioned that ISystem does query check faster

hmm

is it actually check one by one?

[rotund token]

it should be the same

[rustic rain]

in sync?

as in

[rotund token]

all it does to check a query is

public bool IsEmptyIgnoreFilter => _QueryData->GetMatchingChunkCache().Length == 0;

[rustic rain]

can I literally see in profiler that timeline

[rotund token]

        {
            if(!MatchingChunkCache.IsCacheValid)
                ChunkIterationUtility.RebuildChunkListCache((EntityQueryData*)UnsafeUtility.AddressOf(ref this));

            return MatchingChunkCache;
        }```

[rustic rain]

when it checks whether system should run?

[rotund token]

if you have a valid cache all it's really doing is a bool check and an int check

it's pretty fast

[rustic rain]

wdym valid cache?

how to keep it valid?

no structural changes?

[rotund token]

yes structural changes will invalid caches

[rustic rain]

hmmm

[rotund token]

(another reason static archetypes are important for performance)

[rustic rain]

this is all part of my main thread fast system group

in which all I do are structural changes

hmmm

that's no good I guess

meanwhile implementing ECB will probably be not as good

hmmmm

[rotund token]

i believe it has to be a structural change that affects this chunk

but i assume you aren't doing these changes every frame?

[rustic rain]

nnnah

[rotund token]

only when you make game state changes

[rustic rain]

I also made a some kind of pattern

where I use one entity per

toggle state

as in EnabledTag is switched on and off

on one entity

while it also has some other tag, that says that this Entity is this EnabledTag holder one

singletons basically

Trade off of memory for perfomance I guess

kek

[opal lynx]

I am using ecs 0.51 with unity 2020.3 2021.3.5f1
and it shows same 6 messages but I don't understand why

how can I fix for these messages?

[rotund token]

You're using a class not a struct

[opal lynx]

Thanks, oh god.........

[rustic rain]

hmmm, I wonder how I can implement dynamic building for required Query with multiple constraints

basically I have that same EntitySelectionSystem

and now it requires SelectionRequested singleton

if it exists, selection is performed and result is written to other component

but here's the thing

maybe instead of SelectionRequested I can implement a way to dynamically require any sort of components for selection to be performed

goal is that it can be done without changing types (since modders don't have access to adding interfaces to types)

[dreamy glade]

Is there any ways to have the entity manager itself hold custom data? I want to be able to reference a value that all components can read from. Thanks.

For instance, a global integer.

[rustic rain]

EntityManager is partial

you can reference assembly and add new fields to it

but I'm not sure

whether it's safe

[dreamy glade]

Then rather than an entity manager, how would you go about it?

[rustic rain]

I just avoid it as pattern

on the very low end

I use World for such modifications

it's managed anyways

[candid epoch]

You can just keep the data as a singleton entity. Or is there a reason why you don't want to do that?

[devout prairie]

If the value doesn't change you can use a blob asset, or as Dot suggested above a singleton entity with data holder component

[dreamy glade]

I've been trying to learn on how to utilize singletons. Regardless, I'm developing a voxel game where I want, essentially, a global **changing **variable that represents the player's most recent voxel chunk position. Then, the system can automatically regenerate the entities around the player.

[candid epoch]

Then a singleton entity is probably your best bet. It allows you to stay burst compatible

[dreamy glade]

And how would I go about access said entity when performing a 'foreach entity' call?

[candid epoch]

There's a convenience method GetSingleton which you can use to get the entity before for each

[dreamy glade]

I'll look into it, thank you for the help.

