archived-dots

[rustic rain]

so basically

example
type: "Asteroid"
name: "Default"

and then I assign blob reference to that entity

aaaand after I load subscene
I have prefab with blob asset reference that points to that "Asteroid" or name "Default"

so if I instantiate it, I always know what prefab it was instantiated from

now if I want to add that prefab to some database

I CreateAdditionalEntity()

and assign Entity reference to it

reference to that main prefab

and then my runtime system loops over those additional entities and adds them to dynamic database

[devout prairie]

well what i'm doing is fairly simple, but it's just giving me problems at every turn

[rustic rain]

no blobs needed

[devout prairie]

i could find a way to not use blobs but there's really nothing wrong with that approach, and when i'm not using subscenes, it works perfectly

in fact the problem isn't even related to the blobs themselves..

[rustic rain]

there's problem with blob database

it's static

you can't make it bigger

or smaller

[devout prairie]

GetPrimaryEntity for the prefab is returning the wrong Entity

[rustic rain]

did you reference your prefab?

[devout prairie]

i'm not making it bigger or smaller i'm building a blob, as i say that's not even the issue

[viral sonnet]

at conversion stage?

because that's normal sadly

what i mean it's different to the runtime entity

[devout prairie]

yeah it references and creates the prefab, it just doesn't return the correct primary entity for this particular prefab, whereas for the other prefab it does return the correct entity

[rustic rain]

Entities are remapped only if you assign them to Components

so if you assign them to blob

they won't be remapped

[devout prairie]

hmm let me think this one out

so why is one working and not the other

[rustic rain]

because if you convert in runtime

you create entities in actual World

subscene conversion is done inside it's own conversion worlds

at least that's what I assume

could be wrong

[devout prairie]

i wish i could demonstrate without flooding the chat will all of the code

[viral sonnet]

post the file :)

i mostly create blobs at runtime start.

pretty easy and you just have to put them in your own blobassetstore which needs to be disposed

[devout prairie]

well i could but that then introduces problems with the order because the subscene is loaded after systems OnCreate

so i'd need to create a system which looks for subscene to be loaded, and then builds the blob in OnUpdate

and then tell my other systems to wait until this is done

[rustic rain]

why OnUpdate

[devout prairie]

or, manually create/add my systems only once subscene data is loaded

[rustic rain]

OnStartRunning works too

[devout prairie]

hmm

[viral sonnet]

i'll ask something else. why blobs?

[rustic rain]

but I feel like you overcomplicate things xD

[devout prairie]

is that guarenteed to be after the subscene is loaded though?

[rustic rain]

if you define RequireForUpdate

it is

[devout prairie]

kinda going in circles tbh

i think i'll just not use subscene, and avoid all of these problems

😛

[rustic rain]

rip perfomance

I mean

loading time

[devout prairie]

i'm going to give it one more shot

but i'm not happy about it 😛

so, GetPrimaryEntity is clearly not working as expected inside my subscene

the thing is, the exact same code works perfectly outside of subscene, this is the annoying part

i literally take my authoring object, out of the subscene, and my blob is built perfectly and as expected

[rustic rain]

sir, it's because you wrap it in a type

that doesn't get remapped

it's as if you grabbed index and version from Entity

and expected them to change after conversion

Entity references are remapped once subscene is loaded

[safe lintel]

shouldnt be storing entities inside of blobs

[devout prairie]

why not?

[devout prairie]

i get what you're saying here, and it makes sense, but it does work outside of the subscene, during conversion

[rustic rain]

because when you inject game object

you are converting during runtime

[devout prairie]

i'm literally just storing a struct containing an Entity field inside a blob

[rustic rain]

welp

you are literally just storing int2

in a blob

and you expect it to be valid entity reference xD

[devout prairie]

exactly

no the point is that it works fine, during conversion, when not in a subscene

[rustic rain]

because during conversion

you create blob

in actual world

so the entity reference you put in there

is valid for this world

but subscene world only exists during conversion

that's why entities must be remapped

and it's done on components

[devout prairie]

there's no issue with storing the GetPrimaryEntity to the blob struct during conversion and then it failing to remap correctly, if it's not in a subscene

[rustic rain]

you can just store all your entity references in component

that creates blob in runtime

because this way

this component will be remapped with valid references

[safe lintel]

as issue said entities can be remapped but when stored inside of blobs they wont be

[devout prairie]

yeah i get the concept, i just don't get why it does work okay when not in a subscene

[rustic rain]

because in runtime

you create in actual your world

not in temporary world

[devout prairie]

i'm guessing it could be what you're hinting at above

yeah that

that makes some kind of sense

[rustic rain]

think of them as relative and absolute references

subscene references are relative to subscene

runtime conversion references are absolute

[devout prairie]

the prefabs once converted do still exist within the subscene

[rustic rain]

so when you load subscene and read references as absolute you are certainly getting an invalid result

[devout prairie]

so i guess by my logic, that relative reference inside the subscene's Convert would resolve in exactly the same way as when it's not in a subscene

[rustic rain]

if you serialize it

[devout prairie]

but i guess in the case where, a subscene creates it's own conversion world it possibly makes some kind of sense

tbh, i'd still expect it to work as is

[rustic rain]

how can you expect pointers created on one machine

work on other?

[devout prairie]

conversion that is not in a subscene still uses it's own conversion world, this is why i don't get there's a difference

[rustic rain]

can you actually check?

read dstWorld

in your runtime conversion

and compare

to your actual game world

is it same reference?

[devout prairie]

right so i'm guessing it isn't

is, sorry

but in subscene, dstWorld is not the same

[rustic rain]

yep

thus references are invalid

when you wrap them

in blob

store them in component instead

and they will be valid

because they will be remapped during subscene loading

[devout prairie]

haha okay i think i get it

damn

[rustic rain]

man, I just realised

[devout prairie]

still infuriating though

[rustic rain]

that Unity has entity remapper

and they didn't expose it

sadkek

would make creation of save/load so much easier

[safe lintel]

why the desire to store in blobs anyway? I sort of get it when I want to do something one way and it doesnt work, its sometimes hard to let go

[rustic rain]

I don't get it either

in fact

you could probably just use chunks as your database

and use temporary queries

to get all lists of prefabs

[devout prairie]

well why not use blobs?

[rustic rain]

hehe

[devout prairie]

they are specifically for static data

[safe lintel]

well the whole no remapping thing 😅

[rustic rain]

and your prefabs aren't

[safe lintel]

yeah but entities themselves arent necessarily static

[devout prairie]

well the idea is, once the prefabs are created by conversion, they are static at least until they are destroyed

[rustic rain]

so

why can't you just do

[devout prairie]

so all i'm doing is storing my unit data ( health etc etc ) along with the prefab entity

[rustic rain]

query.GetNativeArray()

if it's static

you will have persistent array

of any valid component you want

health, entity, etc

[devout prairie]

again, why not use blobs

it's what they're for

[rustic rain]

I wouldn't agree

[devout prairie]

say i have 10 types of unit right..

each one is basically similar, but have different health, etc etc

i could create an entity/component setup just to store all of those templates yeah

or, i could just store it in a blob

i don't get the problem

it's literally just a struct of all of the unit data, that i can dip into when i'm spawning a unit

[rustic rain]

well, the obvious problem is that you create blob at the wrong time xD

[devout prairie]

spawn system says 'ok i need to spawn x unit', it then dips into the blob and pulls the data for x unit, gets the prefab entity, gets the health, etc, and spawns it

[rustic rain]

why store health differently?

shouldn't be it's on prefab already?

[devout prairie]

there's not a separate prefab for each unit

there's lets say a base model prefab, which applies to all units

[stone osprey]

Btw does anyone here knows any good quadtree library ?

