archived-dots

[safe lintel]

@rustic rain I have this ancient code from my project, that I dont use anymore but this is how I set that up for sprite rendering

            var spriteArray = m_SpriteQuery.ToComponentDataArray<SpriteRenderData>(Allocator.TempJob);
            //var spriteEntityArray = m_SpriteQuery.ToEntityArray(Allocator.TempJob);
            // We need slices
            for (var i = 0; i < spriteArray.Length; i += SliceCount)
            {
                matrixList.Clear();
                uvList.Clear();
                m_MaterialPropertyBlock.Clear();
                
                var sliceSize    = math.min(spriteArray.Length - i, SliceCount);
                for (var j = 0; j < sliceSize; j++)
                {
                    var sprite = spriteArray[i + j];
                    matrixList.Add(sprite.Matrix);
                    uvList.Add(sprite.Uv);
                }
                
                m_MaterialPropertyBlock.SetVectorArray(m_mainTexUv, uvList);

                Graphics.DrawMeshInstanced(
                    m_Mesh,
                    0,
                    m_Material,
                    matrixList,
                    m_MaterialPropertyBlock,
                    ShadowCastingMode.Off,
                    false
                );
            }
            spriteArray.Dispose();

[rustic rain]

what is SpriteRenderData?

[deft stump]

yuuup

I forgot lmao XD

[safe lintel]

fairly certain I got it from https://coffeebraingames.wordpress.com/2020/04/05/one-million-sprites-more-than-120fps-dots-not-required/ but I think it was a forum post

    public struct SpriteRenderData : IComponentData
    {
        public Matrix4x4 Matrix;
        public Vector4 Uv;
    }

If i were to redo this that should be float4 and float4x4 now 🙂

[rustic rain]

hm

what is spriteArray in the first place?

Ok

It's array of individual components

oh I see

[safe lintel]

slicecount is 1023

[rustic rain]

I have my Material stored in shared component

it varies between entities

material is used to store data for shader and spritesheet itself

kind of global data per spritesheet

so I'm not sure how I can just loop through entities

and slice them without splitting them into chunks first

[safe lintel]

oh yeah, for that you need to iterate on the SharedComponent, get it in a similar way

[rustic rain]

which is what IBatchJob does perfectly

but how?

the only way I see is through IBatchJob

I just got a dirty idea in mind

what if I just

store all entities per batch lol

but then

that would be List of lists again

oooor

List of fixed lists

[safe lintel]

wait i have another system to do sprites but without animation

https://pastebin.com/37twp1CL

at some point I need to integrate the old stuff, so many things on the backburner

[rustic rain]

particleSharedRenderMeshes
What is this?

[safe lintel]

private List<SpriteRenderMesh> particleSharedRenderMeshes;

and sprite rendermesh is basically the same as rendermesh

[rustic rain]

so it's a collection of unique sharedcomponents?

[safe lintel]

just a reusable list

[rustic rain]

I'm interested in whether they are shared components

kek

cause rn it's the only thing I'm struggling to get outside of jobs

[safe lintel]

its shared public struct SpriteRenderMesh : ISharedComponentData, IEquatable<SpriteRenderMesh>

[rustic rain]

all right, that's good

let me go through that system again

hmm

[safe lintel]

the first loop of foreach actually is where the drawmeshinstanced happens

[rustic rain]

it's not instanced tho

oh

it's custom class

I see

[safe lintel]

https://pastebin.com/YzH8Uu4w

i got this idea from joachim's gpu animation(3d) system

[rustic rain]

can I get all entities

that have this chosen chared component?

[safe lintel]

that might be a more succinct example of using drawmeshinstanced if you want to dive into it

[rustic rain]

I'v done indirect instanced before

no worries kek

it's dots that makes me struggle rn

[safe lintel]

but i mean of using it with dots, and getting the arrays of entities

from the guy who is cto of unity

https://github.com/joeante/Unity.GPUAnimation/blob/fd196ce763eef3486e1a91c448ead2aa3f7f050a/com.unity.gpuanimation/Unity.GPUAnimation/GpuCharacterRenderSystem.cs#L194

his repo

[rustic rain]

I'll leave it for now, I need to make a simple rendering first, no need to go into more complex stuff

if I can't beat the easiest

so, is there a way to get all entities that have chosen shared component?

[deft stump]

what versions of packages do you have? I think the latest packages breaks compatability

[rustic rain]

I simply install .42 hybrid renderer

it installs all dependencies itself

[storm raptor]

dots

Lol

[deft stump]

So

Entities : 0.17.0-preview.42
Jobs : 0.8.0-preview.23
Burst : 1.6.4
Collections : 0.15.0-preview.21

[rustic rain]

What are Unsafe Containers?

What do they allow compared to normal ones?

[rustic rain]

could you also send GatherParticlesJob?

I think I'm starting to crack

how to do this

        Job.WithCode(() =>
        {
            var matrixHandle = GetComponentTypeHandle<LocalToWorld>(true);
            var animatorHandle = GetComponentTypeHandle<AnimatorSpritesheetComponent>(true);
            var spriteSheetHandle = GetSharedComponentTypeHandle<AnimatedSpritesheet>();
            var archetype = EntityManager.CreateArchetype(staticQuery);

            var chunks = query.CreateArchetypeChunkArray(Allocator.TempJob);
            for (int i = 0; i < chunks.Length; i++)
            {
                var batchInChunk = chunks[i];
                var ltw = batchInChunk.GetNativeArray(matrixHandle);
                var animators = batchInChunk.GetNativeArray(animatorHandle);
                var spriteSheet = batchInChunk.GetSharedComponentData(spriteSheetHandle, EntityManager);

                float[] indices = new float[animators.Length];
                Matrix4x4[] matrices = new Matrix4x4[animators.Length];
                for (int j = 0; j < animators.Length; j++)
                {
                    indices[j] = animators[j].currentIndex;
                    matrices[j] = ltw[j].Value;
                }
                MaterialPropertyBlock mtb = new MaterialPropertyBlock();
                mtb.SetFloatArray("_CutData", indices);

                Graphics.DrawMeshInstanced(mesh, 0, spriteSheet.material, matrices, animators.Length, mtb);
            }
            chunks.Dispose();

        }).WithoutBurst().Run();

All right, so far I got this
But I have no idea how to setup dependencies

since if I run it just like that

all those componentdata I access

gets invalidated

and it throws exceptions

[safe lintel]

        /// <summary>
        /// Todo later: figure out way to multithread this, assuming scheduling the drawmeshinstancedindirect can be figured out
        /// Job gathers particle buffers as native arrays, and adds them to a mega list. Jobs are separate to be able to observe performance
        /// cost in profiler, as well as leave door open to figuring out a better way to optimize the data transformations.
        /// </summary>
        [BurstCompile]
        struct GatherParticlesJob : IJob
        {
            public                             NativeList<Particle>              List;
            [ReadOnly]                  public BufferTypeHandle<DynamicParticle> ParticleBufferType;
            [DeallocateOnJobCompletion] public NativeArray<ArchetypeChunk>       Chunks;

            public void Execute()
            {
                for (int i = 0; i < Chunks.Length; i++)
                {
                    var chunk               = Chunks[i];
                    var particleBufferArray = chunk.GetBufferAccessor(ParticleBufferType);
                    for (var index = 0; index < particleBufferArray.Length; index++)
                    {
                        // Add reinterpreted to temp list
                        List.AddRange(particleBufferArray[index].AsNativeArray().Reinterpret_Temp<DynamicParticle,Particle>());
                    }
                }
            }
        }

i think this could be optimized

but also this is a bit specific to my own usage: particles are buffers on entities, as when I started this system I didnt want to touch EntityCommandBuffers when creating/destroying particles

[rustic rain]

nah, I only want to draw entities itself

so I think it's a bit simpler?

uugh

so you pass empty List to job and in job you fill it with data

[safe lintel]

pretty much

[rustic rain]

oh man, my brain is baked, but I want to try and finish it today lol

[safe lintel]

I still think the Job.Run might be a problem, i mean you are fairly close and its just one thing to rule out as far as potential problems

[rustic rain]

I tried without running at all

like

pure code in OnUpdate

I even put system in late update kek

[UpdateInGroup(typeof(PresentationSystemGroup))]

[safe lintel]

what was the error?

[safe lintel]

is that the error from Job.Run or just OnUpdate code?

[rustic rain]

both

it's same error

[safe lintel]

var archetype = EntityManager.CreateArchetype(staticQuery); try moving this outside of the job or to oncreate

[rustic rain]

wait a second

it's unused

huh

[safe lintel]

also I think you should just write it as:

OnUpdate() {
            var matrixHandle = GetComponentTypeHandle<LocalToWorld>(true);
            var animatorHandle = GetComponentTypeHandle<AnimatorSpritesheetComponent>(true);
            var spriteSheetHandle = GetSharedComponentTypeHandle<AnimatedSpritesheet>();
            var archetype = EntityManager.CreateArchetype(staticQuery);

            var chunks = query.CreateArchetypeChunkArray(Allocator.TempJob);
            for (int i = 0; i < chunks.Length; i++)
            {
                var batchInChunk = chunks[i];
                var ltw = batchInChunk.GetNativeArray(matrixHandle);
                var animators = batchInChunk.GetNativeArray(animatorHandle);
                var spriteSheet = batchInChunk.GetSharedComponentData(spriteSheetHandle, EntityManager);

                float[] indices = new float[animators.Length];
                Matrix4x4[] matrices = new Matrix4x4[animators.Length];
                for (int j = 0; j < animators.Length; j++)
                {
                    indices[j] = animators[j].currentIndex;
                    matrices[j] = ltw[j].Value;
                }
                MaterialPropertyBlock mtb = new MaterialPropertyBlock();
                mtb.SetFloatArray("_CutData", indices);

                Graphics.DrawMeshInstanced(mesh, 0, spriteSheet.material, matrices, animators.Length, mtb);
            }
            chunks.Dispose();
}

[rustic rain]

I'll comment it out for now

yep, progress

different error

at least now it tells what to do

I'm really uncertain how to create dependency

on smth that is not a job at all

oddly enough it does recognize archetype I use in system

[safe lintel]

I think it still is an issue where you are trying to use the job scheduler even with run, when it shouldn't be a job at all(in any case you cant use managed arrays with burst)

[rustic rain]

I'm fine with not using burst or jobs for rendering

since I can't pass native containers to draw call anyway

I do call it like you sent me btw

no job wrappers

just pure code in OnUpdate

[safe lintel]

ok how about

var renderMeshList = new List<RenderMesh>();
EntityManager.GetAllUniqueSharedComponentData(renderMeshList);
foreach (var renderMesh in renderMeshList)
{
    if(renderMesh.mesh==null)
        continue;
    query.SetSharedComponentFilter(renderMesh);
    var spriteAnimators = query.ToComponentDataArray<AnimatorSpritesheetComponent>(Allocator.TempJob);
    var localToWorlds   = query.ToComponentDataArray<LocalToWorld>(Allocator.TempJob);
    
    var indices  = new float[spriteAnimators.Length];
    var matrices = new Matrix4x4[spriteAnimators.Length];
    for (int j = 0; j < spriteAnimators.Length; j++)
    {
        indices[j]  = spriteAnimators[j].currentIndex;
        matrices[j] = localToWorlds[j].Value;
    }
    var mtb = new MaterialPropertyBlock();
    mtb.SetFloatArray("_CutData", indices);
    Graphics.DrawMeshInstanced(renderMesh.mesh, 0, renderMesh.material, matrices, spriteAnimators.Length, mtb);
}

note this doesnt address the 1023 limit for drawmeshinstanced

[rustic rain]

I cracked it

not entirely remember last thing I changed

xD

oh wait

now I do

[UpdateInGroup(typeof(PresentationSystemGroup), OrderLast = true)]

I order it to be the last

so it runs after HybridRenderer

which itself forces out all other jobs to complete

BUT

my instancing not working lol

it draws by individual calls

[safe lintel]

its always something 🙂

[rustic rain]

but not when I draw instanced

[safe lintel]

there is EntityManager.CompleteAllJobs but its a bit of a nuclear option and I wasnt sure if it was the chunk usage or something

[rustic rain]

Well CompleteALlJobs should happen at some point anyway

what is better moment

rather than drawing?

