GPU Zen 3: Virtual Shadow Maps

[mystic lark]

Okay, either we outsource everything to you or I learn PowerPoint 😅

[sweet nimbus]

Guess who's back

[mystic lark]

I did nufin, irl stuff came up ☹️

[sweet nimbus]

all good. I'll get back to it soon™️

[vestal snow]

I'll see if I can move it into google's slide thing

no idea if that will work lol

[mystic lark]

Don't worry about it that much, I have PowerPoint from uni

Should be fine

Deadline is Sunday right?

[sweet nimbus]

uh

April 14th, 2024

Second Peer review period and all articles are final

April 28th, 2024

We start uploading software to GitHub

May 17th, 2024

Release on Amazon.com

For some reason I thought the draft was due Saturday

[acoustic bobcat]

All articles are final?

That's pretty soon lol

Crunch time

[sweet nimbus]

I asked over email and the editor said it can be in a rough state at that point (the 14th) 🤷

[mystic lark]

28th is very soon

Wait

Or am I misreading and 14th is final

Tbf we were given a 3 weeks notice

Hard to navigate around

[sweet nimbus]

It says final but not really (for us)

[mystic lark]

Okay some breathing room

At least

[sweet nimbus]

But we should still try to get all the content in

[mystic lark]

Yes

I agree

I just need to cope

[humble shadow]

Good luck folks, i'm sure it's going to turn out nicely

[acoustic bobcat]

Read the article aloud to yourself, it will help examine the flow

[sweet nimbus]

Duck moment

[acoustic bobcat]

We should call it rubber frogging to motivate GPers to do it more

[mystic lark]

Read the article out loud in VC to increase the pressure

[wooden jolt]

the VSM share thingy is the most up to date article right?

the official™️ thing doesn't compile for me (timeout )

[sweet nimbus]

yes

you have to compile in draft mode

[wooden jolt]

isn't the clip equation wrong

currently it's max(ceil(log2(T_w/T_0)), 0) shouldn't it be ceil(log2(max(T_w/T_0), 0))

[mystic lark]

Max over division of sizes?

The division will always be positive

I'm confused

[wooden jolt]

sorry, you're right I don't know my own heuristic

it's ceil(log2(max(d / c_0, 1.0)))

but yeah the max there is pretty useless

can very well be shrimplified to clamp(ceil(log2(d / c_0)), 0, C)

[mystic lark]

True true

Does the stuff I wrote about their respective benefits/disadvantages make sense?

[wooden jolt]

yes

so far I only wrote one comment

and it's a minor thing

[mystic lark]

Very good

[sweet nimbus]

just keep familarizing yourself with the article

and read one or two of the other articles to get a feel for the style and organization

[sweet nimbus]

[acoustic bobcat]

For a technicial writing tool, overleaf sure sucks at autocorrecting technical language

[sweet nimbus]

ye

also I hate writing introductions

so much

I wonder if the editors would be bewildered by a little cheekiness and alliteration

gonna have a shakespearean intro and the rest will read like the vulkan spec

[sweet nimbus]

the discussion in #1147711082702589973 made me realize that we should at least briefly mention why we decided against using tiled/sparse textures

[acoustic bobcat]

Would tiled/sparse textures replace the virtual memory, or just allow parts of the "physical" memory store to not be resident or something?

[sweet nimbus]

API sparse textures just do the virtual stuff for you

[acoustic bobcat]

ah gotcha

[sweet nimbus]

but it requires API calls on the CPU which is yucky

[acoustic bobcat]

I assume you don't use that just because you cede a lot of control over the algorithm?

ah

[sweet nimbus]

yeah pretty much

so we'd have to do readback (latency) and making lots of small pages resident is incredibly slow

[sweet nimbus]

I structured the intro and previous work in a way that hopefully makes sense now

intro: explains problems encountered in shadow mapping
related work: explains CSM, SDSM, and screen-space shadows, then explains how Epic added VSM to UE5 and our motivation for creating our own version of it

the UE5 stuff had to be mentioned somewhere and it felt like this flowed better than putting it with the other stuff in the intro

[sweet nimbus]

@mystic lark you prefer "physical page", right? I think it's better than "memory page"

[mystic lark]

Yeah I like the naming physical and virtual but we need to introduce the naming and change everything to use it

[sweet nimbus]

shouldn't be too difficult

"memory page" would need to be defined anyway

hmm your allocation scheme is much more complex than mine tbh

which is ok as long as it's adequately described

I also see the term "shadow map page" used sometimes

is that a virtual page?