All i find are pretty outdated and not "high performance"... Wanted to use one in tie with dots basically

[devout prairie]

and then there's different attributes for each unit type

[rustic rain]

ah

I plan to have spawn pipeline for this kind of thing

where I could have simple

Spawn method

with prefab name

and settings struct

in which I define my desires hehe

[devout prairie]

it's really not difficult, complicated or unusual

it's an analogue of using ScriptableObjects in mono

but as you said, i'm basically creating the blob at the wrong time, in the case of subscenes, which was my whole frustration

subscenes seem to be buggy, they don't update until you've coaxed them in multiple ways to do so, they seem determined to cause problems, and now the blob problem comes along 😛

i don't often get annoyed, but when i do 😛

[rustic rain]

well

[devout prairie]

the blob thing was of course my problem, not a problem with subscenes themselves

[rustic rain]

subscenes is future

so better learn how to work with them

[devout prairie]

but after several attempts and constantly being tripped up by subscenes, i was a tiny bit annoyed

[rustic rain]

runtime conversion is slow

[devout prairie]

it is

i mean i get it, i understand the benefits of using subscenes etc

that's literally why i decided ok i have everything working, i'm going to convert it to a subscene

the way i did this previously was probably more in line with what you're suggesting, and i did use a subscene for the prefabs

so i had an authoring component inside the subscene, which had a list of all of my GameObject prefabs

i then created a single prefab holder entity, and added a dynamic buffer, with data on my prefabs etc

i'm pretty certain the buffer also contained references to the prefab entities

[rustic rain]

buffers get remapped

[devout prairie]

so i think the problem in this case is exactly as you're saying, inside the blob they're not getting remapped

whereas inside a component or buffer yeah they do

[rustic rain]

ComponentType has a field HasEntityReference

pretty sure for this exact reason, hehe

[viral sonnet]

i'll ask again. why blobs! what do you gain?

[devout prairie]

i think for simplicity, i'll just switch back to that, but inside the buffer i'll hold a reference to the prefab entity alongside a blobassetreference which will hold the actual data for each unit type

[narrow field]

Trying to install the Jobs package by itself but get this error immediately.

I'm using unity 2020.3.12f1

[rustic rain]

wrong version

at least

[devout prairie]

i just figured it's what they're for.. basically to convert my scriptable objects ( which can be created/edited in the editor ) to a storage that i can dip into from my spawner system

[viral sonnet]

unit data makes total sense, i just don't get the need for an entity in there

and I'm pretty certain you could split that

[devout prairie]

simply just to hold the prefab entity that the spawner will spawn

again, it's just when i'm creating the blob, because the entity isn't remapping

[narrow field]

What version do I need?

[devout prairie]

the idea of storing the entity in a blob in itself i think is fine

alongside the rest of the unit data

[viral sonnet]

i don't agree with that.

a spawner can consist of a prefab + blobAssetReference

it doesn't have to be all in the blob

[devout prairie]

of course it doesn't have to be

[narrow field]

anyone?

[devout prairie]

but why split that up

[viral sonnet]

because a blob is not designed for entity remapping 🙂

or any kind of dynamic data

[narrow field]

Error happens upon trying to install any of these previous versions too

[viral sonnet]

(unless you build at runtime)

use 2020.3.30+

[devout prairie]

well exactly, i was building at runtime, which was fine, it worked perfectly

[viral sonnet]

i'm confused, i thought you had the problems at conversion stage?

[devout prairie]

yes basically, it worked perfectly when building at runtime

but once in a subscene, not

[narrow field]

guess not.

[devout prairie]

because the subscene wasn't remapping to the runtime entities i guess

[devout prairie]

i think this:
#archived-dots message

[safe lintel]

@narrow field can you paste the contents of your package manifest json somewhere?

[viral sonnet]

use 2020.3.30+

[robust scaffold]

Either upgrade your unity version or use 0.11 Jobs.

On the topic of blogs, I haven't found a reason to use any personally. If I could encode textures, maybe, but currently I just have all may data in components.

It helps that in 2d, all the meshs are very simple

[narrow field]

Got it installed now, however it seems I cannot use Unity.Mathematics? It's appears in the package manager as installed however I cannot use it in any of my scripts

https://docs.unity3d.com/Packages/com.unity.mathematics@1.2/manual/index.html trying to follow the documentation that has an example that has "using static Unity.Mathematics.math;" but that STILL doesn't work.

[rustic rain]

why are you using static?

just do using

and leave .math

[deft stump]

using static is for later versions of unity I believe

[narrow field]

Doesn't change anything

I'm literally just following what the official documentation on my current version is doing

[rustic rain]

do you have Assembly Def?

[narrow field]

no.

no mathematics or math in the Unity namespace

[rustic rain]

I mean

if you have defined an assembly def

you need to reference it manually

[robust scaffold]

If you have one of these defined, you need to add Unity.Mathematics to the Reference list.

I've gotta say, this is one of the most beautiful burst outputs I have ever seen.

[rustic rain]

how does it look in c#?

[robust scaffold]

It's the one line at the start of that loop.

[dreamy glade]

In an ECS for each loop, how would I go about getting entities WITHOUT a certain tag? Thanks.

[robust scaffold]

.WithNone<Tag>()

[dreamy glade]

Thanks Korn!

[robust scaffold]

Ive gotta hand it to the 3 guys working on dots physics over at unity, combining 4 AABBs into a single node is ingenious.

The issue is that from what I can tell, everyone working on DOTS physics were either reassigned or fired. The last response from a DOTS physics programmer on the forums was during the 0.17 days and 0.50 brought no new changes to the API.

[balmy thistle]

I don't think you're correct

[robust scaffold]

I'm just saying, during the ~1 year between 0.17 and 0.50, nothing was changed in physics beyond the bare minimum to continue functioning.

I think it got "deprioritized" like DOTS animation and audio.

[viral sonnet]

a shame, physics could need some love. but due to the big code changes further extensive development was halted

dots physics is one of their best releases without much patching needed

[narrow field]

Trying to fix a bottleneck in my system. I'm trying to set a bunch of points in a nativearray and then setdata on a computebuffer using the nativearray. Issue is that I'm using 1 million points and currently most of the CPU frame time is going to SetData on the compute buffer

[robust scaffold]

If each point is a float3, that's 12 MB of data being transferred to the GPU. I dont think you're gonna get any faster unless you reinvent electronic engineering.

[narrow field]

I'm using half3s

[robust scaffold]

Then that's 6 MB, not much better.

You need to get really creative in packing bits or space it out over multiple frames. 1M points is a lot of data being transferred.

Modern video games have thousands of MB of textures in the GPU because they only transfer the data once at loading, or stream it in over multiple frames. If you're trying to transfer 1M points every frame as update, that's physically impossible.

I recommend transferring in the source data at startup to the GPU and computing whatever you need in a compute buffer. Dont try computing things in a Job and then transferring it over.

[umbral veldt]

You can try using some time in the cpu end to tightly pack the data and send it to the gpu, and then also take some time on that end to unpack them. You'd be transfering less bits, but I don't know if you'd manage to compress it that much, and then, compression also adds some overhead. But you can check some articles about demo scenes, those guys really know how to compress things. For example, check Inigo Quillez's website.

[devout prairie]

the guy with the username something 'havok', was he one of the unity devs? i was never quite sure when reading his replies if he was a havok guy helping them with that or an actual unity dev

[robust scaffold]

Havoc is an intel physics system. That dev was loaned from Intel to integrate the havoc API into dots but has since disappeared from the face of the forums since the release of 0.17.

[devout prairie]

yeah that was what i kindof assumed

[narrow field]

Also, I'm not updating the entire nativearray per frame, but I'm having trouble figuring out how to only set a certain amount of data in the computebuffer

I tried using BeginWrite/EndWrite but it ended up not working and I wasn't sure why (yes I had it set to ComputeBufferMode.SubUpdates)

[robust scaffold]

Subupdate, in my experience, seems very unreliable.

Otherwise, set data seems pretty self explanatory

@viral sonnet How does one go and find where a component is being changed for debugging change filters?

[rustic rain]

wdym?

[robust scaffold]

I have a change filter on a job (the chunk.HasChanged(Comp, Version)) and it's being activated despite nothing moving. I need to find out why.

[rustic rain]

is smth acessing component data of this chunk?

[robust scaffold]

Apparently yes. And I need to find out what is.