UNITY_DEFINE_INSTANCED_PROP(float, _CutData)
Do you happen to know

if I'm setting my data correctly?

mtb.SetFloatArray("_CutData", indices);

[safe lintel]

that part looks normal, id probably try to make a really simple test with a monobehaviour and render one sprite on its own(without dots, just draw meshinstanced just to be sure the shader works)

wait what about trying SetVectorArray? (edit forget this as you are indeed using a float array)

[rustic rain]

            for (int j = 0; j < animators.Length; j++)
            {
                indices[j] = animators[j].currentIndex;
                matrices[j] = ltw[j].Value;
                Graphics.DrawMesh(mesh, matrices[j], spriteSheet.material, 0);
            }

this works

like, I just put drawmesh call

inbetween

and it actually draws stuff

with same shader

[winter depot]

I just reread and watched all of the content in the https://learn.unity.com/course/dots-best-practices?uv=2020.1 course. ```I had already been working with dots for some time, but took a break due to frustrations with the idea of complexity vs scalability and how they can coexist. I kind of got most of my questions answered or at least clarified some things that I was foggy about before but just working through the motions.

My biggest question now is what is the official best practice on how to deal with the problem of heavy entities?
In the early parts of the course they are kind of adamant on using components themselves as tags for flags instead of using a component with a bool to early out of a system. Later they mention how each time you add/remove a component and a new archetype is created it lives forever as a chunk possibility, and the more variation of archetypes/chunks the less you are actually utilizing the cache friendly idea of ECS. At times they even say a better approach instead of frequent add/remove of components is to use a bool in a component as a flag.

Towards the end they briefly propose a solution is to break down your entities into sub entities and although they do not directly say how, I assume nesting where an entity points to another entity for say offensive info, and another entity for defensive info, etc. held inside of a component. If I am reading this correctly, that means that when actually dealing with these "parent" entities, we will need to use some sort of nested ComponentDataFromEntity reads right?

I feel like this is kind of the opposite direction of what they told us to do though, there just seems to be a lot of conflicting best practices in general. Does using the parent/child entities approach take up less overhead when it comes to reading the component data inside a loop than I am thinking?

Thanks for reading this wall of text, I just am hoping to finally get solid clarification.```

[safe lintel]

Are you actually running into issues with the amount of data for your entities?

[rustic rain]

I draw about 5 entities now xD

but it's about learning stuff

besides

I love gpu instancing

[winter depot]

I am concerned that as the project grows the variation of archetypes will grow considerably, and I am thinking I have been using components as tags too loosely. Its quite rare that I actually have the majority of archetypes being the same at any given time already. Between having components for things like AttackReady, Auras, MovementRequest, etc, I have insane variants. I assume maybe I should instead be creating entities and tagging them with references to the entities in question like in a DamageSytem instead of tagging the main entity itself, but then again I feel like its dirty because I am using more of the .EntityExists/ComponentDataFromEntity methods which I thought were reads that caused cache misses

[rustic rain]

all right, tried to do the same from mono - same thing

I guess smth wrong with shader?

[safe lintel]

@winter depot you can count the archetypes with something like

 var archetypes = new NativeList<EntityArchetype>(Allocator.TempJob);
 EntityManager.GetAllArchetypes(archetypes);
 Debug.Log(archetypes.Length);
 archetypes.Dispose();

I dont think theres an issue using exists/cdfe but I guess its fairly clear ideally you dont change components willy nilly, and they are doubling down on the enable/disable component state later on(some point prior to 1.0)

im currently at 600 for a simple fps personally, and trying to cut down my own use of component changes

[winter depot]

Yeah I was really really excited to see that they were working on the enable/disable component state. Thanks for taking the time to respond. I will check the amount at peak and get back to you shortly for your opinion. I guess I am just concerned in general because Mike was really adamantly against the notion that you shouldn't overoptimize early. I am just trying to not fall into that whole runs "good enough" approach early on, so later when it doesn't run so well its an entire refactor.

Also I am really hoping they get the package version synchronized soon, I can't stand that updating one package breaks all the others

[coarse turtle]

I get this notion, but I feel like the idea is really just to profile regularly and grab as much info you can to make decisions on which slice to refactor 🤔

like sometimes grab more detailed profiling data in builds so you can make better decisions instead of a major refactor. And to also test on your target hardware as a usual routine

[winter depot]

Right, but when it comes to where should the data live on ECS that's where I am most worried about my future. If I am doing things fundamentally wrong early in that regard I won't just have to optimize a few loops in a few spots, I will need to break every major system apart and go through days of rebuild to get it back together with references pointing everywhere.

[coarse turtle]

well that's where I think tools help w/ this. Things like windbg helps with seeing memory addresses and how your data is laid out

but yea I definitely get what you're saying tho. tbh I started w/ more or less just overanalyzing small problems and seeing how I can make whatever small problem faster, but that's the approach that works for me 🤷
I didn't really care if they were fundamentally right/wrong

I do wanna say for this:

Towards the end they briefly propose a solution is to break down your entities into sub entities and although they do not directly say how, I assume nesting where an entity points to another entity for say offensive info, and another entity for defensive info, etc. held inside of a component. If I am reading this correctly, that means that when actually dealing with these "parent" entities, we will need to use some sort of nested ComponentDataFromEntity reads right?
The way I take "sub entities" is entities associated via a key (could be an index or some other value)

that's the approach that I use for my own project instead of nesting entities (I prefer flat hierarchies when I can in my own game since it's easier for me to debug with w/e existing toolsets dots editor gives)

[winter depot]

So for an early on stage in my project that feels pretty simple with not much manipulation to entities I saw ~40 chunk types with ~100 entities. That's what is making me feel I am doing something very wrong.

[winter depot]

So instead of say OffensiveEntityComponent { entity = Entity} you have a key somewhere else? Or still in a component. I know in the debugger you can at least click an entity ref and it will take you to that entity

[coarse turtle]

i have a struct Key : IComponentData { public uint Value }

i do have things in "fixed" size so I know I will never exceed a certain value

and then entities are mapped by Key

query things in parallel, write to index & prepare data for some gamelogic

and schedule it

[winter depot]

What does that gain you as opposed to using Entity and linking it directly? Also this is for purely static data then?

[coarse turtle]

no statics, just all entity data

i don't need to link an entity if i can just read/write things in parallel to a container with a unique index as a key

[winter depot]

I think maybe I am just confused, that feels less "ECS" to me because then systems aren't really running stateless right? It feels like sort of taking out some of the magic and just writing stricter code with jobs

[devout prairie]

Is there any particular advantage to using a key rather than what Enraged suggested ( ie entity = myEntity inside the component ) ? I noticed 'nesting entities' mentioned above in the discussion.. my understanding is having an Entity field inside a component is essentially storing the entity 'id' not nesting or referencing - or am i wrong about that?

[coarse turtle]

i don't think there's any particular advantage tbh. 🤷 just made more sense to me to generate logic with unique indices over linking entities & public data

[coarse turtle]

don't know how it feels less ECS, I think I'm more just describing EC, your system does whatever it needs to do 🤔

[devout prairie]

my thinking has always been - Entity is i think a value type essentially, and is already and always a unique identifier for an Entity..

[winter depot]

No, I think your right, I think nesting was a poor choice of words. It just helps me kind of visualize what is occurring. There's no data in an entity after-all.

[devout prairie]

so i pretty much use components with Entity fields quite a lot.. for example in my RangeWeapon component it holds an 'Entity shootPos' field which i can then use as a key to grab the shootPos later

( probs cheaper to just store pos/rot of shootpos obviously but for now that's what i do 😛 )

[coarse turtle]

yea you're not wrong. The thing for me was that I needed to rely on open street map data. Linking entities together with open data is kind of a mess for that work I did. It'd mean querying and figuring out which entity I needed to sync together for data that came in asynchronously at varying points. To solve that, each entity was given a unique id. If another entity shared the same unique ID, then they were associated with each other.

[devout prairie]

ahh right yeah, that makes sense in that case - sounds interesting btw!

[winter depot]

Ok @devout prairie I think that is exactly the sort of scenario I was describing based on what the documentation kind of leaned towards. I was more so thinking that ALL the offensive data min/max/range/attacktype/etc is included in one component field named Offensive or something that just stores an entity, which then has those components tied to that entity linked.

So I think I am understanding right, but like I said I was kind of worried about the amount of lookups I would have to do to actually get that data in the end, and was that worth the breakdown of the "heavy entity"

[devout prairie]

ah yeah so separate data entities essentially

[coarse turtle]

yea it was a fun project (we had to write our own parser too which was fun). But with that I ended up carrying that idea into my own in house game w/ "weakly linked" entities

[devout prairie]

i do wonder what the consensus is on data heavy components, splitting them off onto separate entities, or even splitting the heavy components into separate components.. or even what is considered to be a heavy component.. for now, my core components all have a number of fields with different data types ( maybe 4-8 fields with ints/bools/float3's/whatever ) and i'm happy with it.. it's already complex enough building DOD/ECS systems in a complex game without adding additional layers of optimization complexity

[winter depot]

Yeah, that's exactly what I wanted to really learn when I started going over the content for "Best practices". I feel like they touch on so many things and sometimes extremely redundant or in depth, then just kind of make a quick pass at an actual practical use scenario of a best practice.

Complexity and Scalability are tightly coupled in my opinion

[coarse turtle]

oh for reference, in that project, we optimized our component sized

because we wanted to pack as many entities in a chunk

[devout prairie]

definitely interested in the idea of separate data entities though i have thought about this.. say an archetype/entity-array for just 'combat data' for each unit in the game, maybe a separate entity array for health data, etc.. but i'm not 100% if there's an advantage or not there

[coarse turtle]

but some entities were pretty large to the point we could only store a few hundred (maybe around 112 was the min iirc) into a single chunk

[winter depot]

This is the exact quote on how to solve the problem

This means the solution to the problem of heavy entities is to simply break them down into a larger number of lighter entities that can pack into chunks more efficiently. For example, if you divide your character’s components into different entities according to which SystemGroup processes them (for example, AI, pathfinding, physics, animation, rendering, etc.), those SystemGroups can iterate over chunks filled with components that contain only the data they need (for example, the chunks which the physics simulation access have room for more physics components because they aren’t also filled with AI state machine data).