[mystic lark]

Yeah shadowmap page == virtual, memory page == physical

[sweet nimbus]

alright those should be updated

btw I made a bunch of edits and comments

some of the comments are action items

we need some code listings then some pictures

unfortunately pictures are kinda screwed up right now because we cannot do a full compile to see them

but code and tables are perfectly visible in fast compiles

when you're done reviewing my tiny edits please accept them to remove the spam (I kept the spam because I want you to see what I changed)

alright I'm going to bed now

@wooden jolt you best be looking too. not much more needs to be written, it's just a few smallish sections at the end

[wooden jolt]

I am lurking worry not

[sweet nimbus]

there's still a couple giant blocks of text that have no home but I'm sure we'll find them a home or relocate them to dead

[acoustic bobcat]

What's the status of the share doc, it looks the same as when I saw it last

mostly anyways

[mystic lark]

I'll copy it over before I go to bed again

[acoustic bobcat]

Make sure you've incorporated the comments since they might get overwritten when you do that

I saw lvstri added a bunch

[mystic lark]

Good point

[soft island]

if ya'll want another pair of eyes, i'd love to take a look (mind you I'm a person very inexperienced with shadows) 👀

[sweet nimbus]

o

@wooden jolt you should be looking at the main article

[acoustic bobcat]

added a bunch of comments, not text to be clear

[sweet nimbus]

ye the thing we sent you was in a sorry state

[mystic lark]

Still gave us very useful feedback

[sweet nimbus]

I was wondering where lustri's comments were lol

[vestal snow]

(vvv is this ok to post here??)
this is the first attempt to move saky's mapping drawing to some kind of image or diagram thing

I haven't been able to move everything over since I'm not entirely sure what the state of the version we're using is

e.g. exact page entry bits, are we using the same visited/marked strategy and buffers, etc.

oh crap I just realized if you have dark mode enabled you'll have to open it in browser to get around dark on dark text

[mystic lark]

Oh this looks very good!

[mystic lark]

Virtual page has allocated, visited and dirty

Physical page has allocated and visited

And they both have the respective coordinates to the virtual/physical page

Ie physical page has coordinates of the owning virtual page

And virtual page has coordinates of the backing physical page

[vestal snow]

that sounds like

basically two page tables right? bidirectional

[mystic lark]

Yep, the reason for that is the allocator

And it's just handy in general, sometimes you like to iterate over the physical pages and sometimes over the virtual ones

[vestal snow]

what's the format of each page table

do they both use R32 UINT, then R32F for actual depth memory

[mystic lark]

Both are u32 images

Yes, exactly

Maybe I use R32UNORM for the depth, I'm not actually sure

[vestal snow]

i'm guessing that while you have NumCascades virtual page tables
you probably only have 1 physical page table right, assuming a single memory pool?

[mystic lark]

Yes, that's why it's so handy for the allocator

Instead of iterating over all the x cascade Virtual pages, potentially for multiple light sources, you just run over the memory pool

[vestal snow]

ok that makes sense

let me modify the diagram. I'll use VPT (virtual page table) and PPT (physical page table) to shorten

[mystic lark]

Okay, we can make those abbreviations in the article also

[sweet nimbus]

I haven't needed a bidirectional allocator tbh

But my allocation scheme is dead shrimple

[mystic lark]

It's mainly needed for the deallocation

So when I find a physical page that is allocated but not visited this frame I need to modify the previously owning virtual page to no longer hold the allocation

And the only way to sensibly do that is by storing a pointer back to the owning virtual page in the physical page

[sweet nimbus]

Pershrimps

I use a bitmask to represent which physical pages are free, and scan it in the allocator to find a free bit. Deallocation is just looking at the page table to find pages that are backed but not visible (after marking visible pages), then clearing their bit from the bitmask

[mystic lark]

Ye but you clear all not visible pages each frame no?

[sweet nimbus]

I clear the visible bit only

If they are marked visible once more, they remain backed and don't have their physical contents cleared

[vestal snow]

mine is similar to jaker's except with atomic queue

[sweet nimbus]

There's no point in explaining mine though because the article describes tido

[vestal snow]

are we shipping both allocators

ohh

[mystic lark]

Aha I see, yeah but then you have to scan over all the virtual pages to find something to deallocate

[sweet nimbus]

I have one pass that deallocates them in bulk and it takes basically 0 time to execute

My allocator is also completely moronic but takes 0 time lol

[vestal snow]

I actualy hadn't ever considered a bidirectional allocator

this is the first time I'm thinking it over lol

[sweet nimbus]

It's kinda funny how much time we spent bikeshedding allocators

Thinking it would be slow

[vestal snow]

yeah, we each ended up with our own

actually I still need to ctrl+a, backspace my sparse allocator

no need for 2 allocators anymore

[mystic lark]

Uhh yeah, I think I was prematurely optimizing, but this actually allows me to do some cool stuff later

[sweet nimbus]

But I think saky's allocator permits off-screen page caching which could be sweet

[mystic lark]

Yes it does that

[vestal snow]

you can do offscreen with linear allocator too

though I ended up disabling offscreen after we collectively decided it wasn't that helpful

[mystic lark]

It's cool as a fallback

Especially if I make only the higher clips be able to stay off screen cached for example

[vestal snow]

well at least to an extent

up to 15 frames unless you use more PT bits lol

[mystic lark]

Which is trivial with the allocator I have rn

[vestal snow]

let me upload the new diagram with bidirectional

only thing still missing is what looks to be some sort of

queues in the first diagram

[sweet nimbus]

Lol in dark mode the arrows on the right look like a portal from Portal

[mystic lark]

Don't worry about the queues for now

I've still yet to decide whether it makes sense to add them to the article

But the diagram looks very good now I like it a lot

[sweet nimbus]

What are these queues you speak of

[mystic lark]

One is the allocation request buffer, then I have free pages and not visited pages buffers

[sweet nimbus]

ah ok

[mystic lark]

Allocation request is just a buffer of coordinates to pages that need allocation this frame

[sweet nimbus]