[rustic rain]

just search for

GetComponentTypeHandle();

with type of what you think is the reason

[viral sonnet]

oh boy, only a good IDE will help here. comb the code for access with writes

[robust scaffold]

oh fun

[viral sonnet]

what also helps, some job that will debug.log when the comp has been changed and then disabling systems one by one. when the line stops outputting you're nearly there

[robust scaffold]

Well, I dont think I need this change filter anyways. There's gonna be one BVH for the entire world so if anything moves, it gets rebuilt.

[viral sonnet]

riggt, the unity one. yeah no change filter needed

[robust scaffold]

It's not that expensive, only 1.5ms for 10,000 entities. Scales linearly though so it's pretty bad at 50,000 since I use a better splitting algo than unity. If I were to use the same split as Unity's I get similar 0.75ms performance for 10,000.

Yea, there's refitting and incremental reconstruction of the BVH tree but.... is it really needed?

[viral sonnet]

do you need that much or just optimizing for optimizations sake

[robust scaffold]

Stress testing. I dont know if I'm gonna make every bullet a circle body or just a raycast line segment.

[viral sonnet]

bullets shoukd be raycasts

[robust scaffold]

They might be slow bullets though

[viral sonnet]

i think even in that case raycasts are more performant

[robust scaffold]

Probably, hrm. Now with a BVH, I can get back to actually programming physics.

[robust scaffold]

Yea. Raycasts is probably the way forward. I dont want bullets to be colliding with themselves and I havent implemented a collision filter yet.

[rotund token]

oh yeah saw this the other day relating to 2022.2

There are also some exciting changes to support newer Burst versions, e.g. the groundwork to switch Burst off dynamically when you set breakpoints with a managed debugger

[robust scaffold]

Imagine using a debugger. Debug.Log is enough for the honest down to earth coder.

At least assert works in burst.

I think burst is in a great position right now. I cant think of anything that it can improve on atm. I dont have that many methods so I havent hit the iteration speed wall let.

[rotund token]

my only request is always compiled burst jobs but break points switching to mono my remove my need for that

[robust scaffold]

A stretch goal maybe would be a hover over tooltip on the inspector over all the commands. A small blurb when hovering over an operation that says what it does, what the throughput is. The stuff that can be found on the Intel Intrinsics database so I dont need to keep going back and forth to see if any improvements actually improved anything.

[rotund token]

oh inspector could have a bunch of improvements

[viral sonnet]

yeah burst inspector is pretty bare bones

[robust scaffold]

And trimming out all the unity code as well

[viral sonnet]

amd uprof made it right. it shows the c# lines on top and in the bottom window the assembly. when you click on a c# line it highlights the corresponding assembly lines

exactly, there's soooo much bloat

[rotund token]

great to be able to see changes between compiles

so i can stop copy/pasting it into notepad++

[viral sonnet]

then i need a simple ctrl+f

[robust scaffold]

Yea, I saw those screenshots but I think short of coding their own IDE, the inline yellow comments in the inspector is the current best they can do.

[viral sonnet]

oh yeah! change highlights between compiles. mmhhh 😄

[robust scaffold]

I got a dead discord that I post my before and after screenshots for comparison.

[viral sonnet]

you what mate? 😄 do you run on a terminal or smth? 😄

[robust scaffold]

And maybe hints about why this loop couldnt be unrolled or vectorized. What's blocking the magical purple VP from occurring.

[viral sonnet]

we can make quite the list for the inspector 😅

the compiler itself 👌

[robust scaffold]

I'm sure they've got their own. There's little they can actually improve with the compiler IMO.

Like the method hierarchy is currently just a flat list. There needs to be some foldering and ordering by file.

[viral sonnet]

it's at a really good place already

[robust scaffold]

And maybe a burst option that allows a method to be hidden in the burst inspector list (with an additional toggle in the inspector to show the hidden methods). That way unity can hide all of their shipped methods and the inspector can focus on the user's code.

[viral sonnet]

i maybe have a dumb question about float precision. if i have a high number, like 1.0e52 and add +1 to it e52 times. is the result 2.0e52?

i have test it with e10 but the loops is too big lol

[rotund token]

1e5 = 100,000
2e5 = 200,000

i.e. 2.0e52 = 2x1.0e52

[robust scaffold]

Not possible:

[rotund token]

but yeah you are way past max float

[viral sonnet]

oh sorry, ignore the limit. take e31 or smth in range

[robust scaffold]

This question is possible with double.

[viral sonnet]

point is, is something that high even precise enough to pick up such a small number

[rotund token]

that's your precision table

[robust scaffold]

https://www.keil.com/support/man/docs/c51/c51_ap_floatingpt.htm

[rotund token]

at 16.7million (16e6) your precision is 1
anything above that and you're precision is worse than 1

so adding 1 does not guarantee changing the value

[viral sonnet]

so at one point if the numbers are big enough i'd need to store 2 values to even catch an increment of 1.

one variable just for precise accumulation

one to store the big number

[rotund token]

anything above 16,777,216 you need doubles for accuracy for 1 increments

but if you're goal is int accuracy, why not just use an int ^_^'

[viral sonnet]

because it doesn't go that high 🙂

incremental games struggle with data types 🙂

[robust scaffold]

Use a decimal then.

[rotund token]

just because float can go that high, i don't consider it useful for anything above about 100k if you care about accuracy

[viral sonnet]

simple example. you have a metric, you get +1 every 0.1 second (or something very small). you want to accumulate this. now the game is played for a year 24/7. it shouldn't stop working at some point is what i'm thinking about.

[rotund token]

yeah

this is why time is stored in double

not float

[robust scaffold]

I'm pretty sure decimal max value, or the value in which precision starts being an issue, is longer than the current age of the universe.

[viral sonnet]

incremental games often go over the double limit and need custom implementations

[rotund token]

ah

that's unlikely

[robust scaffold]

just reduce the incremental factor then

Pull a zimbabwe and chop off a few of the digits on the increments and then restart the count.

[rotund token]

if you increment +10/second
it will take
5.696113e+299
years
to exceed double max value

[viral sonnet]

it will stop working long before that. the above screenshot shows a double precsion of 1 with e15 or smth that I counted

[rotund token]

yeah so

the max value of 9007199254740992 for 1 precision
is 285,580,191 years if incrementing +1/second

[viral sonnet]

but yeah, at some point the increment has to get higher otherwise it's unrealistic to take any meaningful effect

haha, yeah if you put it like that 😄

as I said in the opening. maybe a dumb question 😄 anyway, the precision table helped to wrap my mind around it where it starts breaking down

[rotund token]

i have that table saved

it's super useful

especially when deciding if i want to use half

though its missing a few steps not hard to do the maths yourself

[viral sonnet]

yep, saved it too now. i remember you posting this some time ago

i want to add more stat data types now. i already have half and float. need double too and bring it all together

double for health and damage amount is super useful

at least as option

[rotund token]

haha i only offer half and short for modifiers

in a union

[viral sonnet]

and i need to find out how good burst is with jump tables. could get somewhat annoying to first get the data type and then the corresponding array of values

ah right i remember. yeah pretty cool. so you had an extension or how did you determine which stat is which data type?

[rotund token]

this is for modifiers

add is always short (+5 int)
increased/more is always half (30% increased damage etc)

the actual final stat is stored in an int always

i don't think it'd be healthy for a game to have a single stat modifier 8000% increased

😄

nearly always going to be <1

[viral sonnet]

depends on the type of game

[rotund token]

i was struggling to find any example

[viral sonnet]

those korean grinders i played went nuts with numbers

[rotund token]

8000% is still only a value of 80

which still has 0.5 precision

and at those crazy scales you don't need fine precision anyway

also personally i hate games with stupid numbers so i dont need to worry about that 😄

[viral sonnet]

eh, most shit the bed with bad number formating 😄

WoW did a huge stat squish because numbers were in the million range.

and they were sooo happy when numbers were again in the 1k range

i just thought. why not show 1234k? readable. good enough

but there are other reasons and those are all solid

it just went on for a very long time like this

and it always ended in the same way every expansion so everytime they needed a stat squish which also means numbers are totally relative and don't mean anything over the course

which makes me sad

[robust scaffold]

Dynamic buffer accessor is so big. Adds nearly 400 bytes to the thread's stack. I can't fit my own stackalloc in there.

[viral sonnet]

that's why i always saw such a performance decrease. what are those 400 bytes?