[candid epoch]

https://docs.unity3d.com/Packages/com.unity.entities@0.51/api/Unity.Entities.ComponentSystemBase.GetSingleton.html#Unity_Entities_ComponentSystemBase_GetSingleton__1

And then just use it inside your Entities.ForEach

[rustic rain]

if you singleton entity is meant to be never changing (as in, it exists as long as world exists)

you can use this poggie woggie approach

        [NotBurstCompatible]
        public static Entity GetOrCreateSingletonEntity<T>(this EntityManager em)
        {
            using var query = em.CreateEntityQuery(ComponentType.ReadOnly<T>());
            return query.IsEmptyIgnoreFilter ? em.CreateEntity(typeof(T)) : query.GetSingletonEntity();
        }

so you just keep Entity reference

in any system you want to access some data

and anytime you need it

you just do

EntityManager.GetComponentData<PlayerInput>(_inputEntity)

in case if you need it in jobs

you can access it through ComponentDataFromEntity

but sadly for best efficiency you'd need to use IJobEntityBatch

but if you really just want static value

burst supports SharedStatic<T>

[devout prairie]

what do you mean with this bit?

[rustic rain]

getting that value outside of job is sync point

[devout prairie]

( i've never used or looked at those jobs )

[rustic rain]

getting that value in any job but batch

equals getting that value for every entity

meanwhile in batch you access it once for chunk

[devout prairie]

ah right

so with for each, probably the best approach is just get/cache the value to a local variable first, and then pass that into the foreach right

[rustic rain]

well

in case where you need it in loop

you are simply doing smth wrong

kek

that's kind of against the pattern imo

[devout prairie]

not sure why you'd say that, it's often necessary to get data from elsewhere and use it inside foreach

say for example deltaTime, or some global value

you get it in OnUpdate, store it to a local value, then use it inside foreach

[rustic rain]

well, for time they literally built that into World

unmanaged one

hmmm

do you guys know any way to implement this:

I have a special singleton entity, which is loaded from subscene/async

because of conversion

I need to convert it

can't make it manually

and so as result I can't get reference to that entity at OnCreate

meanwhile that entity is basically static, once loaded never unloaded

Is there any solution to this?

goal is to avoid Getting singleton in loops

[devout prairie]

sometimes if i can't get something, due to ordering with creation of systems and mono's and conversion etc, i just put a hack inside OnUpdate like:
if (systemFieldEntity == Entity.Null) systemFieldEntity = //get entity code//

so often just fighting with how and when things are created etc, easier to just do that, and then maybe later fine tune it

just my 2c 😛

[rustic rain]

yeah, that's what I avoid...

OnUpdate is bursted

getting singletons is not possible here

[devout prairie]

i'm not sure i get what you mean

you're only actually getting it once right

[rustic rain]

yeah

and I can't use unburstable methods inside bursted method

even if it's behind if check

[devout prairie]

how then would you get a singleton inside OnUpdate?

[elfin spire]

i miss those, in early days we could access global variables which was way faster than copying large arrays

[rustic rain]

you are not copying large arrays tho

[elfin spire]

i'm not using the ECS part of dots, still using mono so i'm copying back and forth arrays to jobs

[rustic rain]

ah

that's not really what we talked about though

accessing data from jobs is equally troublesome

no matter if you use ECS or not

kek

[robust scaffold]

You bring up a good point. If Im reimplementing the transform system, I might as well roll my own 2D transform components as well. Trim down the number of not used properties.

[robust scaffold]

Im pretty sure you can now. Just cache the query containing the singleton at OnCreate().

Also, OnCreate() will get bursted when Unity finally makes entity query creation burst compatible.

[rustic rain]

huh

[robust scaffold]

Singletons were a pain point in 0.17. One of the improvements in 0.50 was burst compatible get singleton. Might be the only one in fact...

[rustic rain]

Is there an Entity journal yet?

[robust scaffold]

?

[rustic rain]

where you can see where certain entity was instantiated?

[robust scaffold]

Never used the journal before. So no clue.

[rustic rain]

@rotund token btw, I checked that approach of using DestroyEntity in order to clean it up from all components but system states
Here results:

every frame I destroy same entity

for 0.02ms perfomance

kind of curious whether I can optimize with query check

but I can't think of a way to check whether entity has any components but this

creating query with way too many component types also didn't really work

[drowsy pagoda]

Can I create BlobAssetReferences (that point to BlobArray wrappers) from within a Parallel job?