[devout prairie]

yeah i suppose then it's maybe a difference between, are you looping 100k entities, or 10k, in terms of really superoptimizing chunks etc

[winter depot]

But as I said, my fear is that when I actually need to reference that data I am going down some nesting in a sense and I feel (maybe wrong) cache misses by doing this

[coarse turtle]

i think at max we only had to deal w/ 30,000 so we optimized for that

[devout prairie]

either way, just using dots/burst is going to be 100x faster i think

[coarse turtle]

but the median was about 12,000 entities

[devout prairie]

so i guess the returns on that level of optimization are only really relevant at quite large entity counts, because up to that point, you're already getting massive gains without really super-optimizing

i think this needs to be more clearly communicated and understood, because it's hard to know what's overkill optimization and what's 'more than enough' in ecs

and in what contexts

[winter depot]

I am not a forum post kind of guy, but I do think I need to go there for higher up clarity . I have a pretty lengthy set of notes that all have contradicting information from that best practices guide that I will try to lay out in a more concise manner and tag you when I post it so you can hopefully see the outcome.

[coarse turtle]

yea that's what we did with weakly linked entities, split off building data away from like traffic data

each chunk only had data relevant to what it needed for the heavy jobs

[devout prairie]

it's funny, ecs/dots has this dichotomy of being at the same time simpler/clearer and also more complex and difficult to reason out

[winter depot]

Major points summarized are they say avoid bools as tags/flags, instead use empty data components (future we get enable/disable state components), they also say do not have "heavy entities" because they cause too much fragmentation of chunks so you are causing cache misses. They also say avoid adding/removing components too often, if you find yourself doing this make bool flags instead. They then say avoid using the ComponentDataFromEntity too much, because it also can cause cache misses (need to find the direct wording they use in this one). They then say break down entities into more efficient chunks.

[coarse turtle]

yea ComponentDataFromEntity does random access

you might not be accessing an entity in the same chunk

[winter depot]

Right, which means there's a very fine balance where you have to choose a path, and neither path feel perfect.

It just feels like they teach this massive performance first idea. It really is great, and it can scale to insane proportions. So long as you limit complexity

[coarse turtle]

i will say im using ComponentDataFromEntity in my own physics systems cause im too lazy to optimize it 😛

[devout prairie]

i basically store uhh is it ComponentFromEntities and pass it into entities loops as readonly, or prebuild nativearrays with the data i want and pass that in.. i'm not sure exactly what is best but often there's just no other way of doing it

[coarse turtle]

yea - that makes sense, im just too lazy

and it runs decently on a few thousand projectiles on a core i3-5th gen

[devout prairie]

i only ever really add/remove components/tags on init though ( ie if i have to do some setup on some entity ), basically i understand it's slower than using the bool/flag approach

[winter depot]

Yeah, I think that's the right way to go about it, and I do this also for the most part when I actually do reference another Entity in a system. I think I need to see how big those prebuilts really are in my own project to see if its causing actual issues and if it can continue to be solid as complexity and variation grows

[winter depot]

This part is what puts me at analysis paralysis. Is it better for a system to be scheduled and all that data to populate to just early out, or better to never run at all?

The solution is we need those component enable/disable changes

[devout prairie]

For example i mentioned the 'Entity shootPos' field inside my RangeWeapon component.. so in my weapon firing system i'll first grab an EntityQuery and store all the shootPos Translations, then pass that into my main RangeWeapon loop and use RangeWeapon.shootPos as key to grab the Translation inside the loop

[devout prairie]

Yeahhh

[safe lintel]

I think: if you have several hundred entities its not a big deal to throw a bool into a component to just do a soft disable, but if you are dealing with 100k+ or some insanely large number, then a component change makes more sense to not even run in a query, but its all dependent on your own project's needs

it still comes back to profiling often to ensure there arent hiccups

[winter depot]

Ideally though, you run an entity query that checks the component data itself for that flag so that the system doesn't run if none of them match the flag. Then all of the headaches I have I think would dissapear

[safe lintel]

im just talking about workarounds for now as we dont have enable/disable

[winter depot]

Yeah, I got ya. Thanks for the discussion guys, I think I will start breaking my entities into smaller entities based on what I gathered here.

[safe lintel]

but what is your project anyway? kind of has a huge bearing on just how much optimization is needed

[winter depot]

Its an RTS

[safe lintel]

I would only do that if you are running into an actual issue though? what constitutes a heavy entity for you?

[devout prairie]

How about using a flag component with Change filter, i wonder if that would be a good compromise..

[coarse turtle]

i wonder if nordeus can give some insights: https://github.com/Unity-Technologies/UniteAustinTechnicalPresentation

It's a pretty old demo at this point, doubt it'll compile correctly on the 2020 LTS 🤔

[winter depot]

Well, I mentioned earlier I think I am making the entities heavy because of how often I am adding removing components just to get a system to run once. For example, adding AttackReady component to any entity that has certain parameters that determine if it should be attacking. Then removing that component once an attack entity is created and all the data for that interaction is populated.

I do this with just about everything, whereas I could always be creating micro entities to just schedule something once. I then worry about the overhead of creating so many entities so frequently

[safe lintel]

ah found this post by joachim on adding/removing components
https://forum.unity.com/threads/archetype-memory-layout-alternatives.1209346/#post-7724272

[winter depot]

Yeah, this is amazing. They really need to update the documentation and learning materials to reflect this change coming

[devout prairie]

definitely in my understanding, adding a component for something like attack seems like it might be less optimal, given that it could be adding removing that component several times in the space of say 30 seconds for potentially many 'units/characters' within the game

[safe lintel]

I dont view the adding/removing components to changing into separate entities to be solution to "heavy entities" necessarily. To me this is more an issue of state changing rather than a lot of data.
Like you could just keep a component that is queried in a system to see if that unit could attack. if you split this into another entity, its sort of tangential to the issue of adding/removing components.

[devout prairie]

essentially i think you're causing it to shuffle memory/arrays around constantly, rather than just, loop em all and check if they can attack this frame

[safe lintel]

in my mind, a heavy entity is one where you literally only have 3 entities per chunk

[winter depot]

Yeah, I think you are right

I guess I will have to do some math to see how heavy my entities would actually get later on

[devout prairie]

additionally you are forcing the creation of more archetypes based on components being added/removed especially if that's happening a lot with a wide range of different tags

[coarse turtle]

this is my network buffer entity, which im considering of nuking 🥲

[winter depot]

Like, for example I found doing Auras best if I use a buffer instead of many little aura components. This buffer could get pretty big since I want to hold all Aura data until they expire because of Aura levels of strength + duration reasons.

[safe lintel]

you can check this in the entity debugger, i sort of am nearing it for my weapon entities(7 per chunk if I am reading this correctly)

[winter depot]

Ok, I can clearly now see how badly having a fixed string is

[safe lintel]

also https://forum.unity.com/threads/chunk-info-ui-explained.586729/

[winter depot]

I think fixed strings are most of my heavy entities

I am going to switch how that's laid out, I guess blobs are the right direction

[safe lintel]

yeah can imagine they eat up a lot 🙂

[winter depot]

Thanks @safe lintel

[devout prairie]

that's interesting so are you using an active ragdoll variation of your ragdoll setup to bob the weapon around etc?

[safe lintel]

this is just the first person player weapon stuff, no ragdoll stuff for the player

[devout prairie]

yeah i was just inferring from some of field names - is it a physics spring setup with IK on the arms holding the weapon?

[safe lintel]

theres a spring system but its a ported version of the ufps spring system. my player weapon code started off from porting ufps to dots

[devout prairie]

ah right k

[safe lintel]

hm according to that thread I linked, joachim said 0.5 is already branched off

[winter depot]

Ah I didn't see enable/disable bits not til 1.0 dang

[winter depot]

So in refactoring to the new bit flags, I assume they are going adjust the whole .WithNone, .WithAll, etc and they will add .WithEnabled/.WithDisabled or something like this. For the time being I guess this boilerplate should mostly just be commented out and rewritten for me and I just check the bool flag I am manually making. This sound like the solution most are going with in preparation?

[deft stump]

wait so enable/disable is otw in 0.5?

[safe lintel]

No @deft stump

[deft stump]

damn

my hope has been crushed again

[safe lintel]

tbh never even thought about it @winter depot

[deft stump]

trying to make my own conversion system.
Looking at the Entity Debugger, unity's own Convert To Entity just runs always?

[coarse turtle]

yea, but it's only for runtime conversion

[river meteor]

hey question for dots people in here

say i started up a URP project with dots, and just created a spawner that spammed capsules with physics bodies and physics shapes and nothing else. what would be the expected max # of capsules you'd expect to see before <60fps frame drops?

[craggy orbit]

i was able to get around 20k entities before any frame drops. i did have lots of additions on the CPU and GPU side though. you could probably get a lot more if you're not doing any complex work (sounds like you're not)

[river meteor]

i have 5k

and i'm down to 20fps with that

so i tried to figure out what i was doing wrong, i assume it's logically

but the person i was talking to lastnight was just like suggesting project configurations endlessly and i'm like i'm pretty sure it's not that lol

do you have an example of some boilerplate you'd use for a simple spawner? i'm sure i'm missing something simple

so here is my bullet prefab setup

old picture (i've since removed the capsule collider)

https://paste.myst.rs/a5cwsw0b and this is my spawning system

i've enabled gpu instancing on the material, i've enabled hybrid rendering v2 in the scriptping symbols

[worn valley]

Have you disabled all the safety stuff in the editor or run it in a non dev build?

[rustic rain]

Do you guys know whether there's any reason to use SRP?
Cause it appears SRP is incompatible with GPU instancing

[half jay]

Any tips what can cause this perf issue with physics on android? Any tips what should i check first?

[north bay]

Are you using IL2CPP?

The ms counts are pretty crazy but I wouldn't focus solely on the fixed-step simulation group that you have highlighted in that screenshot, it is running 19 times per frame to catchup because the FPS are that low
Better look what besides it is taking that much of your frame time

[half jay]

Yes im using IL2CPP. Im a bit confused because i see only this place with huge numbers.

[half jay]

one think that i can also add that in scene it was about 2000 static colliders without any rigidbody and also im not detecting their hitting

[rustic rain]

why do you think it's an issue in the first place?

maybe that's intended perfomance on your device

[half jay]

Im not sure just said some additional facts that may be will help

[rustic rain]

bruh, hybrid renderer sucks

I hope to see between URP and HDRP another one

ERP

kek

I can only imagine how perfomant it'd be, fully gpu instanced

[river meteor]

yeah i built a regular build and ran it. same thing

[safe lintel]

@river meteor as you are using ecb's you can schedule that with burst

[river meteor]

ecb?

[safe lintel]

just after the foreach, use _commandBufferSystem.AddJobHandleForProducer(spawnJobHandle);

entity command buffers

well you cant use UnityEngine.Random in that though

there is the Unity.Mathematics.Random for burst jobs

[river meteor]

oh ok. i did modify it to use Mathematics.Random I think

that's a good point though. i might have been previously screwing myself so it wouldn't run unless i put withoutburst

[rustic rain]

Is there a way to create authoring component that runs code inside of editor?