[rotund token]

i only see 20 bytes + 41 from safety

[viral sonnet]

and how did you see how much it puts on the stack?

[robust scaffold]

With and without

[viral sonnet]

are you talking about the size in the archetype ?

ah, nice

[robust scaffold]

300 bytes, not 400. Still, big.

Unity stackalloc 256 + 512 in their AABB overlap test. I just move mine to a native array. Not like I'm doing any vectorization on the actual nodes.

For comparison, this is 456 bytes.

[viral sonnet]

haha my biggest job has 936 in its execute

so methods that are inlined don't need an additonal stack alloc, right?

[robust scaffold]

If they're inlined, they get added to the method's stack. Most of the time, that's what you want but if you have a really big but short lived manual stackalloc, you want to have it NoInlining so it gets cleared at the method's return.

[viral sonnet]

i see thanks

[robust scaffold]

There's one stack alloc per method (not inlined) and all of them summed up equals the thread's stack alloc. Which is finite so dont use too many variables.

Too many as in stackalloc int[512] like what unity is doing...

[viral sonnet]

ok i see, i haven't accounted for that the memory is just moved to the parent method. what about variables that are calculated in the main method and then used as parameter for a sub-method. does this occupy 2 times the memory or will the method just read from the register of the main method?

[rotund token]

512 is not particularly large, 0.05% of the stack

oh wait 0.2%

[robust scaffold]

multiplied by 4 because of int

Probably depends on the comp running it as well. My shittop dies at 256 max for that screenshot bove.

[viral sonnet]

wait, is my math off. what cache size did you take? i get 6,25% for l1 with 32.768

[rotund token]

stack has 1MB

per thread on x64

[viral sonnet]

ah ok

[rotund token]

1mb = 1024x1024=1,048,576 bytes
he's allocating 512x4 bytes
2048/1,048,576 = 0.001953125 or 0.1953125%

hopefully my math checks out ^_^'

(I originally missed the x4)

[viral sonnet]

so uhm, how does the stack size differ from L1/2/3 cache?

[balmy thistle]

stack is just a section of main memory that employs a well understood convention of growing down and automatic cleanup

[viral sonnet]

oh man, why did i assume stackalloc puts data on l1 ... 😄

[rotund token]

stack is stored in ram but it has a high probability of matching in cache

[robust scaffold]

And if you shift the pointer off stack, you get the dreaded stack overflow!

[viral sonnet]

i was a master of stack overflows when i learned programming in c++. those damn recursive learning tasks 😄

[minor sapphire]

They called you the great flooder

[balmy thistle]

I used to effect magical memory optimizations back in the day by editing the default link script to shrink the stack size from 1mb to 64k

[viral sonnet]

those were the times when you had to restart your pc on some stack overflows or wrong pointers. and don't get me started on the dos IDEs. Borland C++ was it called

[timber ivy]

what exactly is burst doing to make my code run faster?

because I get very little performance gains using it. Infact im seeing an overall loss in performance just having dots installed before adding any user code

my editor stays at 90 fps vs the 600ish i would see before

[robust scaffold]

Using hybrid renderer?

[timber ivy]

yeah

but

even before installing that

[robust scaffold]

Thats why. It has quite a bit of overhead rendering nothing. That's why I dont use it.

[timber ivy]

How do you render then?

[robust scaffold]

Manual draw calls. SRP.

Not too difficult. That section of code draws circles.

[timber ivy]

is unity working on a better renderer?

[robust scaffold]

Yes. In 2022.2, it's looks to be completely overhauling HR to be far better.

[viral sonnet]

short answer, yes

[robust scaffold]

https://forum.unity.com/threads/new-batchrenderergroup-api-for-2022-1.1230669/

[timber ivy]

Awesome

[viral sonnet]

hybrid renderer will be in version 3 (soon)

[robust scaffold]

It is in version 3 right now, just DOTS will use it (soon)

As for Entities, the other half of DOTS, it's a storage medium. It shouldn't impact your FPS any more than using List<>s or Dictionary<>s

What burst does is translate the C# code into direct machine intrinsic operations. Making it faster and better.

Well, C# IL code.

[timber ivy]

so its basically an il to native jit compiler

[robust scaffold]

AOT in build. JIT in editor. But yes.

[timber ivy]

ok cool

maybe I should try a build mode and see what my fps is

[viral sonnet]

not needed. turn off safety checks and leak detection. there's little difference then to builds

at least when it comes to burst jobs

[timber ivy]

Then thats not the issue I already did this

maybe dots not optimized for cpus that make use of efficiency cores?

I got one of the new i7s with 8p+4e cores

[robust scaffold]

Dots without HR is basically a pile of arrays that's a bit flexible about typing. It doesnt care what the processor architecture it's running on.

[timber ivy]

oh i figured it was doing more than that behind the scenes

[viral sonnet]

when you mention performance issues. best to just post a screenshot of the profiler here with the timeline

[robust scaffold]

All the fancy things are mimicking dynamic typing of the arrays. You can roll your own Entities entirely without any packages. NativeArray is a core struct.

[timber ivy]

Im just talking about how if you go from a blank unity project with no code in it then install dots and press play your fps is significantly lower than before.

[timber ivy]

Is native array just a wrapper around T**?

[viral sonnet]

dots is quite a few things. as we already mentioned, HR is not that great

but you also don't need HR really to do something with burst, jobs and entities

[timber ivy]

entities jobs and burst installed alone reduces fps in the editor in a blank project by 50 fps for me

thats with safety checks and leak detection turned off

thats with a urp project

[viral sonnet]

there's something else going on. burst and entities itself don't do anything

[robust scaffold]

The default systems i guess?

[timber ivy]

if you install entities the default systems run

but

according to profiler

[robust scaffold]

The checks though are just boolean

[timber ivy]

they aren't taking that much time

[viral sonnet]

they shouldn't run really with no entities present

anyway, 50 fps don't mean anything when the framerate is 300 or 350. what counts is the milliseconds spent in cpu

fps is a reciprocal number and therefore relative

[elfin spire]

anyone knows of a compression repo that uses burst/jobs?

[rotund token]

unity just uses l4z

via CodecService

works in burst but it's just a native call - not burst optimized or anything

[winged bane]

Latios has some compression stuff that might be interesting
https://github.com/Dreaming381/Latios-Framework/blob/8e67573423d51e115a121e6cf4fb2ed5fd04e6f0/Kinemation/ACL_Unity/AclUnityCompression.cs#L243

[robust scaffold]

3ms just to find overlapping AABBs for 10,000 entities simultaneously. Not good.

[winged bane]

ouch

[robust scaffold]

I press multithread and it halves the time (cant multithread BVH building) but I dont know if it's valid.

[balmy thistle]

is this simdified

[robust scaffold]

Yea. Compares 4 AABBs at once.

I think the main issue is the adding of the entity that it overlaps with to the dynamic buffer. That isnt vectorized.

[winged bane]

Im allergic to dynamic buffers

[robust scaffold]

It's a great automatically managed NMHM.

[winged bane]

I havent found a use case where I needed something dynamic yet I guess

[robust scaffold]

An alternative option would be a fixed buffer struct and it'll allow for vectorized writes of the target entity, but that decimates chunk capacity.

[winged bane]

fixed buffers only good if you can control the size

but they are my bread and butter

[robust scaffold]

it's just so much empty space though. Dynamic buffers is great in that it's dynamic. Memory isnt wasted on empty slots.

I have 10,000 entities, each with a fixed Entity[16]. I want to be able to record every possible overlap but only in rare cases where all 16 would be used. Rest of the time, it's just sitting there empty

[winged bane]

my use case lets me fill them up pretty good

can every entity overlap with every other entity?

[robust scaffold]

Yea. I havent implemented collision flags yet so they can overlap with any other entity with a physics component

[winged bane]

what are you doing with the results of the job?

[viral sonnet]

have you considered writing to a temp array and then use AddRange? the temp array could be a list that you allocate once in job per thread

[robust scaffold]

I'm doing that right now in fact. (Also trying to figure out why unity is crashing).

[robust scaffold]

It's the broadphase of a physics collision detection algo

[robust scaffold]

Do you have dynamic buffers in your project ATM?