[robust scaffold]

No. The only memory you can allocate from within a job, parallel or not, is with Temp allocation. BAR with Temp will be deleted the next frame.

You might be able to create and allocate Blob builders outside a job, send it into a job to populate, move it out and then finish it off with create references outside the job.

[drowsy pagoda]

I see. Thanks for that.

[robust scaffold]

My inspector is getting really really long. Anyone have any tips on either a custom inspector or something to compress it?

[rustic rain]

split logic? xD

[robust scaffold]

no

18 components but each component is small:

[gusty comet]

I've just started to dabble with DOTS and I've noticed a convert to entity bool on the item inspector, is this meant to replace the ConvertToEntity script? if so, does anyone know where the API is to access it?

[rustic rain]

it's literally adding mono script automatically, kek

[gusty comet]

ok, so its a check-in-editor thing

?

[rustic rain]

pretty sure it's just Editor script that adds component to editor scene game object

[robust scaffold]

I am 99% sure that button is on the list to be killed by DOTS 1.0. Replaced entirely by subscene conversion.

That button is a carry over from... 0.12? 0.9? Ancient.

[gusty comet]

ok, ill avoid now and just do a layer and add the script in a foreach or something

[robust scaffold]

No no, you should use subscenes. Trust me, once you get it nailed down, it's pretty smooth.

[balmy thistle]

Word

[gusty comet]

Oh yeah, that automatically converts everything inside it to entities anyway right?

[rustic rain]

what's more

it fully serializes data

before you press run

[robust scaffold]

Yea. And helps with conversion with a bunch of side features.

[rustic rain]

making loading way faster

[gusty comet]

sweet, im planning a hybrid approach first, gonna convert all the static geometry to entities for a start

do normal raycasts work on entities?

[rustic rain]

no

nothing

[robust scaffold]

It's just a surprisingly advanced DOTS feature. Not really something that a beginner to the land of ECS or Unity's flavor of it can easily pick up.

[gusty comet]

ah, special API for it?

[rustic rain]

separate physics package

[robust scaffold]

Completely different physics engine.

Join me in coding your own physics engine. Free from capitalism unity interference.

[gusty comet]

so i wont be able to detect ground or shoot walls or anything using standard raycasts? or even have access to API that enables me to? 0-0

[rustic rain]

DOTS is still considered alpha, kek

[gusty comet]

unless i install physics that is

[robust scaffold]

Well, yes. If you completely convert your physics scripts to DOTS format.

[rustic rain]

a lot of things

A LOT

will have to be reimplemented from ground 0

[robust scaffold]

Are missing

Like rendering, if you hate the hybrid renderer like I do.

[rustic rain]

bruh

why so much hate?

even tertle's own fully bursted renderer had similiar perfomance as HR

[robust scaffold]

DOTS without HR = fast. DOTS with HR = slower than mono.

[rustic rain]

that's not true at all

[gusty comet]

i suppose i could use the seperated collider objects, since they are seperate from the ECS'd renderer, i can still detect the colliders using standard raycasts, I just wouldnt be able to do anything fancy like have specific sounds for specific surfaces.....

[robust scaffold]

Trust me. For something as big and integral as physics, you do not want to roll hybrid.

[rustic rain]

pretty much anything Hybrid = pain, imo

[gusty comet]

this is assuming im not using DOTS physics

[robust scaffold]

I tried with something as "simple" as 2D tilemaps. It's really clunky and slow.

So bad, I'm implementing my own 2D physics. DOTS integrated.

[rustic rain]

I decided to do pure ECS, so far I feel like it's the best decision

learning is though

but

I work on very simple game visually

so that might not apply to others, kek

[robust scaffold]

Yea, crossing over is just ugh. The one part I think is actually fairly decent is stock rendering as compute buffers and compute shaders have NativeArray hooks and can easily be attached.