The article calls them buffers, not queues

[mystic lark]

Free and not visited is what allocator puts out

[mystic lark]

Oh ups, sorry

I copied the naming from the convo here

[sweet nimbus]

Np I was just confused

[mystic lark]

(at least I think that's what J Stephano was referring to)

[vestal snow]

yeah I think that's what I was looking at in the original

i wasn't sure how to incorporate them into the new one

[mystic lark]

They are not really necessary for this diagram I think

I feel like it has enough as is haha

[sweet nimbus]

I think the top right table can give the wrong impression of how the cascades are stored

Because the lower levels are overlaid onto the higher levels in it

[mystic lark]

I think we can have a single cascade in the diagram

[sweet nimbus]

if they had a 3D tilt effect and were slightly spaced so you get a hint of the pages being covered, it would convey what happens more accurately

[mystic lark]

And then note in the description that you have one vpt per cascade and multiple vpts can point into the same ppt

[vestal snow]

I think I see what you mean

would it be like

almost like the old pyramid diagrams we were seeing when struggling with clipmaps

but more compact and slightly turned on the side?

[sweet nimbus]

Saky

Are there any places you want to add code listings

Which might make the description clearer perhap

[mystic lark]

I am really lost with the code snippets

I think we should do one for sampling perhaps

[sweet nimbus]

Sure

[mystic lark]

Show how you reproject, fetch clip level, Virtual page into physical page info

Idk if we want to do something more

Maybe the culling?

[sweet nimbus]

Lemme think

Hpb?

[mystic lark]

You mean building it?

[sweet nimbus]

Well I'm asking you lol

What part of culling

Something like this function maybe?
https://github.com/JuanDiegoMontoya/Frogfood/blob/2f821bfc6634ab94eb6444b296006844c06f0343/data/shaders/shadows/vsm/VsmCommon.h.glsl#L130

[mystic lark]

yeah something like that

but I'm not sure, seems kinda random

maybe just pseudo calls of fuctions to describe the process

[sweet nimbus]

It can also be used to visually break up the text hehe

Also

People seem to be super interested in the vsm drawing shader

We should at least provide that

[mystic lark]

that is a good one yes

simple enough so that it's not wall of code

but important and shows a bunch how we use VSMs

the whole lookup process

I like it

[sweet nimbus]

The vsm sampling one is good too

Then we need to sprinkle diagrams throughout

And move those big chunks of orphan text

And fill out the last couple sections

Then we'll be ready for review 😎

[mystic lark]

Easy crunch tomorrow

we are also missing text for stable adressing, drawing and culling

[sweet nimbus]

Fuggit Imma crunch today

[mystic lark]

hpb building

[sweet nimbus]

Yeah

[mystic lark]

https://tenor.com/view/april-fools-joke-dog-its-fine-this-is-not-gif-16757454

[mystic lark]

Okay I'm going to bed so I can rise early tomorrow

[sweet nimbus]

Eh I can crank those parts out

[sweet nimbus]

Gn sweet prince

You did good

[mystic lark]

We all are! It's gonna be one fine article

wait I just realized

you are behind right, not ahead (time wise)

so it is tomorrow

the deadline I mean

[sweet nimbus]

yes I'm 9 hours behind you

it's only 4:44pm on Saturday here

I'm not sure what timezone the deadline is supposed to be. I'm guessing it's quite fuzzy in reality

[prime ice]

west coast frog

[acoustic bobcat]

Let me know if you want some style linting done at any point

Finished my taxes so I'm just gonna work on my shit otherwise

It's very irritating that track changes isn't enabled

[sweet nimbus]

yeah

[acoustic bobcat]

When I edit papers I usually touch nearly every single sentence in some way

[sweet nimbus]

is there different software we can use

[acoustic bobcat]

Which is pretty hard to manage without track changes

Idk

[sweet nimbus]

I guess I could get a subscription real quick so you can make edits without receiving psychic damage

[acoustic bobcat]

I wonder if it would work if you paste a modified version over an old version, will it do a diff and generate the tracked changes?

Or will it just say the whole sentence was replaced

Nah don't get a subscription just for me

[sweet nimbus]

ah there's a free trial

[acoustic bobcat]

Ok your choice

[sweet nimbus]

it will just say the whole thing got changed even if the contents are 99% the same

[acoustic bobcat]

Darn

It's more to protect you from psychic damage than me, I can just edit it or comment out every sentence I edit to keep the old copy, it will just be annoying to go back and see what changed

Reading every sentence twice to catch the changes

[sweet nimbus]

fyi @vestal snow @wooden jolt @mystic lark I made a copy of Saky's copy of the article (lol) except I got a trial for the standard version, so it has track changes enabled

everyone got an invite and demongod is already in it

[vestal snow]

oh ok so does that mean we should write it in the new location, then copy over to GPU Zen when we're done?