[viral sonnet]

yes, i use them for stats. inlined in chunk at fixed size

[robust scaffold]

I dont know why but setting internal buffer capacity for a buffer component to 0 results in unity hard crashing on launch.

[rotund token]

o_O

[viral sonnet]

are you doing something weird with it? setting to 0 means data is allocated outside of chunk. not sure if you want to do this in that kind of algo?

[robust scaffold]

I dont think this is because I'm abusing pointers everywhere. I dont even run anything until play.

[robust scaffold]

Outside the chunk is fine, it's only storing an Entity for the opposing entity that it overlaps with.

[viral sonnet]

writes will be slow though if you do it single

[robust scaffold]

That's the stack trace when I set the capacity to 0 and launch unity.

[viral sonnet]

how do you get the ptr?

[robust scaffold]

I dont, not for a dynamic buffer

[viral sonnet]

hm, that's very unusual then

[robust scaffold]

Top of the SS is the access. Bottom is the write.

[winged bane]

I havent had a chance to dig into physics much but there are some options for passing the data into your narrow phase

[viral sonnet]

never ever had a crash because of it

[rotund token]

where's the free occurring in entities

we have over 100 buffers or something stupid at work all 0 size

[robust scaffold]

Top of the SS is the componenttypes containing the buffer (Overlaps). Bottom is the add components. It works, if I set the IBC to anything above 0.

[viral sonnet]

and the struct of the buffer only has Entity in it?

[rotund token]

oh

i think i know your bug

can you make that a property =>

[robust scaffold]

Target?

[rotund token]

these here

[robust scaffold]

Oh, the components?

[rotund token]

yeah i had this bug a long time ago

[robust scaffold]

Why would properties change anything, these are all managed objects

Nope

[rotund token]

it's because typeof(X) => ComponentType.ReadWrite<X>() which requires using TypeManager
return FromTypeIndex(TypeManager.GetTypeIndex<T>());

[robust scaffold]

And it crashed...

[rotund token]

and typemanager doesn't get setup in time

oh you doing this in burst?

yeah that won't work

[robust scaffold]

The incremental builder is in burst, but the conversion works even with these managed arrays

[rotund token]

i'm pretty sure what's happening is outside of burst, your component has 0 size and exists outside the chunk

inside it has the default size

so thinks its putting it into the chunk

[robust scaffold]

But if I make it 4, no crashes

[rotund token]

yeah because that still exists in the chunk

[robust scaffold]

Oh fuck, the conversion system really does not like any dynamic buffers of size other than 4

Well then, I guess I'll be using a fixed buffer. Thanks unity for making the choice for me

[viral sonnet]

huh, i use 33

[rotund token]

you can't use ComponentType

before TypeManager has been initialized

a static field can cause this to happen

[robust scaffold]

Hrm, alright. Im gonna try this

And then convert that to componenttype[] inside the foreach

[rotund token]

yeah that's what i was suggesting because that's instantiated on use

[robust scaffold]

hrm

that didnt fix it

It's creating the component types inside the Entities.ForEach()

So this has to be after TypeManager been initialized.

Huh, same issue with my other dynamic array also created at conversion.

Seems that dynamic buffers can not have a size other than default (4) during conversion. After startup, it's fine.

[viral sonnet]

i have no idea why you have these errors, only thing i can say is that i can set all kinds of internal caps in and out of conversion

[rotund token]

all my buffers are setup during conversion

a wide variety of sizes

    [InternalBufferCapacity(6)]
    public struct ClusterConnection : IBufferElementData
    {
        public Entity Value;
    }```

```        void IConvertGameObjectToEntity.Convert(Entity entity, EntityManager dstManager, GameObjectConversionSystem conversionSystem)
        {
            dstManager.AddComponent<Cluster>(entity);
            dstManager.AddComponent<ClusterDirty>(entity);
            dstManager.AddComponent<ClusterConnection>(entity);```

that's a 6

    public struct BuildPreviewData : IBufferElementData
    {
        public Entity Value;
    }

        protected override void CustomConvert(EntityManager dstManager, Entity entity, GameObjectConversionSystem conversionSystem)
        {
            var previewData = dstManager.AddBuffer<BuildPreviewData>(entity);```
giant 128

[robust scaffold]

I'm making mine in a GOCS, not a IGCOTE

Im gonna see if it changes anything if I try a IGCOTE

[rotund token]

i have a bunch in conversion systems

    public struct SavableLinks : IBufferElementData
    {
        public Entity Entity;
        public ulong Value;
    }```

```#if UNITY_PHYSICS
    [UpdateAfter(typeof(EndColliderConversionSystem))]
    [UpdateAfter(typeof(PhysicsBodyConversionSystem))]
#endif
    [ConverterVersion("tim", 1)]
    public class SavableConversionSystem : GameObjectConversionSystem
    {
        protected override void OnUpdate()

// ...

            var savableLinks = this.DstEntityManager.AddBuffer<SavableLinks>(entity);

[robust scaffold]

hrm

None of my scripts touch conversion though, what did I break?

Alright, i tried a IGCOTE and it didnt crash... huh

What I have, which might be the issue, is a duplication script.

I think I might be overflowing whatever internal system they have for clearing memory during conversion

[robust scaffold]

Yea, it's the duplication. I'll need to do that on awake instead of during conversion it seems like

[robust scaffold]

hrmmm, actually. I think it's DOTS Hierarchy in the editor (before pressing play) that is crashing unity

Alright, nevermind. Just changing the IBC crashes unity. To fix, reopen unity, clear entity cache, and hope

[timber ivy]

what are you using this for?

[robust scaffold]

Broadphase tests. For physics.

[timber ivy]

doesn't unity handle this internally?

[robust scaffold]

Unity has their own physics package is if that's what you mean internally. I choose to make my own.

[robust scaffold]

Doing this cuts runtime by about 0.2ms, out of 1ms total. Not bad.

[viral sonnet]

nice, pretty good

[chilly crow]

sooo apparently this system queries for SettingsMenuState in order to run (I assume because of the GetSingletonEntity call), but I want it to run regardless.
should I be going about this a different way? I'm basically just using the component in another system with RequireSingleton and OnStart/StopRunning to control my UI state

public partial class ToggleSettingsMenuStateSystem : SystemBase
{
    protected override void OnUpdate()
    {
        if (HEVS.Input.GetKeyDown(KeyCode.Escape))
        {
            Debug.Log("derping");
            if (HasSingleton<SettingsMenuState>())
                EntityManager.DestroyEntity(GetSingletonEntity<SettingsMenuState>());
            else
                EntityManager.CreateEntity(typeof(SettingsMenuState));
        }
    }
}

[rotund token]

Add
[AlwaysUpdateSystem]

[chilly crow]

pog thank you

[robust scaffold]

I think this works. It should hopefully without exploding.

Yep. It does. Nice. Love it when code works first try.

[viral sonnet]

what's going on here? not seeing the use of this

might have to do that i have no context what compIdx, compDat is

[robust scaffold]

CompX is the "compressed" arrays of entities.

Compression is basically is vectorized select that also shuffles the true components to the left.

[viral sonnet]

hm, i don't think you need to pin them btw

[robust scaffold]

I'm using Span<> instead of direct pointers for safety.

These are random hardcoded array sizes and if any of them overflow, span will throw an error whereas a pointer will do random things until it hard crashes unity (learned that the hard way).

[viral sonnet]

and that works with pinning the ptr to it?

wouldn't that require using the actual span struct?

[robust scaffold]

Yea. Span is built into C# and has neat properties. It's basically a wrapper around a pointer.

And you use fixed to get the original pointer out of Span.

[viral sonnet]

i know, but you are using pointers then and not the span. how can that be safe?

[robust scaffold]

hrm, true. I dont actually index the span.

Used to before I wrote that memcpy stride section

but then I rewrote it, and now I dont use the span...

Well, guess I dont need the span.

[viral sonnet]

yep, for that kind of safety. hardcode a const somewhere convenient

and use an assert

[robust scaffold]

I was originally using it alongside this span as well (which is being indexed here)

[viral sonnet]

oh wow, math.compress is left packing. i actually have a simd left packing in my code that's not working because i have no idea how to actually write such a thing and the lookup table

great name ... who calls something like that compress

[robust scaffold]

Yea. Math has some really neat functions. I'm using this because that's what Unity physics does.