I want to make preview in scene view for my custom rendered sprites

[karmic basin]

Yes ? Nothing to do with DOTS though. Read the manual to learn more about editor scripting. Then people in #↕️┃editor-extensions would probably help

[safe lintel]

just use a monobehaviour @rustic rain

[prisma anchor]

I was trying to prototype my level with ProBuilder, but I can't get the player to collide unlike a regular primitive. Does ProBuilder work with Unity Physics?

[stone osprey]

Quick dumb question here...

Is it "okay" to cache entities ? Or should we query for them everytime we need a certain one ?

[safe lintel]

ive found most issues get resolved when you switch mesh colliders to physic shape equivalents, dunno whats happening behind the scenes but meshcollider to me is somewhat unreliable @prisma anchor

[worn valley]

Oh weird

[river meteor]

EntitySpawnerSystem.cs(29,13): error DC0001: Entities.ForEach Lambda expression uses field '_random in a reference type'. Either assign the field to a local outside of the lambda expression and use that instead, or use .WithoutBurst() and .Run()

i'm not sure what this means. so i shouldn't get random values inside the lambda?

wouldn't the random values be the same for every object then?

[prisma anchor]

Thanks fam. I was using the Physics shape equivalent, It just doesn't seem to wokr

[safe lintel]

@river meteor can try

OnUpdate() {
var r = _random;
Entities.ForEach
}

[river meteor]

oh ok. what's the appropriate way to seed this version of random?

[safe lintel]

@prisma anchor and the appropriate mesh, and the mesh has read write enabled?

[river meteor]

wrong chat sorry lol

[worn valley]

Lol that has nothing to do with random. 😛

You need a randomizer that can be used in parallel. I think unity's new math library has something like that

You are trying to bring in a reference type to a strut which can't access that code during the job. Does that make sense?

[river meteor]

yes it makes sense

so i'm moving the references to _random outside of the lambda

but that means every entity is going to receive identical random values

that's what doesn't make sense

[worn valley]

You need to do the randomization inside the lambda which means you need to use math.random I THINK

[prisma anchor]

Rubber Ducky moment: I forgot to add Convert Entity....

[worn valley]

I forget how that works in burst

[river meteor]

you mean Unity.Mathematics.Random?

that's what i'm using

[winter depot]

I thought you stored random in a component on each entity that’s going to use it so the seeding works

However, the simplest way to generate good quality random numbers is simply to have one instance of Random per Entity, stored in a component. Jobs can use this instance to generate any random numbers which are required for data transformations on a particular entity, and the state will be unique and persist as long as its component exists.

[karmic basin]

It's okay to cache an Entity as long as it is still alive. If it's supposed to die at some point you might surprise yourself with an outdated reference

[rustic rain]

How to exclude certain components from conversion completely?

is there such option?

I mean native components like renderer and etc

[frosty siren]

you can implement your own GameObjectConversionSystem and use update in group attribute GameObjectAfterConversionGroup and just remove component you want to exclude

[rustic rain]

huh

btw

about your 2d renderer

is it based on built in or?

I'm really uncertain atm whether you can create gpu instanced shaders in SRP

[unkempt abyss]

(0,0): Burst error BC1064: Unsupported parameter `Unity.Mathematics.float3` `position` in function `(blah).Invoke(Unity.Mathematics.float3 position, float time, float vertexColor, void* payload)`: structs cannot be passed to or returned from external functions in burst. To fix this issue, use a reference or pointer.
just wanted to double check - I can't even pass a float3 through? so I should just pass in 3 floats seperately? 🤔

[rustic rain]

void* payload?

[unkempt abyss]

ignore that part 😄 that bits working as expected 🥲

[rustic rain]

what about return type?

allthough...

float3 is indeed a struct

[unkempt abyss]

good point, return type is float3 too

yeah just kinda thought it would handle float3 since it's all blittable and stuff 🤔

[rustic rain]

I wonder how to use pointer in this case

but reference would be way easier

[unkempt abyss]

bah yeah, return type is a problem 😦

[rustic rain]

just do ref

[unkempt abyss]

yeah think i'm gonna have to

[rustic rain]

Is there a way to keep resizable collection of entities prefabs?

so you can kinda keep serialized list in your game object singleton

and it will create same list but of entities

[karmic basin]

Dynamic buffer

[river meteor]

that's actually a good idea i didn't think of. so it just comes through in the lambda yeah

[worn valley]

Are you trying to get a float3 out of a for each loop? You may have to pass out a one length native array.

[rustic rain]

Is there any need to keep blob asset store that is used during manual game object conversion? Or should dispose of it as soon as I got my entity?

[rustic rain]

is Dependency property after running Schedule on job in SystemBase contains job handle of that scheduled job?

I want to add my own system as dependency to other system

[tepid abyss]

Hey, is there a way to edit entities during play mode? all fields seem to read only in the inspector

[rustic rain]

sadly nope

[north bay]

0.50 will finally allow you to edit entity fields in the inspector...

But who knows when that will drop

[rustic rain]

in 2028

[rustic rain]

hmmm

any idea how to cause triggers

to run main thread code

from inside job?

        Dependency = Entities
            .WithReadOnly(interactables)
            .WithReadOnly(ground)
            .ForEach((ref DynamicBuffer<CollisionEvent> collisionBuffer, ref PlayerVelocity playerVelocity) =>
        {
            if (!collisionBuffer.IsEmpty)
            {
                var hit = collisionBuffer[0].colliderHit;
                if (interactables.HasComponent(hit.Entity))
                {
                    var data = interactables[hit.Entity];
                    playerVelocity.value.x += data.velocityBoost;
                    playerVelocity.value.y = 20f;
                    buffer.RemoveComponent<PhysicsColliderBlob>(hit.Entity);
                }
                else if (ground.HasComponent(hit.Entity))
                {
                    // Game should be over
                }

                collisionBuffer.Clear();
            }
        }).Schedule(collisionHandle);

in that commented area

I need to run

        shootCannonSystem.Reset();
        GameState gameState = new GameState
        {
            state = GameStatus.OnStart
        };
        SetSingleton(gameState);

such code

where Reset() is also interacting with EntityManager

I don't care at what point code will run, as long as it will run in current/next frame

But I need to avoid sync point

where I have to do Run()

I guess I could add another buffer for Events kek

feels a little bad tho

that I'll have to create system update that will query through every frame

just to know whether something is triggered

[deft stump]

It's not actually.
it's the intended approach.

[rustic rain]

so they want me to query through?

[rustic rain]

wait what

SetSingleton changes original entity?

        gameStateEntity = EntityManager.CreateEntity();
        GameState gameState = new GameState
        {
            state = GameStatus.OnStart
        };
        EntityManager.AddComponentData(gameStateEntity, gameState);

I have this code in OnCreate()

oh wait nvm

I referenced wrong entity

[deft stump]

a system that queries and checks every frame if something happened.
that's the intended approach

[rustic rain]

I wonder how it works internally

does it actually checks queries every update

or

has some sort of event system during structural changes

[deft stump]

it queries every update

and then do work

[deft stump]

structural changes is done via ecb

[rustic rain]

I mean

what is inside of all those structural changes

I'd guess they internally disable certain systems during structural changes

kinda like put bool on query

whether it's active or not

[deft stump]

???????
all it does is add / remove stuff to the internal database

[rustic rain]

if that's true then it's a waste

since queries are known whether they should run or not after structural cahnges

and I can't think of a way for empty query to run without another structural change

tbh, I think it does work like I think

since queries have property

IsEmpty

which is bool pointer

how to correctly get ComponentType of DynamicBuffer?
ComponentType.ReadWrite(typeof(DynamicBuffer<CollisionEvent>))

this gives me an error

nvm

you need to reference buffer element type

[gusty comet]

Does DynamicBuffer.Reinterpret<T> have any sort of safety mechanism? Like, if the input struct is

public struct Test1
{
  public Entity entity;
  public float2 location;
}

and the output type is

public struct Test1
{
  public float2 location;
  public Entity entity;
}

will an exception be thrown, or will the data be incorrectly casted silently?

[rustic rain]

nnnope

no checks

[worn valley]

Roll the dice.

[rustic rain]

GetComponentDataFromEntity<Translation>(true)
I assume reading it outside of job is unsafe?

[rotund token]

Yeah you'd need to complete the dependency first to be sure

Outside of a job you should just use entity manager

[rustic rain]

oooh

I forgot about using EM

kek

[gusty comet]

Thanks

[tepid abyss]

Hey, I am moving a lot of entities towards a target, and I am trying to avoid them being overlapped, is there a way for me to access the Translation of other entities inside Entities.ForEach (of the same Archetype)?
(Or if anyone has an efficient way to avoid overlapping sprites in 2D without using the physics engine?)

[worn valley]

Are you changing the translation at the same time you are looking at it for each entity?

[tepid abyss]

No, I only change its velocity

[worn valley]

Cause you should be able to grab the translation component within the lambda

If you only want the same archetype you should use whatever Chunk for each is now using

Like check the chunk

[tepid abyss]

That way I could access other entities translations?

within the same foreach loop

[worn valley]

Yup. You should think about in the instance you are doing a forEach loop what is being read, what is being written and what is being both read and written

So if you are only reading things it is safe to read whatever

I believe you use the IN on the arguments in the lambda

that means it only reads and doesn't write

[tepid abyss]

Yeah okay I just read IJobChunk (Just started using DOTS), one thing is bothering me, I want to check for collisions and "push" away the entities from the collider, the IJobChunk iterate in chunks, what if one entity is colliding with another and they are in two different chunks?

[worn valley]

You can also get an array that is referenced by entity name

I did a lot before they added the lambdas so I can't tell you how t get that info

Then you want to use the getComponent by entity reference

It's slower because it's random access though but will still be pretty fast

Sorry I can't remember the function call off the top of my head. It's something like GetComponentDataFromEntity

[tepid abyss]

Yeah I tried using it but it kept throwing exceptions because I indexed into another chunk

[worn valley]

Oh weird that should be fine. Maybe something has changed.

In a previous job you could pull all the data you need into a native array

And then pass that into the next job

[tepid abyss]

Wait, I did something wrong on my end, I didn't specify ReadOnly on the entity list

Now it is working as intented! 😄

[worn valley]

There it is!

[tepid abyss]

thank you!

[worn valley]

Yay

I find ECS and DOTS is like 90% figuring out how to get the data you need

And not the actual code

[tepid abyss]

var enemyArray = GetEntityQuery(typeof(EnemyTag)).ToEntityArray(Allocator.TempJob);
needed to grab the enemy array and mark it as ReadOnly to iterate over it

Yeah, I am still new to this

[worn valley]

That isn't the function I remember but they may have changed it

[tepid abyss]

That's another one that I needed to use (There might be more that I still don't know of), but I did use the one you specify to get the Translation's

[worn valley]

Cool. I find the translations weird because they reference the LocalToWorld instead of being the actual current data. So sometimes you can get weirdness because local to world and translation is out of sync

I ended up dumping translation and rotation because you can get that info from LocalToWorld except it's a little more complex.