[acoustic bobcat]

I think the opposite probably

If you guys all have access to the real version, the other one is just for those of us who aren't official authors

[sweet nimbus]

actually it might be better to write in the new place because the official one doesn't even compile

[acoustic bobcat]

lmao

[sweet nimbus]

but we'd have to copy the book config which might be a pain in the ass

or not

[vestal snow]

I'll start adding figures to the

[sweet nimbus]

o I was about to do that

well from the main article

I'm just copying the folder structure

[vestal snow]

oh ok cool

[sweet nimbus]

if you have more content to add then go ahead

[vestal snow]

will be nice to compile it and write feedback on them

I did but I realized one diagram I made might actually not make sense at all

so I think I need to redo it first

[sweet nimbus]

ok so I'm going to copy all the stuff from the main article into this one, then we continue in there

I renamed the article to reflect that

I will review some stuff in the main article (there are a bunch of tracked changes and comments) then copy them over

[vestal snow]

wait

is it just me or does the copy have multiple copies in it

[sweet nimbus]

I'm about to overrite everything

I'll preserve your 2 new figures

just gimme a sec (I'm doing this between turns in d&d )

[vestal snow]

no problem lol

For the depth analysis section, the paragraph says something like "fragment is reprojected into the cascades uv coordinates" -> does the current impl still do uv wrapping like before

[sweet nimbus]

we'll have to wait for saky to wake up to answer that I guess

[vestal snow]

sounds good. I'll upload the diagrams and we can edit them when he comes back

[sweet nimbus]

everything is copied over and compiles now

pictures render 🥳

[acoustic bobcat]

Nice

Is it ready for edits

I can take a look after I eat dinner

[sweet nimbus]

hmm I want to do more work before I ask for your time

[acoustic bobcat]

alright

[sweet nimbus]

I'm copying comments from the other document too

so our thoughts are in sync

hmm

ah I remember what I wanted to say

I feel like being liberal with pictures right now

[acoustic bobcat]

This seems like a lot of work for something that is ostensibly due tonight lol sorry

[sweet nimbus]

it has a soft due date of tomorrow

it's fine to look now. the obviously crusty stuff are clearly marked by comments

[acoustic bobcat]

Alright

[acoustic bobcat]

Btw when you say tomorrow do you mean tomorrow evening

[sweet nimbus]

it's unclear but I'm somewhat confident that nothing will be looked at by the editor until, say, monday morning

the guy editing our section was clear in email that our draft can be quite rough

[acoustic bobcat]

Ok cool

I'll just edit tomorrow then

[sweet nimbus]

yup that's perfectly fine

[acoustic bobcat]

I'll just look at it when I wake up and you can see the results when you wake up lol

[sweet nimbus]

just got mega confused by how figures are not necessarily placed in the location in the text where you would expect

fortunately the solution is shrimple
https://tex.stackexchange.com/a/32605

[mystic lark]

Uh yeah, ! Usually works but the latex gods need to bless you

Sometimes it doesn't care

[sweet nimbus]

you suggested a glossary earlier but I'm not sure the ideal way to do it

there is a glossary package but the main project doesn't use it

I guess we can do a good old list

[mystic lark]

I wouldn't do a glossary

it usually is enough to just introduce the term first in text

[sweet nimbus]

alrighty

[mystic lark]

and then in brackets the abbreviation

seems like there are no comments in the official GPU zen doc

is it just me?

[sweet nimbus]

wdym

ah

yes

we are using a new doc

I moved the comments over manually

#1168692074447642664 message

[mystic lark]

I see, I misread when I woke up and thought we ought to still be using the old one

[sweet nimbus]

I got a trial account while you were gone so we could have track changes lol

[mystic lark]

hahaha okay smart smart

[sweet nimbus]

and we can actually compile the thing

can I get a clarification on how the clear pass works

I presume you generate a list of pages to clear, then to an indirect dispatch where one workgroup clears one page

that's what I do anyway

[mystic lark]

I just read the allocation requests

[sweet nimbus]

right as long as you have a list somewhere

[mystic lark]

yep, it's basically the indirect dispatch struct as for the allocate pages

[sweet nimbus]

ok I'll note that

[mystic lark]

first pass generates the allocation requests buffer - allocation pass tries to give all these a physical page - clear pass looks at the pages in the requests buffer, if they got allocated it clears, otherwise allocation failed so it ignores

[sweet nimbus]

I will note at the beginning that our implementation is gpu-driven and does not use sparse/tiled resources

that is an important detail imo

[mystic lark]

at the beginning of the implementation section?

[sweet nimbus]

ye

[mystic lark]

perfect, yeah good idea

[sweet nimbus]

the beginningn't 😄

[mystic lark]

I shuffled some of the parts that were out of place

I'll have breakfast and do a review pass rewording the senteces

I am not happy at all with the stuff I've written

[sweet nimbus]

ye fuel up

[sweet nimbus]

I'm rewording what needs to be reworded

what you gave is, at the very least, a great skeleton to work from

[mystic lark]

oh btw, before I go - hear me out

I know you vote for simplicity in the allocator, but I don't like that the allocator setup phase has two buffers - one with the set of visible pages and one with the set of free pages

[sweet nimbus]

whose, yours?

[mystic lark]

yep

[sweet nimbus]

we can cope

[mystic lark]

what I want to do eventually, is have them all be in the same buffer and sort them in ascending order based on index where: free pages get index - 1, not visible pages get index = clip level of the page

too complex for the article/fine as is for now or do I make that change?