[robust scaffold]

Yea, no clue.

[viral sonnet]

oh i know, leftpacking was too verbose ... oh wait

[rustic rain]

Sooo, span is like reference to stack memory structure?

[viral sonnet]

it's a convenient (safe) struct for a stackalloc

so yes. no idea if it can resize. never used it really

[robust scaffold]

And apparently the other "hidden" neat math functions are countbits and tz/lzcnt for bitwise ops.

[robust scaffold]

Cant resize.

Span can also be used to fix regular arrays but that will never come up in burst as you can not make managed arrays.

[rustic rain]

So stackalloc just let's you allocate any amount of data and span let's you to cast this allocated data to your chosen type?

[viral sonnet]

you can do it with pointers or reinterpret too. also unsafeutility.As

[robust scaffold]

Stackalloc allows you to allocate any type of data and returns that data as a pointer to the first value. Span is just a neat wrapper that contains the length of the stackalloc as well for safety when indexing into it.

[rustic rain]

Hmm

So span is like unsafelist?

[robust scaffold]

Since I'm not indexing into the span (and using it's safety), i changed the stackalloc back to int pointers.

[rustic rain]

Just a fixed sized pointer to allocated memory?

[robust scaffold]

No, it's like a NativeArray. It contains the length of the final array, not the number of elements.

[viral sonnet]

dat leftpacking must be nice for early outs on something. that at least was the use case i wanted to tey but failed

[rustic rain]

Oh, so it is size of arraycount multiplied by type size?

[viral sonnet]

why are you making this complicated. it's like a nativearray but allocated in stack

[robust scaffold]

The array count in the stackalloc is like a regular array, not in byte size. In that SS, i can fit 64 ints into compIdx or 256 bytes.

It's a NA(Allocator.Temp) that disposes automatically at method return, instead of whenever unity feels like it.

[rustic rain]

I just don't really understand it 😄

[viral sonnet]

it's really fast and on par with preallocated mem. should be used instead of temp nativearray in jobs

[robust scaffold]

And yea, leftpacking is purpose built it seems like for tree testing. BVH, Octree, Quad, any type of tree.

[viral sonnet]

i have tested it comparing stackalloc with a pre-allocated nativelist and there was no real improvement

[rustic rain]

Any examples how it looks like from allocation to disposal?

[viral sonnet]

no dispose needed, when the method returns it's gone

[rustic rain]

Hmm

[viral sonnet]

float* arr = stackalloc float[16]; for example

[rustic rain]

Is there a limit to amount of data?

[viral sonnet]

i wanted leftpacking for these kind of tests

[robust scaffold]

1 MB per thread generally.

[viral sonnet]

yeah 😄 the stack size

[robust scaffold]

Left packing results in 3 vectorized operations. No loops. Very clean.

[rustic rain]

So, 1 mb is size of stack?

[viral sonnet]

what do you say, would this work for my case?

to at least simd this test

[robust scaffold]

https://docs.microsoft.com/en-us/dotnet/csharp/language-reference/operators/stackalloc

[rustic rain]

Ok

Now it finally makes sense

Just temp native array but faster hehe

Combined with span

[robust scaffold]

I've posted above how I'm using to compare 4 (or less, it's a bool check) entities against 1. So if you can guarantee a multiple of 4 entities you're comparing, left packing will work.

[viral sonnet]

ok, i can do this with 2 loops then. i don't always have multiples of 4

thanks, i need to go to bed 😄 o/

[robust scaffold]

You can create a guaranteed next multiple of 4 array then memcopy in. Then do all these checks.

[rustic rain]

stackalloc byte[MaxStackLimit]

Huh

[robust scaffold]

yea, just dont use too much because this is how you get those infamous stack overflows.

[rustic rain]

I feel like I'm learning cpp at this point xD

[robust scaffold]

Yep. Burst is C++ brought to the world of C#.

At least you dont need to write header files.

[rustic rain]

But span is indeed a nice learn

Along with stackalloc

Microoptimizations for performance critical applications

[robust scaffold]

This aint just micro. This is how you can conduct physics on thousands of individual objects simultaneously in real time.

Might be micro for each entity but summed up over 10,000 or 50,000, it's a lot.

[rustic rain]

I assume this means crash?
Security

The use of stackalloc automatically enables buffer overrun detection features in the common language runtime (CLR). If a buffer overrun is detected, the process is terminated as quickly as possible to minimize the chance that malicious code is executed.

[robust scaffold]

Yep. That's the downside. Writing C++ means losing the nice soft crash and informative error messages when things overflow.

And you hard crash unity as well. I spent 7 hours yesterday figuring out that I mistyped j instead of k inside a for loop indexing a raw pointer. It kept crashing unity with no error messages outside of Null reference exception (buffer overrun). And slowly walking back to where the issue is.

[rustic rain]

Hehe

Btw. Native array allocated to heap?

[robust scaffold]

Yep.

[rustic rain]

Welp

Smth I didn't know

[robust scaffold]

You can tell because adding additional capacity to a NA doesnt change the stack allocation quantity inside burst.

[robust scaffold]

Span's not just limited to stackalloc. I've wrapped NativeArrays around these to get a pointer like access with the buffer overrun safety of a NA.

[timber ivy]

Is there away to set an entity as a navmeshmodifier?

[robust scaffold]

The GO nav mesh? No built in way. There is a DOTS nav mesh though, open source and free.

[timber ivy]

I was under the impression navmesh was compatible with dots

im using the expirimental ai package if that makes a diff

[robust scaffold]

https://github.com/dotsnav/dotsnav

[timber ivy]

is it good?

[robust scaffold]

Never used either nav systems myself. I use flowfields for my pathfinding.

I plan for a few thousand entities and flow fields scale better.

[timber ivy]

I just need something simple for pathfinding tbh. My game is a top down 3d game and the ground is 100% flat

[robust scaffold]

Cant go wrong with A* with a few dozen at most pathfinders. Beyond that, nav meshes are faster up to about a thousand. Then flowfields win out.

[timber ivy]

Id have about 250 units pathfinding at most

thing about a* is I can't understand how I go about marking area of my world as unwalkable

[robust scaffold]

make your world a grid. Larger grid sizes mean faster A* performance but worse quality of path.

Unity's nav mesh query, i think the thing you're looking at, seems to be a burst accelerated GO system, not a DOTS system.

If you want to use the nav package, you'll need to go hybrid and dip into GO land to update the nav mesh using monobehaviors

[timber ivy]

yeah I dont want to mess with game objects so that isnt an option

[rustic rain]

Pretty sure there's a navmesh in dots

Since tertle discussed it with someone else

@rotund token could you clarify?

[robust scaffold]

dots nav was someone in the dots community, it's fully featured and very robust but not Unity's package

[rotund token]

Unity's default navmesh works from dots

[jaunty hornet]

Yeah you can use this API if you want to use inside job. https://docs.unity3d.com/ScriptReference/Experimental.AI.NavMeshQuery.html

[rotund token]

^

[timber ivy]

But can you generate a navmesh at runtime on an entity?