But you don't have to worry about translation not updating LocalToWorld at the same time

[prisma anchor]

Is there a good technique for one-off systems to be invoked? like spawn object or create X. Right now I create data with the system, and delete that entity

[worn valley]

You can use entities with components to do this. You create an entity and add a component that is something like "SpawnNewBoss" and a system that requires that component to operate. At the end of the frame destroy the component or entity you made and the system stops running

That's roughly how the DOTS net stuff works for messages

[prisma anchor]

so pretty much what i do now

[worn valley]

Simplest solution is sometimes right

[prisma anchor]

yea

[worn valley]

The only downside is you are creating sync points but you will be doing that anyways by creating new entities

And you can use an entity command buffer to create and destroy the entities later and in parallel

[prisma anchor]

ok thanks!

[rustic rain]

You get component data from entity (system method) and iterate through it using another entity as index. Doing it yourself in scalable way would a tough task tho. Have you considered using physics package? (At least just for collision detection)

[rustic rain]

You guys

is there particles in dots?

I see so many mentions

but not any actual tutorials

oh nvm, i finally found unity manual

[rustic rain]

bruh, it's onlyy for Tiny

[rustic rain]

I guess in order to create my own particle system

I'd just need to create particle prefab

and authoring component that stores reference to it and settings (like, how long it lives, how many and etc...)?

[still pewter]

Hi everyone, I'm trying to make a 3D following system, like a homing missle. Simple enough in default Unity, but with Unity.Mathematics it's a bit trickier I find. I got it working in 2D but when it's 3D cross vectors and quaternions make my brain all jumbled and I'm afraid I'm making this way complicated than it actually is.
Has anyone gone before me? Any advice or tips or code?

[rustic rain]

it's all the same as normal unity 3d space

but

for some reason quaternions in dots are using different euler conversion

not even sure, but so far

all you need to figure out is your correct euler angle in code if you are converting between those and quaternions

otherwise it's all the same

            .ForEach((ref Rotation rotation, in LocalToWorld ltw) =>
            {
                //Modify holder's X rotation
                Quaternion rot = rotation.Value;
                float x = rot.eulerAngles.x - 15f * comp.deltaMouse.y * delta;
                //Angles are weird in dots, requires complex clamping
                x = (x > 200) ? math.clamp(x, 269.9f, 370f) : math.clamp(x, -10f, 90.1f);
                rotation.Value = Quaternion.Euler(x, 0, 0);

            }).ScheduleParallel();

here example of my mouse X axis clamping (up/down)

in classic unity I'd just do

Mathf.Clamp(transform.rotation.eulerAngles.x, -90f, 90f);

but in dots I need to do this nonsense kek

[vagrant lotus]

dots use radian

[rustic rain]

dots has no euler conversion

I use Quaternion's one

[vagrant lotus]

lower case quaternion have

[rustic rain]

ahem

it has conversion from euler

not to

[vagrant lotus]

oh

thats cause ambiguity when doing so

[rustic rain]

maybe, but so far it's the only way

unless you implement

to euler conversion yourself

[vagrant lotus]

ye

[rustic rain]

which I wasn't looking for

maybe you happen to know any other way to clamp only certain axis without conversion to euler? xD

actually the question I was looking for a long time

[vagrant lotus]

are you trying to rotate something along an axis?

[rustic rain]

it's for mouse rotation

up and down

so you are clamped

[vagrant lotus]

quaternion.axisangle?

you can multiply 2 quaternion together

[rustic rain]

I know how to rotate smth with quaternions

but how you rotate clamped

so you don't rotate over certain angle

[vagrant lotus]

oh

[rustic rain]

basically so you won't be able to do somersault with mouse kek

[rotund token]

are you looking for a free look camera?

because yeah if you need to clamp, just convert back to euler

var localEulerAngles = mathex.ToEuler(rotation.Value);
var look = math.radians(input.LookRequireEnable()) * camera.LookSpeed;
var newRotationY = localEulerAngles.y + look.x;
var newRotationX = localEulerAngles.x - look.y;
newRotationX = math.clamp(newRotationX, math.radians(camera.XRotationMin), math.radians(camera.XRotationMax));
rotation.Value = quaternion.Euler(newRotationX, newRotationY, localEulerAngles.z);```
my code for clamping my free look rotation

[vagrant lotus]

mathex?

[rotund token]

oh thats probably from my core library

[vagrant lotus]

cool

[rotund token]

yeah looking at it

it's the ToEuler from the Unity.Physics package

[vagrant lotus]

still a valid dots only solution

👍

[rustic rain]

phyiscs?

huh

I don't get it

where do I get it

what is mathex?

[rotund token]

mathex is just a class in my core library

the ToEuler just had a comment i wrote

        /// <remarks> Taken from Unity.Physics. </remarks>
        /// <param name="q"> The quaternion. </param>
        /// <param name="order"> The winding order. </param>
        /// <returns> Euler angles in radians. </returns>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public static float3 ToEuler(quaternion q, math.RotationOrder order = math.RotationOrder.Default)```

i just took this method from the Unity.Physics package

[rustic rain]

is that all?

[rotund token]

?

[rustic rain]

[MethodImpl(MethodImplOptions.AggressiveInlining)]
        public static float3 ToEuler(quaternion q, math.RotationOrder order = math.RotationOrder.Default)

all you need to define?

[rotund token]

no

the method is huge

i can't post it here too long

[vagrant lotus]

which version of unity.physics

[rotund token]

hence im saying if you want it, i literally just copied it from unity physics somewhere and you could find it there as well

[rotund token]

whatever the latest is

[vagrant lotus]

under which class?

(I might be blind)

[rustic rain]

could you just send file txt here? xD

I promise to repost it for future generations kek

[vagrant lotus]

kek

[vagrant lotus]

https://forum.unity.com/threads/is-there-a-conversion-method-from-quaternion-to-euler.624007/

one of the last comments

[rustic rain]

When I'm doing this but without auto generate, is there a way to ensure my entity prefab will be converted only once?

So when I put my autohring mono on several GO

and add to them same GAmeObject as prefab

all of them will be same entity in the end

not multiple equal entities instead

[rustic rain]

if anyone wonders about question above: IDeclareReferencedPrefabs

Soooo

Do you guys know any better idea how to resolve trigger events on collisions?

rn I do this by quering through dynamic buffer with collision events elements

but in the end I clear buffer

so it only triggers once

and if I want to add another trigger related to collision

that would mean I need to expand my query

aaand the list can go on

Meanwhile I kind of want to avoid creating such God System

that takes 15 components to ref

hmmm

Can multiple same components be added to same entity?

[rotund token]

No

[rustic rain]

if I try to do that through buffer or entity manager I'll get error or?

I just want to avoid exceptions in this case

[rustic rain]

        var buffer = bufferSystem.CreateCommandBuffer().AsParallelWriter();
        Entities.WithAll<SpawnParticles>().ForEach((Entity entity, int entityInQueryIndex, in ParticleSystem particleSystem, in LocalToWorld ltw) =>
        {
            Random random = Random.CreateFromIndex((uint)entityInQueryIndex);
            for (int i = 0; i < particleSystem.numberToSpawn; i++)
            {

                Entity newParticle = buffer.Instantiate(entityInQueryIndex, particleSystem.particlePrefab);
                Particle comp = new Particle
                {
                    timeLeftToLive = random.NextFloat(particleSystem.minLifetime, particleSystem.maxLifetime),
                };
                buffer.SetComponent(entityInQueryIndex, newParticle, comp);
                Translation translation = new Translation
                {
                    Value = ltw.Position,
                };
                buffer.SetComponent(entityInQueryIndex, newParticle, translation);
            }
            buffer.RemoveComponent<SpawnParticles>(entityInQueryIndex, entity);
        }).ScheduleParallel();
        bufferSystem.AddJobHandleForProducer(Dependency);

Any idea what I'm doing wrong?

oh looks like I didn't add component to prefab

[safe lintel]

anyone find the rotationeulerXXX components dont update in the inspector?

[rustic rain]

there are no such components in base entities tho

[safe lintel]

by XXX I meant all the combinations of XYZ/ZYX/YXZ etc, and its in the transforms namespace

[rustic rain]

oh

Are there any system to update them then?

[safe lintel]

its part of the builtin transform system, euler types dont seem to update in the inspector though(also shows as a 4x4)

[rustic rain]

you can check used by systems

in inspector

[safe lintel]

I mean its actually working and rotating, but the values only report as 0, even when they are not 0

[karmic basin]

com.unity.dots.editor have these : class RotationEulerXYZInspector : Float4x4ValueInspector<RotationEulerXYZ> I guess they should be involved in the editor field drawing ? 🤷‍♂️

maybe you found a bug 😁

[pliant pike]

I find that components in general don't update in the inspector you have to click on and off them and then maybe after a while they might update

[safe lintel]

@karmic basin thanks, if you replace it with Float3ValueInspector it works. Also NonUniformScale is 4x4 as well 🤔

[robust scaffold]

ugh, reopened my ongoing dots project for the first time in months and god damn it nothing works. thats what I get for following 2022b...

time to get everything working again, pain

[gusty comet]

hello, I recently started studying dots and I had a question: everywhere it says that you need to call dispose on NativeArray, but if my Job uses NativeArray to get data from entities, do I also need to Dispose it?

public class MovementSystem : SystemBase
{
    struct MoveJob : IJobEntityBatch
    {
        public float DeltaTime;
        public ComponentTypeHandle<Translation> TranslationTypeHandle;
        [ReadOnly] public ComponentTypeHandle<Rotation> RotationTypeHandle;
        [ReadOnly] public ComponentTypeHandle<PlayerTag> TagTypeHandle;
        public void Execute(ArchetypeChunk batchInChunk, int batchIndex)
        {
                NativeArray<Translation> translations = batchInChunk.GetNativeArray(TranslationTypeHandle);
                NativeArray<Rotation> rotations = batchInChunk.GetNativeArray(RotationTypeHandle);
                for (int i = 0; i < batchInChunk.Count; i++)
                {
                    translations[i] = new Translation { Value = translations[i].Value + (math.mul(rotations[i].Value, new float3(0, 0, 1)) * DeltaTime) };
                }
                translations.Dispose() //?
                rotations.Dispose() //?
        }
    }
}

[robust scaffold]

No. You do not need the Dispose() for NA obtained from ArchetypeChunks. Those are managed by Unity in the back.

Manual deallocation of NativeArrays only need to be done for NA you create / allocate yourself,.

[karmic basin]

Did you have the chance to see if any other 4x4 is broken ?

[gusty comet]

Thank you

[karmic basin]

Weird. We might touching the experimental part here :)

[safe lintel]

well localtoworld 4x4 works as expected, I expect its just where the actual data is float3 and its a 4x4 inspector it wasnt showing correctly

[karmic basin]

Ooooh then maybe they started as a 4x4 because of quaternions but they forgot to switch it for the euler representation ones, do you think ?

[safe lintel]

quite possible

@pliant pike I have experienced that in the past

[robust scaffold]

Is that with the fancy inspector or the built in one? I've never experienced any delay in updating entity values with the editor built in inspector.