[sweet nimbus]

how does it work when multiple visible pages are in the same level

[mystic lark]

it doesnt really matter

[sweet nimbus]

um

I think we should just stick with what you have

[mystic lark]

the idea is that you try to free the pages that are living in lower clips as these get cleared super often anyways and are super fast to draw (culling goes brr on these)

[sweet nimbus]

the only thing I want to maybe try hacking in before the code is submitted is support for dynamic objects

[mystic lark]

so you offscreen cache only the higher mips (if there is enough memory to do that)

[mystic lark]

okay

[sweet nimbus]

I quite like that idea but I don't think it's critical enough given our time constraints

I think supporting dynamic objects is the minimum we need for anyone to take it seriously

[mystic lark]

oh yeah, we will need to make that stuff up and implement later

hehe

Well I have a pass that is not yet in, that runs as the first one in during the Setup phase

which clears the wrapped pages

[sweet nimbus]

wats a wrapped page

[mystic lark]

uh how do I explain

when you move one tile forward, one row of tiles gets out of the vsm frustum and the front most row of tiles wraps onto it

[sweet nimbus]

ah

[sweet nimbus]

I remember worrying that it would be difficult to support that case, but then my allocator being so simple that it doesn't care

[mystic lark]

this

[sweet nimbus]

yeah

[mystic lark]

I have that in

[sweet nimbus]

in the impl?

[mystic lark]

yeah

[sweet nimbus]

but not the article

[mystic lark]

not yet

I add it today

[sweet nimbus]

hm aight

https://tenor.com/view/make-it-so-picard-gif-5445593

[mystic lark]

but I was thinking we make the dynamic object tile invalidation part of this?

[sweet nimbus]

yes but most of that would be on the CPU I imagine

[mystic lark]

yeah exactly

[sweet nimbus]

I was thinking the CPU would generate a bit for each page saying whether an object moved in it

[mystic lark]

this pass already takes info from the cpu (this tile offset - prev tile offset = offset we need to clear in this pass)

[sweet nimbus]

for each object that moves, project its AABB into each clipmap and set the bit for every page spanned by it

[mystic lark]

yeah, not really GPU driven, but no time to figure that out

[sweet nimbus]

I think it would scale up fine tbh. we only worry about gpu-drivenness for things that would otherwise require readback

[mystic lark]

hmm true

[sweet nimbus]

some impls might already have this info on the GPU so doing it there would be a natural fit, but we do not

[mystic lark]

we could abuse the rasterizer to do this for us, rasterize at page table resolution so each pixel shader invoc would be one page

and then we'd atomic or the dirty bit

[sweet nimbus]

that's indeed what we're basically doing, except we'd need conservative raster

[mystic lark]

yeah so that we don't kill something that is subpage size

true

[sweet nimbus]

let's keep it simple and do it on the CPU

[mystic lark]

I guess it's fine to do it on the cpu for now

[sweet nimbus]

yeah our demo will have like 1 object moving at most, and the worst case is that it's a huge object covering every page of every clipmap

which is only like 32^2 * 16 (16k) pages to set a bit for

[mystic lark]

I was more worried about how it scales with many dynamic objects

• some people do animation and shingus on the GPU

but let's not worry about it that much

[sweet nimbus]

considering that the worst case is 16k tiny ops for a dynamic object, it's not too bad

[sweet nimbus]

UE5 solves this by just considering animated stuff as "always moving"

but there is a flag to override that if the movement is subtle enough

[mystic lark]

honestly we can handwave it a lil - "clear pass additionally takes in a mask of pages that got invalidated due to moving objects. This mask can be generated by projecting dynamic object bounding boxes into the pages and marking all affected pages as dirty. It can be implemented both on the CPU or the GPU. In our implementation, due to its simplicity, we chose the CPU path."

[sweet nimbus]

agreed

gives us a little leeway in the implementation

alright now go eat breakfast

[mystic lark]

we can even not mention the last sentence

yeye I eat haha

[sweet nimbus]

I'm going to change "paging table" to "page table" also
https://en.wikipedia.org/wiki/Page_table

hmm I didn't notice that there are two kinds of page tables in your impl

as long as they are disambiguated every time they're mentioned then it's fine

[mystic lark]

Yeah there is the physical page table and virtual one

[sweet nimbus]

wait

[mystic lark]

We will call them ppt and vpt

[sweet nimbus]

is the physical page table just the physical memory?

[mystic lark]

It's like a header for the memory

[hazy steppe]

I are you adding a section of what stuff you have not considered/tried or think should be explored/or notat all further at the end of le article.. things you couldnt cram into the article because of time, but might inspire future frogs to pick up on that

[sweet nimbus]

yeah we will

[mystic lark]

Ie one physical page is represented by a pixel in physical page table

Which just holds some meta information

[sweet nimbus]

ok

[hazy steppe]

good luck

[mystic lark]

(the pointer back into VPT and allocated/visited bit)

[sweet nimbus]

I suppose it's accurate to call it a page table then

A page table is a data structure used by a virtual memory system in a computer to store mappings between virtual addresses and physical addresses.