[rotund token]

yes

[timber ivy]

how do I add them to the navmesh modifier list?

cause right now I got this and it works fine for a gameobject but once things got put into subscenes it stopped generating the navmesh

[rotund token]

i told someone how to do this hte other day

[timber ivy]

private IEnumerator RebuildNavMeshAsync()
    {
        Bounds navMeshBounds = new Bounds(new Vector3(observer.transform.position.x, 0, observer.transform.position.z), new Vector3(boundsSize, 20, boundsSize));
        modifiers = NavMeshModifier.activeModifiers;

        for (int i = 0; i < modifiers.Count; i++)
        {
                markups.Add(new NavMeshBuildMarkup()
                {
                    root = modifiers[i].transform,
                    overrideArea = modifiers[i].overrideArea,
                    area = modifiers[i].area,
                    ignoreFromBuild = modifiers[i].ignoreFromBuild
                });
        }

        NavMeshBuilder.CollectSources(navMeshBounds, surface.Value.layerMask,
                        surface.Value.useGeometry, surface.Value.defaultArea, markups, sources);


        var navMeshUpdateOperation = NavMeshBuilder.UpdateNavMeshDataAsync(
            navmeshData, surface.Value.GetBuildSettings(), sources, navMeshBounds);
        yield return navMeshUpdateOperation;     
    }

[rotund token]

and it was you >_>

[timber ivy]

no you didn't tell me how to add entities to the navmesh modifier list

[rotund token]

you just have to add your rendermesh

to the list of meshes

you can't use CollectSources

you have to collect them yourself

[timber ivy]

Ok thank you

[rotund token]

just build a list of NavMeshBuildSource

from your entities

sourceObject is the mesh from RenderMesh

transform is LocalToWorld

think that's all you need from memory

Area I guess

[timber ivy]

I am sorry if you did mention this when you helped me build the code before I may have overlooked it cause I wasn't using unities ecs at that time. Decided to switch a few days ago after seeing how much better the api got

[rotund token]

for bounds, you can either use a known area

or you can calculate it from combining all your WorldBounds

apart from that, it just takes the same navmeshdata and buildsettings

[timber ivy]

so the navmeshmodifiers aren't needed when using entities then?

[rotund token]

i don't think they're require for game objects either

they just let you override the area with a different type if you want

if you don't use them it'll just use the default area

[rustic rain]

Has anyone touched FMOD with Dots?

any idea how can you call Events with entities?

[rotund token]

we use fmod

[rustic rain]

any general tip how is it structured in project?

[rotund token]

fmod is basically native

you can just call it

[rustic rain]

yeah, but it has string field in Editor

which kind of makes it a bit harder

for type EventReference

so it comes to resolving guids manually

[timber ivy]

cant you just marshal the string to a ptr?

[rotund token]

you dont pass fmod the string though

(i dont think, could be wrong on that)

but the wrapper is literally just

native calls with a ptr and values

[rustic rain]

oh wait

it's FMOD library?

not your advanced helper utils? xD

[rotund token]

i have no utils for fmod

this is work

i dont use fmod in my own project

this is just the default fmod sdk

[rustic rain]

yeah, interesting

I still wonder about how you structure it in entities

if you even do

[rotund token]

we don't

our audio was written before entities even existed

[rustic rain]

ah

[rotund token]

all the fmod string convert to guid

and you can pass those direct to fmod to get eventids etc

[rustic rain]

fmod has it's own ID manager?

interesting

[rotund token]

i'm no expert

audio guy asks me to hook up some data

i do

that's about the extent i've used it

apart from setting up the audio device switcher i think

[rustic rain]

I see

[rotund token]

i just know it's pretty much a native library

which has the nice benefit of being callable from burst

and the c# wrapper they provide is really tiny

[rustic rain]

hmm

[rotund token]

so should not be that hard to add your own little burstable wrapper instead

[rustic rain]

I cannot really find any actual examples of being called from burst

All I found are managed solutions

[rotund token]

just go into fmod_studio.cs

and look at the external calls

and call them directly yourself

[rustic rain]

and even some dud who made a guide of how to use it with entities also used Managed systems

[rotund token]

or just go look at a tutorial

for c++

instead of c#

[rustic rain]

external calls can be made from burst?

[rotund token]

burst can call native libraries

[rustic rain]

huh

so you can make a cpp method and call it from burst?

that's really interesting to know

[rotund token]

yes

[rustic rain]

I'd assume as long as parameters are unmanaged and burst compatible

[rotund token]

my navmesh library is in c++

and just called from burst

[rustic rain]

and so is return type

[rotund token]

thats a limitation of the native call

[rustic rain]

but Transform

[rotund token]

?

[rustic rain]

it's managed and it's external call

[rotund token]

this is not a DllImport

[timber ivy]

You can pass pointers to structs from c# to c++ as parameters to pinvoke calls

oh nvm

[rotund token]

^

[timber ivy]

i misread

Transform is an object type

[rustic rain]

hmm, can't actually find it

even github is 404

kek

oh found it

I think

https://github.com/fmod/fmod-for-unity/blob/2.02/Assets/Plugins/FMOD/src/fmod_studio.cs

hmmm

So the key to native calls

is System class

with only field of IntPtr

I am guessing this is similiar to EntityManager

as I can get reference to it in OnCreate

and then call it in bursted code

looks like true

private void Update() { Debug.Log(_system.isValid()); }

not a single false

very interesting

    [StructLayout(LayoutKind.Sequential)]
    public struct PARAMETER_DESCRIPTION
    {
        public StringWrapper name;
        public PARAMETER_ID id;
        public float minimum;
        public float maximum;
        public float defaultvalue;
        public PARAMETER_TYPE type;
        public PARAMETER_FLAGS flags;
        public GUID guid;
    }

They even have their own string wrappers kek

[rustic rain]

hmmmm

how would conversion work with it though...

[rustic rain]

UNITY_DISABLE_AUTOMATIC_SYSTEM_BOOTSTRAP_RUNTIME_WORLD
Where do I need to define it so it works?

it still compiles with automatic bootstrap

oh

maybe # is wrong one

Does World get's disposed on SceneChange?

Can I keep it?

[dreamy glade]

System.ArgumentException: Unknown Type:`{0}` All ComponentType must be known at compile time & be successfully registered. For generic components, each concrete type must be registered with [RegisterGenericComponentType].
Anyone know how to fix this issue? In my code, I'm attempted to add a custom dynamic buffer to an entity: ecb.AddBuffer<BlockBuff>(entityInQueryIndex, ent); //Add new block data array. but the error then occurs. From what I've seen online, I have to add an AssemblyInfo.cs in the same directory as the script to define it prior to compile time: ```using Chunks;
using Unity.Entities;

//Used to define component types prior to runtime.

[assembly: RegisterGenericComponentType(typeof(BlockBuff))]
[assembly: RegisterGenericComponentType(typeof(DynamicBuffer<BlockBuff>))]```

I would appreciate some help on the matter if anyone knows any solutions! Thanks.

[rustic rain]

can you show implementation of your buffer element data?

declaration

[dreamy glade]

          public bool isExist;     //Does this block exist within the chunk?
          public int colorIndex;   //Index position of color map
          public int textureIndex; //Index position of texture map
          public int faceCount;    //Index position of texture map
     }```
Sure thing. Thanks.

[rustic rain]

ahem

looks fully legit

Remove RegisterGenericComponent...

that shouldn't do any good

hmmm

are you certain that this exact buffer is causing error?

I'd also recommend clearing cache and restart editor

[dreamy glade]

To my knowledge, yes. The error doesn't show where the issue is being caused at but commenting out those lines removes the error.

[rustic rain]

entities, especially subscenes are really buggy lately

[agile cradle]

Hi. Does dots have any instruments(api) to detect in systems removing or adding new components? I've used State Components and custom tag components to detect this actions but it is clunky solution.

[rustic rain]

no

might want to figure out better solution

structural changes like these are very expensive

when there's a lot of entities

that require such reaction

[solemn hollow]

look into ChangeFilters

[viral sonnet]

in job structs, chunk.DidOrderChange detects structural changes

[agile cradle]

WithChangeFilters in systems foreach? I guess this feature detects only component changes not adding/removing.

[solemn hollow]

i think it also detects structural changes in chunks. but if you only need to look for add/remove look at enzis suggestion.

[agile cradle]

thanks for advice!

[rustic rain]

but it detects all structural changes

not of certain component

[solemn hollow]

very true. a built-in feature to run systems when certain components are added/removed would be quite nice.
Sure you could do that initialisation when you add a component right at that point where you add it but youd not be able to split up systems in a clean way. like if an entity gets a health component id not want to initialize the HP bar right away where i spawn the entity. id want an extra system for UI decoupled from the gamesystems.

what i currently do is adding one "Initialized" component to entities that need to do some stuff in the first frame on spawn. likewise i have tags that request initialisation of certain components (which was @agile cradle clunky solution). havent found anything better :/

[rustic rain]

potentially

I see a way it can be done

if you quite literally write your own ECB

which detects it and stores in some container

as events

[rustic rain]

Unbelievable

FMOD is burstable

holy

btw, physics stutters doesn't exist in build

I have no idea, what is causing this nonsense in editor

[robust scaffold]

You mean the sliding across surface stuttering?