[robust scaffold]

As a side note actually, the built in entities inspector does occasionally crash my editor when I attempt to look at a world that updates asynchronously from main thread graphics due to use of exclusive entities transaction invalidating access to external entity manager. Very annoying.

I think it's due to overloading the log due to large amount of stack trace dumps every frame.

[hollow jolt]

what does it mean to assign it directly ?

ok i think i found it

[pastel field]

If DOTS Animation is inaccessible in Unity 2021.X (dependencies break so no AnimationCurveBlob) what is the solution today to use AnimationCurve in ECS?

We can find 2 or 3 years old codes that reproduce AnimationCurve with weights and wrap modes (https://forum.unity.com/threads/need-way-to-evaluate-animationcurve-in-the-job.532149/), but they are on average 10 to 20 times less efficient than the "original code".
For the moment I use ISharedComponentData on the main thread, but something tells me that this is not ideal.

[tepid abyss]

Hey, I am trying to create a custom shader graph for Hybrid Renderer V2, I followed multiple instructions to make sure I didn't mess up anything, the renderer output error:
A Hybrid Renderer V2 batch is using the shader "Shader Graphs/TestShader", but the shader is either not compatible with Hybrid Renderer V2, is missing the DOTS_INSTANCING_ON variant, or there is a problem with the DOTS_INSTANCING_ON variant.

I am using Sprite Unlit in the shader graph, don't know if that matters.

(Also, I deleted the ShaderCache and reopend Unity after declaring the HYBRID_V2 define)

The generated shader file doesn't seem to be containing the DOTS_INSTANCING_ON define so the error is correct, but why it didn't add it though

Oh, when I switch to Lit it does work, so the problem is with Sprite Un/Lit

And it also seems that URP Material Property Base Color doesn't work on Sprite Renderer, I looked at the code and they querying for Renderer, so it should have worked, but doesn't

[tepid abyss]

Okay, the URP Material thing is just because the Sprite Un/Lit use _Color and not _ColorBase, so I need to create a custom authoring script with the MaterialProperty attribute

I was wrong, its not just the property name, the whole system doesn't work with Sprite Un/Lit shader graphs it seems, and not with SpriteRenderer (Used another mesh with MeshRenderer and the same material + authoring scripts (Urp Material Color) and it seems to work)

[rustic rain]

hm

When you destroy entitiy, is there a way to specify to also destroy all it's children?

transform children I mean

[rotund token]

there's a linked buffer for this built in

[candid epoch]

how are you guess handling static values that you use in bursted jobs? do you move them to static readonly fields? does burst always inline this stuff or should i just keep them as "magic numbers" to make sure they are as close as possible to the code were they are used (to prevent cache misses)?

[rustic rain]

you mean DynamicBuffer<Child>?

[rotund token]

no

https://docs.unity3d.com/Packages/com.unity.entities@0.17/api/Unity.Entities.LinkedEntityGroup.html

this i think

[rustic rain]

so that still means I need to go through and destroy all those entities recursively myself?

[rustic rain]

    private static void DestroyEntityRecursively(EntityCommandBuffer.ParallelWriter buffer, int entityInQueryIndex, BufferFromEntity<Child> data, Entity mainEntity)
    {
        if (data.HasComponent(mainEntity))
        {
            var entityChildren = data[mainEntity];
            for (int i = 0; i < entityChildren.Length; i++)
            {
                DestroyEntityRecursively(buffer, entityInQueryIndex, data, entityChildren[i].Value);
            }
        }
        buffer.DestroyEntity(entityInQueryIndex, mainEntity);
    }

yep yep yep

that works

Kind of annoying tho

that you have to create such method yourself

[rotund token]

that's what i'm trying to tell you @rustic rain , you don't have to. You can use LinkedEntityGroup to do it for you

by default unity will setup LinkedEntityGroup for you if you do conversion

and if you destroy the parent all the children will be destroyed

there's also EntityManager.SetEnabled(entity, enabled) methods etc that will enable/disable an entity but also all entities in the LinkedEntityGroup

[rustic rain]

I'm doing it inside job

[rotund token]

call ECB.Destroy(entity)

[rustic rain]

        Entities.WithAll<Interactable>().WithReadOnly(children).ForEach((Entity entity, int entityInQueryIndex, in Translation translation) =>
        {
            if (translation.Value.x < -60f)
            {
                DestroyEntityRecursively(buffer, entityInQueryIndex, children, entity);
            }
        }).ScheduleParallel();

[rotund token]

and it will destroy everything

[rustic rain]

nah, it doesn't

buffer.DestroyEntity(entityInQueryIndex, mainEntity);

[rotund token]

did you setup a LinkedEntityGroup?

[rustic rain]

that's what I did in the first place

[rotund token]

if you go into EntityComponentStore.DestroyEntities you can see it in action

            ref UnsafeList entitiesList, ref int minBufferLength, ref int maxBufferLength)
        {
            int indexInArchetype = ChunkDataUtility.GetIndexInTypeArray(chunk->Archetype, m_LinkedGroupType);
            if (indexInArchetype != -1)
            {
                var baseHeader = ChunkDataUtility.GetComponentDataWithTypeRO(chunk, indexInChunk, m_LinkedGroupType);
                var stride = chunk->Archetype->SizeOfs[indexInArchetype];
                for (int i = 0; i != batchCount; i++)
                {
                    var header = (BufferHeader*)(baseHeader + stride * i);

                    var entityGroupCount = header->Length - 1;
                    if (entityGroupCount <= 0)
                        continue;

                    var entityGroupArray = (Entity*)BufferHeader.GetElementPointer(header) + 1;

                    if (entitiesList.Capacity == 0)
                        entitiesList.SetCapacity<Entity>(inputDestroyCount * entityGroupCount /*, Allocator.TempJob*/);
                    entitiesList.AddRange<Entity>(entityGroupArray, entityGroupCount /*, Allocator.TempJob*/);

                    minBufferLength = math.min(minBufferLength, entityGroupCount);
                    maxBufferLength = math.max(maxBufferLength, entityGroupCount);
                }
            }
        }```

[rustic rain]

eh

I think I'll go with my way

Child component is automatically added

no need to setup anything

[devout prairie]

If i understand you correctly.. basically any values you pass into a job are ( as i understand it ) copied into the job so to speak.. ie you don't want the job to have to reach out and grab some value from elsewhere in order to use it, so it works with a local copy of it.. so for example in a system OnUpdate, if you are passing a static value or one of the systems field values into a ForEach, you have to create a local copy of that value and pass it into the ForEach.. i think it's the same idea with custom jobs

[pastel field]

For those who like me struggle with this, there is an example in the samples that does exactly that.
https://github.com/Unity-Technologies/EntityComponentSystemSamples/tree/master/ECSSamples/Assets/Advanced/BlobAssetSimple

I come to ECS only because I'm motivated by Havok and it's a very bad entry point 🥶

[rustic rain]

not really, jobs do work with references too

[pastel field]

It doesn't seem to take into account the Wrap mode...

[rustic rain]

native arrays for example

or component data structs

[worn valley]

I would love to know too because copying data is expensive. I wonder if the ReadOnly attribute does something fancy behind the scene like only passing the pointer so you don't have to copy all that info.

[rustic rain]

relax guys, it's not actual copying kek

from what I'v seen ReadOnly still passes data as reference, just forces you out from modifications during compilation

[tepid abyss]

Is the packages in my project not debugable (specifically, the Hybrid Renderer package)? I am trying to set a breakpoint at some method but it doesn't seem to break

[rustic rain]

make sure Burst is disabled

[tepid abyss]

Ok, I'll try that

Doesn't seem to make a difference

Umm, I get a "Didn't find loaded method for..." error

[candid epoch]

this is true but afaik only for "non static readonly" values.
i hope this example system illustrates what i mean:

  public class TestSystem : SystemBase
  {
    private float3 notStaticReadonlyFloat;
    private static readonly float3 staticReadonlyFloat;

    protected override void OnUpdate()
    {
      Entities.ForEach
      (
        (in LocalToWorld localToWorld) =>
        {
          // this works without binding to a local variable
          var otherNewValue = new float3(1) + staticReadonlyFloat;
          
          // this doesn't not work without binding to a local variable
          var newValue = new float3(1) + notStaticReadonlyFloat;
        }
      ).Schedule();
    }
  }

my question is if burst inlines the "staticReadonlyFloat" field to ensure there will be less cache misses. i've seen a data oriented programming talk where they talked about the issues that some compilers "lump" static data together, which can result in cache misses and false sharing. was wondering if this is an issue for burst as well.

[rustic rain]

you must copy your static fields to local

burst doesn't work with static fields

[devout prairie]

Yeah i think i've possibly come at this question from the wrong angle. Definitely you don't want to be accessing managed memory inside jobs and as i say ForEach requires a local scope level variable to be passed in ( can't pass in a system field for example, it has to be assigned to a local var ).. Tbh i'd like to know more of the details under the hood of this also 🙂

[rustic rain]

that's how it's intended

from what I'v seen, unity did magic with pointers and unsafe operations

[candid epoch]

the example above does work tough (the static readonly part)

[rustic rain]

I doubt you'd be passed a local copy

in every job

that would cause GC to just kill itself

and all perfomance would be wasted

[devout prairie]

ah yeah i see, that's interesting

[rustic rain]

hmm, maybe compiler sees it as const in the end?

after all static readonly fields are assigned before .NET runs at all

[candid epoch]

i mean this is not actually an issue of "it does not work". just wondering if i get a performance hit from moving this to a static readonly field. as i understand this highly depends on how burst compiles this. if it moves it somewhere far away from the orgininal code, this could be a performance hit, if it just inlines the static readonly field it wouln't make a difference

[rustic rain]

uuugh, I wouldn't worry about it too much

[devout prairie]

i think where and how it's stored when the job retrieves it is really the issue.. so for example if you have a value in managed, it's going to take longer for the job to pull it out and use it.. i do wonder if the dots system basically takes all native statics when it initially runs and puts them together into fast pools of native memory, which can then be accessed quickly by jobs alongside the other data they are grabbing from the native memory space..

would be an interesting one to test though, run a million jobs using a large static value and compare with a non static version

[devout prairie]

inlining the static readonly would mean it's actually copying the value right

[candid epoch]

basically just replace the variable with the value that get's assigned

if it inlines a static readonly struct field it would recreated the struct every time the code runs

[devout prairie]

mm

there's a couple of people on here that could probs answer this properly i think

@rotund token @robust scaffold sorry for the ping 😛

[candid epoch]

i think inlining would probably be better for stuff that needs to run fast because there is no need to lookup the struct value (which would probably be a cache miss) from somewhere else, but that's just an educated guess

[devout prairie]

I think KornFlaks could probs answer this as he does a lot of custom vectorization and looking at the compiled jobs etc

Maybe out of timezone atm though, not sure

[candid epoch]

no hurry was just wondering how it works 🙂

[devout prairie]

Yeah it's interesting, kinda prompts me to think more about how it's working under the hood, it's easy to just barge on making a bunch of assumptions about what happens underneath

Definitely worth doing a little test tho, million jobs doing it in two different ways

[robust scaffold]

Ugh, what's going on?