[mystic lark]

Hmmmm

Yeeeeah

It's like when you have a block of memory preceded by status block of that memory

But since my physical memory is an image not a buffer I didn't really have a spot to fit this header into

So I made it a separate image

I initially called it memory meta info

But then I shifted it to fit with the other naming scheme

[sweet nimbus]

if it acts like a page table then it's fine lol

just gotta make it clear that there are two distinct page table-like things

[mystic lark]

I'll add it to the implementation preamble

[sweet nimbus]

what are the dimensions of your cascades and pages

mine are 4096^2 and 128^2 texels

[mystic lark]

same as yours

although they are all defines and everything should work if I just change them

as long as the change is reasonable and not something like 2pixel pages

[sweet nimbus]

ye I note that it is flexible

[sweet nimbus]

I'm gonna use "virtual page table" and "physical page table" as previously suggested

hmm the physical page table stores redundant info, no?

[mystic lark]

yeah

the two bits (allocated and visited) is reduntant

[sweet nimbus]

I guess that's fine to prevent a dependent lookup, but I think we should omit that we store this info again because it might be confusing

[mystic lark]

hmm hmm

yeah I agree

[sweet nimbus]

it can just be a comment in the source code

[mystic lark]

might make it a bit harder to word around certain things but I think it will be fine

[sweet nimbus]

well

we can just say "for convenience, x and y bits are also stored in the PPT"

[mystic lark]

It is an implementation "detail" so I think that should be enough yeah

[sweet nimbus]

I'll just write something then let you evaluate 😄

[mystic lark]

I hope I'm not touching the text too close to you @sweet nimbus, I'm trying to avoid the parts you are currently working on

[sweet nimbus]

lol you're right on my tail

🏃‍♂️🌬️

[mystic lark]

I'll hop down for a bit and have to leave in 15 minutes or so, so you'll have time to finish 😄

I also got writeful form my uni (something like grammarly, but aimed more at researchers, so it was trained on papers and such) So I'll do a full sweep once I return

[sweet nimbus]

trying to figure out how to refuckulate the intro again

ok I have un idea

[mystic lark]

okay I'll let you cook

I'll be back in an hour or two

the visual clatter OL produces though

[sweet nimbus]

I need to take a 3-6 hour "nap" soon or I'll become stupid

[mystic lark]

oh right, did you even sleep??

I can take the reins after I come back and you rest

[sweet nimbus]

my sleep schedule is wack so I'm just starting to become tired

anyways I just want to fix the intro before I sleep

[mystic lark]

if you manage to find the time for it, skim over (and accept or reject) the changes I did pls

[sweet nimbus]

realistically that will happen after i wake up 😄

[mystic lark]

okay

thats also fine

[sweet nimbus]

I saw some of the changes already and they were good

[mystic lark]

My main fear is that I change one of your sentences/you one of mine, I don't like it so I change it back and we get into a cycle of us changing the sentence over and over

we should probably start a comment discussion if this happens hehe

[sweet nimbus]

yeah

hmm

I think we can delete the section header of "previous work" actually

then it won't seem so weird that we introduce VSM there

[sweet nimbus]

I'm copying what we have back into the main article in case they start looking at it today

I'm relatively happy with the intro, now just gotta finish up the end

after eep

[mystic lark]

I'll recover and add all missing parts into the implementation section

[mystic lark]

Reading about sparse makes me wonder again haha

hmm nevermind, dumb idea

[acoustic bobcat]

General comment, ideally figures should appear before they're mentioned

I know placing figures is a PITA in TeX and it will probably be changed later

Or if not before they're mentioned then at least on the same page

Also, for the final article the figures should have their font match the article text, ideally

The editor may have general comments about figure style to try to keep the articles consistent with each other

[vestal snow]

What’s the article font/size? I’ll switch the figures to use it

[acoustic bobcat]

Don't worry about it yet

I'm not sure exactly what the font is, it's just the overleaf default

If there's any uncertainty someone could contact the editor and ask if the figures need to be finalized by today or if there will be feedback on style

The figures are quite good btw nice work

Based on the book I saw I'm guessing there's going to be a fair bit of layout and format work yet to actually turn this whitepaper-style text into book content

[mystic lark]

@acoustic bobcat are you editing the part you are on rn, or can I modify it? nvm

[acoustic bobcat]

I'm commenting/making suggestions

I'm not sure what happens when I make suggestions over someone else's suggestions so I might just make comments in those places

[mystic lark]

suggestions? or do you mean changes?

[acoustic bobcat]

Like if someone else has edited something and that edit exists as a "track changes", and then I edit that edited text, idk what happens

Seems like it could become a mess

[mystic lark]

I think it overwrites, but I wouldn't worry about it that much

I feel like the "track changes" just blow up and become very hard to read

although I try to read them still, it just become huge chore

especially when we need to work on it

[acoustic bobcat]

Yeah idk what the best solution is

You guys shouldn't really be using them though

Suggest changes is for secondary editors, if you guys are the primary writers you should just modify it tbh

[mystic lark]

I'd say just edit the sentences when you see an obvious mistake - grammar or syntactic - when you are unsure make it a suggestion comment

[acoustic bobcat]

I restructure whole sentences frequently though

It works well when there's only one person using them but for me to add a bunch on top of existing ones I'm not sure

Jaker should be awake soon maybe he can just apply all his

[mystic lark]

I've been applying Jakers 😅 I just leave mine if he wants to see them

[acoustic bobcat]

Maybe we should just apply them

The version history is tracked anyways

Not like they're going to be rejected

whereas mine very well may be since I don't have all the context and shouldn't make direct edits without review

[mystic lark]