[rustic rain]

no, I mean stuttering due to changing translation value in update

[robust scaffold]

Is the camera following the moving entity?

[rustic rain]

yes

[robust scaffold]

Oh, that's just camera update issues.

[rustic rain]

doubt

works totally fine if I set physics body type to static

[robust scaffold]

Like the update of the position of the camera should be located in late simulation group so it pulls the correct position from the target entity

[rustic rain]

position of camera is done by cinemachine

and it follows companion link

[robust scaffold]

Uh, no clue then

[rustic rain]

it's some kind of weird physics doing

[whole gyro]

I gave up on cinemachine due to camera stutter issues. Switched to a custom monobehavior for the camera controller running in late update and it works fine now

[rustic rain]

you can select where to update cinemachine though

you can even manually update it

but once again, those stutters are from physics

not from camera

[whole gyro]

Its been a while but I remember messing with all of those settings and couldn't get rid of the stutter completely. Didn't try manually updating though so maybe that would have worked.

I'm not trying to say that the problem is from cinemachine necessarily. Never figured out the root cause. But switching to a custom camera controller fixed the stutter issue for me.

[robust scaffold]

Finally, after hours of digging through the Entity source, I have converted my tilemap:

All you need is some asmref to pull out unity's neat function that extends supported companion object types:

And conversion from there is straight forward and works.

[rustic rain]

usually things like that worked for me by default

if you simply AddComponentObject

[robust scaffold]

That works for general objects that dont also need to run their monobehavior update loops. Or it's already on the white list.

[rustic rain]

oh

that explains why I need to update Animator manually

[robust scaffold]

Yea. I do not want to roll my own tilemap and Unity's tilemap depends on the existance of a grid and tilemap component. So companion GO is the only way for it to work.

I think I finally understand the conversion system, after all these years

[robust scaffold]

Finally converted my tilemap. Took.... weeks.

Burst 1.8.0 Pre-2 is out. Key changes I've noticed:

• Setting a breakpoint in an attached managed debugger (Rider/VS Unity Debugger...) on a method that is burst compiled, will switch off the burst code path for that method, allowing it to be debugged as normal.
• Added toggle to filter Unity tests on and off.
• Assembly is now searchable either through CTRL + f or the contex menu associated with the inspector view. Search options include case sensitivity, whole word match, and regex.
• Added focus on current job in the burst inspector.
  @viral sonnet @rotund token

I think they ported the burst compiler not detecting changes from 1.7.2 to 1.8.0pre2 as well. Gonna upgrade now and see this new inspector

Holy shit, it's so clean. No more unity tests clogging everything

[robust scaffold]

A bit more looking around, havent noticed anything else immediately

[rotund token]

oh it's per method

not per job

that's even better

good thing i have an ultrawide o_O

search is new right

can actually find the code i'm looking for like instantly so nice

[rustic rain]

hmmm

how do I quickly find

Lambda's code?

what is key word for that search?

[rotund token]

in burst inspector?

just search for a line of your own code

[rustic rain]

well

oooh

hm

nope, didn't work

[rotund token]

oh yeah they don't always properly generate the line stuff

[robust scaffold]

Switch between full debug and then back to minimal debug (default) to refresh the yellow line comments

[rotund token]

had a corruption within 30min

=S

Already logged

[robust scaffold]

oof, guess the improvements to the hashing system didnt really improve much

[rotund token]

interesting it was in my 1 entities.foreach in the entire library

coincidence? probably not

but maybe

maybe it was just entities codegen corrupted

i didnt actually test if restarting without nuking cache would fix it

[viral sonnet]

you guys made me excited to update. guess i'll wait 😅

nice to find a ctrl+f feature!

[robust scaffold]

Did you know, a rectangle has 4 points? That's 4 floats for X, and 4 floats for Y. 4 means vectorization.

[winged bane]

[robust scaffold]

@viral sonnet Needed the ptr CDFE. Hope ya dont mind me taking a copy of ya repo. What's the improvement for BufferTypeHandleFast?

[viral sonnet]

you're welcome 🙂 i should update the repo again. it's nothing big tbh, uses a lookup cache.

[robust scaffold]

The type names get longer and longer and longer

[viral sonnet]

haha, yeah I think it would be worth it to just rename it to CDFEX 😄

[robust scaffold]

Or CdfeUnsafe

UnsafeCDFE. There ya go

[viral sonnet]

yeah, good one

[robust scaffold]

I dont remember, is this mixing of CDFE and CDH possible?

[viral sonnet]

yes, but you need to disable safety

[rotund token]

you will get alias errors if you are writing

[viral sonnet]

on the cdfe

[robust scaffold]

All of these are readonlys

[rotund token]

it's obviously not safe if you're reading from cdh and writing cdfe to same entities ever

[viral sonnet]

ah yeah, then it should work

but if it complains, you know what to do 🙂

[robust scaffold]

The massive disable safety tags.

Im pretty sure Unity made the tags so ungodly long to discourage people from using it.

or at least the peasants without ultrawide monitors

[viral sonnet]

in times of auto completion i have not much issue with them. all start with Native

[robust scaffold]

Just a suggestion under the chunk extensions:

public static bool Has<T>(this ArchetypeChunk chunk, ref BufferTypeHandleFast<T> bufferComponentTypeHandle)
    where T : struct, IBufferElementData
{
    var typeIndexInArchetype = ChunkDataUtility.GetIndexInTypeArray(chunk.m_Chunk->Archetype, bufferComponentTypeHandle.m_TypeIndex);
    return (typeIndexInArchetype != -1);
}```

unsafeCDFE makes the tower of properties slightly better

[viral sonnet]

it does

forum post suggested they will add the cache to the buffer but it's not in 0.51. just checked

[robust scaffold]

Cache?

[viral sonnet]

the lookupCache

[robust scaffold]

Oh, maybe in 1.0 then. BufferAccessor should just be an array of pointers though.

[rotund token]

the Update() on CDFE/BFE will be nice to have as well

[viral sonnet]

i have added my collections now to the repo

parallel list and array hashmaps

the array hashmap work on either one array with a key offset or 2 arrays. fastest hashmap i could make 🙂 parallel list is basically a multi threaded list writer with 128 nativelists to write to

usage is like a nativestream, just that it expects chunkIndex and don't foget to set the chunkCount before the job is scheduled

the index can be abused for whatever 😄

i should probably make the name more generic. BeginForEachChunk doesn't suggest that any kind of index can be used

[rotund token]

curiously what's the advantage of your array hashmap

over just writing to the key/value arrays directly in a regular hashmap

is it just api convenience?

[viral sonnet]

pretty much, yeah. easier to setup and the biggest difference is the 1 array option that only uses value but no key array

the parallelList can also be indexed with ParallelListHashMap. not using that though in my project

[robust scaffold]

Another suggestion:

public bool TryGetComponentRefRO<TUser>(Entity entity, out TUser* ptr) where TUser : unmanaged
{
#if ENABLE_UNITY_COLLECTIONS_CHECKS
    AtomicSafetyHandle.CheckReadAndThrow(m_Safety);
#endif
    CheckComponentIsZeroSized();

    var ecs = m_Access->EntityComponentStore;

    var hasComponent = ecs->HasComponent(entity, m_TypeIndex, ref m_Cache);
    if (hasComponent)
    {
        ptr = (TUser*)ecs->GetComponentDataWithTypeRO(entity, m_TypeIndex, ref m_Cache);
    }
    else
    {
        ptr = null;
        return false;
    }

    return true;
}``` Allow for the the pointer returns to be cast designated as whatever the user wants. For easy inline reinterpretation but no safety checks.

[viral sonnet]

ah, yeah. makes sense

[robust scaffold]

Here's my use case. CircleReal is just a IComponentData wrapper around a single float. I can now inline the reinterpret without having to use another var.

[viral sonnet]

no need to explain. good idea. 🙂 i have added it

[robust scaffold]

This is what happens when you have 9 components in a single job.

I dont know, should I split this into separate jobs or have one megajob