[devout prairie]

Yeahh sorry you'd have to read the last few comments 😛

[robust scaffold]

https://docs.unity3d.com/Packages/com.unity.burst@1.7/manual/docs/CSharpLanguageSupport_Lang.html#static-readonly-fields-and-static-constructors

Creates, I assume, a native array copy of the managed array at compile time that is passed between main thread and job thread when needed.

Only one creation occurs. And then that "proper" native array is used repeatedly until that system falls out of scope and is destroyed, if it ever is.

When compile-time evaluation fails Burst will try to fall back to compiling all static initialization code into an initialization function to be called once at runtime. This means that if code fails compile time evaluation it needs to be Burst compatible, otherwise compilation will fail

Burst compatible which means blittable values. Which means native arrays backing. I dont have much experience with static readonly fields on practice though so I cant tell you how itll look in the Burst inspector.

[devout prairie]

Interesting i've never read the burst docs! So i was kinda sorta thinking in the right direction..

I knew you'd have a better understanding of this 🙂

@candid epoch

[candid epoch]

thanks for the link. very interesting stuff. will dig into it a little further on the weekend. so it will definitely evaluate it once at compile time. only thing i'll need to figure out is if burst compiles this as a "pass by reference" or "pass by value".

[robust scaffold]

The nonStaticReadonlyFloat is passed by value into the ForEach. The staticReadonlyFloat is passed by reference.

[candid epoch]

and burst puts the value of the staticReadonlyFloat in memory somewhere close to the jobs data to minimize cache misses?

[worn valley]

How does it pass native arrays in? Does it pass the pointer into the burst job then?

[candid epoch]

my guess would be yes. you can treat a pointer as the start of an array and if you know it's length and the datatype you can easily calculate the end pointer and the step size.

[devout prairie]

I would guess as it's a pointer to natively allocated unmanaged memory then yeah

[rustic rain]

but native containers are managed type

[deft stump]

they're not managed by C#'s GC

[broken siren]

Hey guys a quick question please, I just want to know if I can renew my Unity Personal License (which is free by default) after it expires. Or do I have to purchase the next license ?

[robust scaffold]

Wrong channel to ask that but yea. Personal License are permanently free. You will not need to pay for the next personal license. Just reactivate it.

[robust scaffold]

I do not know the exact location of a burst compatible static readonly variable. Give me a min to check the inspector

[robust scaffold]

private static readonly byte[] Zeroed = {(byte) 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10};

So it appears that for a compile time readonly array, what Burst does is just directly copy the value of that array into the bursted function and then sees if it can optimize the method. In this case, it recognized that it is setting sourcePtr values with Zeroed and just mem-set the destination array with the constant Zeroed array.

[robust scaffold]

In the case of a static constructor, it once again recognizes that it is a constant byte array and optimizes the values to a mem-set

[robust scaffold]

And yea, static constructors can not accept parameters. So there is no dynamic setting of static readonly variables.

So basically it's like writing a constant value. There is no issue on whether or not the static readonly variable is located inside or outside the cache because the constant value will be optimized away (by embedding the value into the code itself) at compile time.

So final answer, yes a static readonly value is located "inside" the cache.

[devout prairie]

Interesting stuff Kornflaks, so A) it is copied, and B) it is basically what he was asking, kinda inlined into the job

[robust scaffold]

Yea. I guess it is, in the end, "copied by value" and not "by reference"

[candid epoch]

@robust scaffold thanks for looking it up. good to know that burst pretty much exactly optimizes it the way i would expect it to after watching the data oriented programming talk. And even better that i don't have to be concerned about a performance hit when moving static values to static readonly fields.

[cunning perch]

What's the most recent tutorial or guide for setuping up a DOTs project?

[karmic basin]

Check pinned messages

Yes, even 1 year old resources still applies :)

We are expecting a new release this quarter

Maybe it will break everything, maybe it will come with updated docs... we dont know

[cunning perch]

0.5 right? I figure I’d just learn the current version. The knowledge should mostly translate 🙏

[karmic basin]

0.50 yeah

[cunning perch]

Transfer I mean

[karmic basin]

Yes I would do the same

There's a forum post were they answered a little bit about what to expect

https://forum.unity.com/threads/dots-development-status-and-next-milestones-december-2021.1209727/

[lusty otter]

Any way to use Time.realTimeSinceStartup or other timing mechanism in a Bursted job?

I have a very expensive loop that I'd like to exit out early if it exceeds for example 10 ms, so to not kill FPS.

[deft stump]

put it in a variable and pass it in

[lusty otter]

I mean I need to access the time during the execution of the job, not before it starts.

[coarse turtle]

might be possible through a native dll and exposing a static api to do so 🤔

[lusty otter]

😅

Sounds a bit complicated for such a seemingly trivial need.

[coarse turtle]

wonder if c# datetime works in a job 🤔

[lusty otter]

Burst error BC1045: Struct `System.DateTime` with auto layout is not supported

[devout prairie]

in Burst, you measure time by the distance between atoms 😮

[lusty otter]

I feel like I can't be the only one that have tried to do something like this

Time slicing a long algorithm into multiple frames, except in a Bursted job.

[devout prairie]

yeah i wonder, you'd think there'd be some way

[lusty otter]

I suppose I can fall back to collecting some performance information on low end mobile phones, and see how many loops they can run on average, then split it at that.

It will mean for high end phones it will take longer to complete (or I guess I could adjust the loop depending on the speed of first execution)

But that's just so unnecessarily complicated, when I feel like this should've been a trivial task.

[devout prairie]

i wonder if there's some hack, for example have another job or code running on main thread which writes to a timer value, and the bursted job kinda unsafely reads from that or something

[lusty otter]

That wouldn't work on single threaded devices, no?

[devout prairie]

mm doesn't seem like it'd apply there no

hmm how about have a shared component which is given the time each frame, and also pass that into the job

[pliant pike]

why do you need to access the time during a job 😕

only way to do that is WithoutBurst.().Run()

[lusty otter]

I want to run an algorithm on a huge amount of items, each takes variable and undetermined amount of time
I just want to terminate the execution after 10 ms for example and continue in the next frame, so it won't tank the frame rate.

[pliant pike]

in a proper bursted job time doesn't exist, its completely isolated from the outside data

you'd have to exit early in some manner or run the job for a certain amount of update cycles from the outside

[lusty otter]

Well I don't think it's too crazy to ask for obtaining time inside a Bursted job

[pliant pike]

it kind of is crazy though

[lusty otter]

I mean I could just call a native API like others have suggested.

[pliant pike]

jobs are designed as completely isolated systems that run on the data you give them as efficiently as possible

you can't do that if they have to wait a ton of time to get data from ram

[lusty otter]

I don't really care about the design, for example if you give me a huge file to hash it while keeping the game loop alive, I'll need that.

It's not a ridiculous use case that I'm asking for.

[devout prairie]

ah i can see the problem i think now that i think about it.. so you're essentially locked into a bursted job in the main thread in a single threaded environment, burst obviously does not want to reach out into memory to grab an updated time value as this breaks the whole burst execution

maybe a solution would be to parse files externally or run a job at some other point which analyses the 'source file' and determines a good scheduling routine / chunk count etc

[pliant pike]

yeah that's why you have to set up the data in specific ways, make sure its local to the update and whether it's read or write etc, its also the same with schedule() and scheduleparralell()

[devout prairie]

yep

[pliant pike]

you could put early out code like I said, import an int iterator that counts the number of cycles, if you know roughly how long it takes to calculate one cycle then you can have exit when the iterator equals however long want

that would only work in a single threaded job though, you'd have to create multiple counters and pass them into a multithreaded job

[lusty otter]

The best way is to have a lowball estimate on how many cycles it can safely do in 10 ms even for bad devices, exit out the job and use the elapsed time to adjust how many cycles for next frame

Here's my situation:
I currently have a 2 second fade into black screen, and then do this expensive Bursted job here which can take some time (but it's hidden well because of the black screen), and then fade out into gameplay.
On low end devices the pause can still be felt, hence I'm thinking to move some of the workload across the 2 seconds of fade in frames to improve UX.

But like if this is going to be the cost then I'm just not going to bother, it's so much work for arguable amount of benefit when user can tolerate a bit of loading

I'm just thinking if this wasn't done in Burst, this would take a few lines of code at most.

[pliant pike]

yeah figuring out how to use burst and jobs demands a lot more time and knowledge

[lusty otter]

I mean this has nothing to do with Burst really, it's just straight up not supported use case.

[pliant pike]

yeah people have been asking to do things similar to what your asking since the beginning of dots

[devout prairie]

i would say yeah but it could also take anywhere beyond 100x longer to do without burst

[lusty otter]

That's fair.

The "transform your code so that it operates on isolated local data" part of Burst I would agree, but this is simply just "I can't call a function which is quite literally a native API on any operating system that has nothing to do with Unity at all"

[devout prairie]

I've saw people post things in here they said took 100 seconds and reduced to 3/4 seconds with burst.. so i guess it's kinda trying to find a way to still get all that benefit, without breaking the paradigm that gives the whole benefit

[lusty otter]

Don't get me wrong I love Burst and my game wouldn't be possible without it, but this is about the only time I couldn't get it to work and it's completely out of my control so it's a bit frustrating.

[pliant pike]

oh I know what you mean, it is really frustrating figuring out the limits and what you can and can't do but when you do figure out something that runs way more efficiently than you could do any other way it's great

like there's no way I could figure out how to build code that works multithreaded and as efficiently as it does here on any other engine or from scratch

[lusty otter]

I'm just going to hold it off til later

Also the "do some amount of work in Burst then measure outside to then adjust for the amount of work on next frame" solution relies on two assumptions: 1) you are able to split the work into equal units to adjust and 2) the overhead of starting a Bursted job is negligible.

I don't know about 2), but 1) seems a bit sketchy for my situation.

[worn valley]

Anyone have good articles on how NativeSlice<T> works. I don't understand what happens with it and I haven't been able to figure out a use case from the documentation

[rustic rain]

why exactly you can't just turn that value into variable?
double for example

I don't really see why you want to get that value inside bursted job, not when you just setup that job

[lusty otter]

Here is the actual use case.

[rustic rain]

it's all going to run in merely same ms anyway

hm

so basically you want some async operation

well, I'm not even sure if you can stop job execution at all

[lusty otter]

The job can stop itself

If it can obtain the current time, it can simply check how long has elapsed and stop the loop.

[rustic rain]

seems like a bad design in the first place tho

[lusty otter]

This is not some out of the world idea, it's a pretty common practice to split a huge chunk of work across multiple frames.

[rustic rain]

I get the thing, I'm just not sure if jobs are intended for that usage at all

it will affect framerate no matter what tho

even if you somehow manage to stop it after 10ms, it's 10ms passed already

I think you can do it by forcing job to complete yourself

but instead of using whatever method that simply awaits for it to finish

you'll just check on it in while loop

[lusty otter]

The loop is inside the job, not outside

[rustic rain]

once it reaches treshold you cancel it, assuming you can do it outside of job

you can't access anything of outside world from job tho

only locals

[lusty otter]

You could say that, but nothing is stopping me from directly calling an OS level native API to fetch the current time.