It was mainly meant to show to Jaker what I was working on while he was eeping, but as I said, it is just a soup now

so I'll just apply them all

as we have not really been arguing about anything - the corrections we make over each other are 99% fine (at least from my side that is)

[acoustic bobcat]

Alright

don't apply mine though

What's the difference between light space matrix and frustum

If the light space matrix is a tighter fit does that mean the frustum is too

[mystic lark]

I'd say light space matrix defines the frustum

ugh well, light projection matrix does, light view decides the position of the frustum

btw @vestal snow I think it would be nice to have a diagram depicting coordinate wrapping when using stable addressing and caching

Ie the "Toroidal mapping" as NVidia calls it in GPU gems

something like this

[vestal snow]

Oh yeah that sounds like a good idea, I'll get started on it

in your impl do you also use the uv coord wrapping scheme?

[acoustic bobcat]

I mentioned this in the comment but I think "bookkeeping" (recently added in the text by someone) is actually an even better name for the first stage

"allocation" is better than "setup" but "bookkeeping" is even more general and better captures what's going on

Bookkeeping, drawing, and sampling

[mystic lark]

it is a weird word to use in an article 😄 but it does fit well

I'll make it bookkeeping

[acoustic bobcat]

Nah it's fine

It's a common term

[acoustic bobcat]

Alright I just finished a pass through

I will wait for you guys to continue and can check back again later

[mystic lark]

Thank you I'll address the comments before I go to bed, after I'm done writing a bit more

[acoustic bobcat]

Feel free to discard or question any suggestions I made, they're just suggestions after all

I won't be offended and I don't have full context so sometimes I am guesing in my interpretation of the meaning

[mystic lark]

as well as the camera following the player at a page sized offsets to allow for page caching

[sweet nimbus]

I'm awoke

[sweet nimbus]

It's the proj*view of the light

I guess they're the same lol

[mystic lark]

is that an industry term so to say?

[sweet nimbus]

Yeah I think so

[mystic lark]

I did use that syntax all over the article

okay then it's good

[acoustic bobcat]

Like when you say that SDSMs are a tighter fit, is it the frustum that fits more tightly?

And the matrix is just describing that frustum

[sweet nimbus]

Yeah we should use the word frustum in that sentence

[acoustic bobcat]

Alright

I changed it

I changed that whole sentence so I just changed that word

I was sort of agonizing over whether I could make that swap

[sweet nimbus]

Pain

Btw I think I had a dream about writing the article

[mystic lark]

Pretty much the whole of todays process is just rewording senteces

only like an hour ago I got to start on actually writing new stuff

[acoustic bobcat]

Also sorry about the edit spam from the equations lol I should have modified the whole thing in one go rather than in pieces

The easiest way to review edits is to just click on them in the editor and let it take you to the associated comment

[sweet nimbus]

Yeah that's what I've been doing

[mystic lark]

I attempted to describe stable addressing and page wrapping

but it is very rough

for some reason I really struggle with explaining these concepts

[acoustic bobcat]

Can you explain it verbosely here

[mystic lark]

okay

Stable addressing and page wrapping need to be introduced in order for caching to work

caching just means, that pages no longer get invalidated when not visible. They instead stay allocated until there are not enough free physical pages, at which point they get freed

[acoustic bobcat]

What about wrapping

[sweet nimbus]

Caching also means not having to re-render a page if it's still in view and nothing in it changed

[acoustic bobcat]

Yep

[mystic lark]

Yeah so, stable addressing makes the camera follow the player in a page sized increments

[acoustic bobcat]

Working on caching right now for my foliage

[mystic lark]

as opposed to following the play position exactly

additionally, you need to fix the camera position to only slide along the "near plane"

By doing this, when the player crosses a page boundary and the cascade light frustum snaps to a new page, you can still do the shadow test on pages that were drawn with the previous camera position

ie, you can preserve the cache and don't need to clear it

[sweet nimbus]

The issue can basically be summed up as "how can we permit the light camera to move without invalidating cached pages"

[mystic lark]

The issue that remains is, that when light matrix snaps one tile forward a new row/column of pages enters the frustum and a row/column of pages leaves it on the other side

Page wrapping is an addressing scheme, which maps the pages that enter the frustum onto the location where the pages leaving the frustum used to be

like so

[acoustic bobcat]

That's a handwriting font right

[mystic lark]

yeah

lmao

[acoustic bobcat]

nobody's handwriting is actually that neat

just making sure

[mystic lark]

my actual handwriting

[acoustic bobcat]

That's a handwriting font right, nobody's handwriting is actually that neat

jk lol

May as well be compared to mine

Anyways I think I see the wrapping thing now

How is this related to invalidation though

[mystic lark]

this image sums it up well

[acoustic bobcat]

Right

I've done this in foliage rendering as well

[mystic lark]

The boundary needs to be invalidated/cleared

[acoustic bobcat]

But in the article there's something about page invalidation being based on wrapping

ah ok

ah yeah that makes sense

[mystic lark]

I made a comment on that yeah, I started the section, but then realized there is a bunch more I need to explain before I can talk about that. That is why there was just a one random sentence...

[sweet nimbus]

After I eat this hotdog I'll be on

[mystic lark]

I have like 1-2 hours in me

then I fear I become more trouble than help

[sweet nimbus]

cap

you're always worth

but it's good to get proper rest

[mystic lark]

btw what is "pretty rough state" defined as, do we know?

like can parts of the article just be missing?

probably not right?