[rustic rain]

well, then you'll have to run job without burst and in main thread

otherwise it just won't compile

[pliant pike]

the unity api is stopping you from doing that though 😕

[lusty otter]

You could call native code, no?

[rustic rain]

you can't call nothing but whatever api burst supports

[lusty otter]

But regardless, what I'm trying to say is, this is not something that's fundamentally undoable, it's just a design choice.

[rustic rain]

burst has very limited api, that's why it's so fast

[lusty otter]

I could just go and modify the compiled assembly and add that in.

[rustic rain]

well, are you willing to dig that hard into compiler code?

meanwhle, are you certain that won't make burst worse for everything else?

[lusty otter]

Nope, because I would just "rewrite it in Rust"

[rustic rain]

kek

that's the other solution

but overall I think you're just trying to apply techniques from different programming approach to DOTS

which might be not smth actually intended

[lusty otter]

Burst isn't DOTS specific even though it's primarily used by DOTS.

[rustic rain]

hmm

[lusty otter]

Well, I suppose at least for me I use it to speed up portions of my code where it makes sense to

[rustic rain]

if your code is just really expensive to run by it's own

maybe you might try simply making function pointers of bursted jobs

bursted functions*

[lusty otter]

Most of the things Burst doesn't support are directly because of multithreading, getting the current time from OS isn't really a problem.

At least in my head it works like that, and I think it's a common enough thing that there's not really a good reason why it can't be done.

Alternatively, has anyone measured the overhead of starting a Bursted job with .Run()?

[rustic rain]

well, considering even exceptions aren't really supported in burst I wouldn't be too sure

.Run is just doing code here right now

[lusty otter]

If it's negligible, I could just pull the while loop outside and measure time elapsed on each loop.

[rustic rain]

it won't be threaded

it'll run on mainthread

if that's what you are looking for

[lusty otter]

Right, but there could still be overhead.

[rustic rain]

there's a lot, but burst eats that

and still manages to bring more profit

but

really depends on your code

not everything can be vectorized

and this is where burst shines

I did test some really simple code of doing some transform changes

with and without burst

so far

invoking function pointer is about 10 times more expensive than doing some really simple and cheap math

jobs will probably have even more overhead

but you profit by putting them on other threads, keeping main thread mostly for scheduling or limited stuff

here that comparison

    private static float3 NewPos(float3 curPos, float speed, float offset, float delta)
    {
        curPos.x -= speed * delta;
        if (offset * -1.5f > curPos.x)
        {
            curPos += offset * 3f;
        }
        return curPos;
    }

and here the code I compared

[worn valley]

Jobs are supposed to be put info in, get info out another time. Trying to modify them to cancel just seems like it's outside the scope. If you are trying to get information from the OS, what other things could you potentially grab from other places. Kind of defeats the purpose of creating a job.

[lusty otter]

It's not outside telling to job to cancel, job cancels itself once certain amount of time has passed.

[worn valley]

fetching any information from outside of the job's little bubble that can change is not really what the jobs system is designed to do. I am sure you can hack it to do that, but Unity probably wants to keep the use cases simple

You just shouldn't grab any information after you schedule the job. I think jobs are much more like functional programming than other types. Inputs -> black box -> outputs

[lusty otter]

Technically you could have the job run on a second thread and repeatedly check a flag which the main thread writes to when the time is up.

private void BenchInside() {
    var job = new LoopInsideJob();
    job.Run();

    var startTime = Time.realtimeSinceStartup;
    job.Run();
    var endTime = Time.realtimeSinceStartup;
    Debug.Log($"Inside: {endTime - startTime} seconds");
    // Inside: 5.722046E-06 seconds
}

[BurstCompile]
private struct LoopInsideJob : IJob {
    public void Execute() {
        for (var j = 0; j < 1000000; j++) {
            var v = 670617279;
            for (var i = 0; i < 1000000; i++) {
                v = v % 2 == 0 ? v / 2 : v * 3 + 1;
            }
        }
    }
}

private void BenchOutside() {
    var job = new LoopOutsideJob();
    job.Run();

    var startTime = Time.realtimeSinceStartup;
    for (var j = 0; j < 1000000; j++) {
        job.Run();
    }
    var endTime = Time.realtimeSinceStartup;
    Debug.Log($"Outside: {endTime - startTime} seconds");
    // Outside: 0.6792498 seconds
}

[BurstCompile]
private struct LoopOutsideJob : IJob {
    public void Execute() {
        var v = 670617279;
        for (var i = 0; i < 1000000; i++) {
            v = v % 2 == 0 ? v / 2 : v * 3 + 1;
        }
    }
}

So the overhead of a job is big enough that I can't pull the loop outside.

I'm up for suggestions but tbh I think I'm just going to give up at this point

I thought this problem would take a few lines at most but apparently not, I'll just have slightly worse UX instead.

[rustic rain]

if the only reason your job takes 10+ ms to run is just that because you too many individual items

that can be easily threaded

you might want to consider compute shader

gpu likes that kind of stuff

and it'll probably take WAY less time to compute

[worn valley]

Why don't you record how fast it takes the first time the user does the code on their device and then store that in their player prefs. It would only ever happen once for a player on any device. You could also collect that data and then update a list for future builds depending on phone type.

[rustic rain]

what is intended way for unity to setup dependencies between jobs inside your own project?

do you just expose Dependency or Job Handles

in public properties?

[lusty otter]

They have to be run in a single thread, and doesn't play well with vectorization.

[lusty otter]

Different games will have different "items" to run so it's not the same every time.

[rustic rain]

I'm not even sure if you'll profit from jobs or bust then

[lusty otter]

It does, by a huge amount.

Vectorization isn’t the only thing Burst offers, just less junk in general comparing to normal build.

IL2Cpp generates a lot of junk mainly due to metadata stuffs.

[rustic rain]

I see

well, I don't see any other solutions to you rather than either rework your items or accept job limitations
Either way compromises

[rustic rain]

I need help with this system I'm trying to develop

basically

I have interactable collisions

and what I want is to have some sort of tag that will contain info about what collision type I'll use

so in my collision resolver I simply put tag that this item wants to get his special collision event

and then my for each unique collision type system does unique collision stuff on all items

I'm not sure how to resolve this kind of stuff

since you can't keep Type in component

Any ideas how to that can be done?

[robust scaffold]

Does it have to be Type? Can you not just use an enum and map to collision type?

[rustic rain]

I don't want to be hardcoded to enum

since I might want to add A LOT of other interactions

and every time I'll have to adjust enum which is uuggh

and then I'll have to run huge switch on that enum

not good

[robust scaffold]

What is Type? Is there an integer component? Is that literally System.Type?

[rustic rain]

In perfect world I'd simply contain Type in my Collidable component which will be added during collision

wait a second

I think I got it

I'll have universal Tag

smth like

WantCollisionTag

WantCollisionEffectTag

yeah

and then I'll have query that will run only with that tag

[robust scaffold]

Uhh, to be honest. In my experience, ya dont want to add and remove components on a regular basis.

[rustic rain]

and each system will have it's own resolver

well

you can't really avoid that

that's core stuff

besides

my collisions are basically a goal of game

and they don't really happen a lot

it's only player that collides

and he does it like once per second

on average

[robust scaffold]

Alright then. I dont quite have the context here to suggest anything so good luck.

[rustic rain]

Is there a fast way to simply destroy every single component type on all entities?

basically I have components that only exist for 1 frame and I need to flush them all at once

    protected override void OnUpdate()
    {
        EntityManager.RemoveComponent<CollisionHappenedTag>();
    }

I need smth like this xD

[gilded glacier]

from what I know (although not sure whether correctly), singleton entities are internally handled the same as other entities, that is they have their own chunk

I think

[rustic rain]

no no no

I mea

I can have a lot of entities

with that tag

I want to remove them all

[gilded glacier]

no, I'm just talking my issue here

[rustic rain]

I want that tag to cease to exist in current world

ah

[gilded glacier]

nothing with yours

if singletons are actually saved in their own chunks of their type, isn't there a big waste of memory space by having a single entities in a chunk ?

and is the use of singleton entities any beneficial (other than the require for update with systems) ?

[rustic rain]

it somewhat is, I remember some custom dots framework adressed it

but I'm not sure how whether it's worth to look into at current state

since it all might change in 0.5

[rustic rain]

it's a component

so it can still run through queries

bursted

and paralleled

unlike static fields

which will be forced into your main thread

unless ofc you make Job.WithCode

[rustic rain]

when you schedule some job with Schedule() will that job run on worker threads or?

I'm guessing it will run on first possible worker thread

but not splitted into itself through different threads

I'm basically trying to understand whether ScheduleParallel makes any sense in super small queries (like 1-2 entities)

[rotund token]

depends more on your workload than your entity count

you might only have 2 entities, but if these entities are your AI directors for each team and have to do a bunch of really costly calculations then you probably want them on different threads

but if your workload is small, then yeah definitely makes sense to just use Schedule

also something people often overlook is that, just because their job isn't running in parallel doesn't mean you're wasting cpu time - other jobs can run at the same time as your job occupying other threads

[olive kite]

Is there any reason why the scale of an entity would reset to 1,1,1 after modifying it and then removing an entity in it? There is an animation I am doing and it does it fine with scaling the entity, but then when it is completed and I remove a flag I'm using to say that it has an animation running it resets the scale back to the original

[radiant berry]

What tutorial is good to make my first simple game with dots like pong for example that is up to date with the 2020 editor version?

[robust scaffold]

https://dots-tutorial.moetsi.com/

Not the best "newbie" tutorial but it is fully up to date with DOTS 0.17 lineup

[worn valley]

That is a good one. Goes through everything

[rustic rain]

        JobHandle = Entities.WithAll<PlayerTag>()
           .ForEach((ref PhysicsColliderBlob collider, in Translation translation, in Rotation rotation) =>
           {
               var colInput = new OverlapColliderInput
               {
                   Collider = collider.Collider,
                   Transform = new PhysicsTransform(translation.Value, rotation.Value),
                   Filter = collider.Collider.Value.Filter
               };
               if (physicsWorld.OverlapCollider(colInput, out OverlapColliderHit hit))
               {
                   beginBuffer.AddComponent<CollisionHappenedTag>(hit.Entity);
                   buffer.RemoveComponent<PhysicsColliderBlob>(hit.Entity);
               }
           }).Schedule(Dependency);

smth is really wrong with this
I can't find what's the problem

so far

It triggers twice

in 2 different frames

while it shouldn't

since I remove collider component

and it works good on everything but one type of entity

basically, when it triggers

It's supposed to add tag and remove collider

that happens

BUT, for some reason that tag gets added to some absolutely irrelevant entity

0:1

index

and that ends up breaking game obviously

I suspect that's because it triggered twice

and there could be some bug that adds CollisionHappenedTag to some random entity

I checked with DEBUG which entity is hit.Entity

it's always the correct one

never entity 0:1

[rotund token]

are you running this in fixed update?

[rustic rain]

I don't have fixed update at all

[rotund token]

alternatively, when is the command buffer relative to this update

[rustic rain]

cause 2D physics is outside of it

everything is running per frame

ok, I think I know what's wrong

yeah

my pepega brain forgot to mention WithAll

in job