[sweet nimbus]

Let's try to get the first draft until the April deadline - it shouldn't be super polished, just let's have the first round of the comments then.

[mystic lark]

hmm so uhhh, that tells us not very much haha

[sweet nimbus]

Yeah I will aim to have a version of the missing parts by EoD

[sweet nimbus]

Exactly, but given how loose they've been with the other deadlines I guess nothing is set in stone

That's why I'm not stressing over the article being perfect today

[mystic lark]

I'm just asking if I should make another coffee, or if it will okay-ish to be missing a bit

[sweet nimbus]

Ah don't lose sleep over it if you don't want to or if you need to do other stuff tomorrow

[mystic lark]

we'll see how it goes

[sweet nimbus]

I'll see if I can recruit my irl friend to take a look at well

[acoustic bobcat]

Irl friends

right those exist

Anyways I can definitely give it another read over today

Write more stuff and I'll review it

And then you can go through all my remarks

[hazy steppe]

what timezone and time is it due?

[acoustic bobcat]

yes

[hazy steppe]

: )

[acoustic bobcat]

(nobody knows)

[hazy steppe]

[acoustic bobcat]

Today, in some timezone

[hazy steppe]

i assume wolfgang sits in california

perchance 14th EOB, pacific time

[acoustic bobcat]

Yeah like jaker said they probably just want it on their desk monday morning

[hazy steppe]

raisinable

[sweet nimbus]

ah

they added some comments to our main doc a few hours ago

actually just one

file missing

fixed it

we have a warning that OverallMapping.png is too large for the page

and the glsl language couldn't be found for a listing

[hazy steppe]

are those limitations of overleaf?

[sweet nimbus]

https://gist.github.com/bbenno/b4d7c880453b7967dbf84c9765b965bd

[mystic lark]

we can make it cpp

the listings are borked anyways

( in the Shared version)

because we do not have their styling

[sweet nimbus]

ye

[vestal snow]

I made an attempt at making a diagram for it

[sweet nimbus]

yeah the font in the system overview needs to change eventually

[vestal snow]

I definitely need to edit it

[sweet nimbus]

and the squished arrows in the bidirectional mapping look kinda funny

perhaps use straight arrows to avoid that

unless you can figure out how to make them not squished 😄

[vestal snow]

oh in the

the first diagram

let me change that

[vestal snow]

oh crap

[sweet nimbus]

make the pics really high res if you can so they are at least 300 dpi

[vestal snow]

yeah I think I can

are they all a problem or just some?

Like should I reupload all

[sweet nimbus]

it looks fine tbh, don't worry about it

[sweet nimbus]

I don't know if any of them are a problem. I was just letting u know for if you edit them

[vestal snow]

also I crammed the wrap addressing diagram into the caching section

[vestal snow]

sounds good, I'll try to keep them all high res

[sweet nimbus]

What is an acceptable image resolution?
In general bigger is better. We encourage the authors to think about the resolution rather in terms of printing resolution, dots/inch (dpi). Our target recommended resolution is 300 dpi. This means that a 1080p image would be 1920/300=6.4 inches wide on the actual printed paper. If you need to make your image smaller, this is fine too, because it would lead to higher than 300 dpi. However, all images should have at least 300 dpi pixel density for printing.

[mystic lark]

The bookkeeping flowchart also needs to be remade to preferably fit the style of the other figures

[vestal snow]

I added "px" and "py" to the diagram so we can reference it in text and mention that they are both multiples of a page

[mystic lark]

I've also added a fifth stage, so the old figure is outdated now

[sweet nimbus]

btw is the official doc using the float and listings packages?

[mystic lark]

yeah

I copied that from them

[sweet nimbus]

ok good

[mystic lark]

or well, from their style thingy

and I copied the syntax of the listings from the cyberpunk article

so it should be fine regardless

@sweet nimbus are you gonna go through Demongods suggestions or should I do it?

[sweet nimbus]

I'm going through them rn

[mystic lark]

okay I'll just spit out more text

[sweet nimbus]

sounds good

also I just realized a section describing our debug views would be helpful but we can focus on the important stuff right now

[mystic lark]

hm do you think it is relevant?

seems a bit random to me

[sweet nimbus]

well I think for anyone wanting to implement vsm, it would be nice to know

but yes it's kinda hard to fit into the existing structure, so let's ignore it for now

[mystic lark]

that is true yeah, it will be a single paragraph anyways so we can add it later if we decide to

[sweet nimbus]

maybe we could put it after the conclusion as some sort of "extra" section

idk how papers normally handle miscellany

[mystic lark]

Appendix

is usually the way to do it

[sweet nimbus]

ok I looked it up and that does seem helpful/like the way to do it

[vestal snow]

alright the portal arrows were replaced by a double ended straight arrow

if that diagram is too big I can squish it down

[sweet nimbus]

we will need to remember to re-upload all the pics in the main doc too if you edited any in this one

[vestal snow]

oh yeah they're pretty much all out of date in the main one

and they still all have to change at least 1 more time to fix the font

[sweet nimbus]

making green and yellow highlights disappear gives me a dopamine rush

[mystic lark]

it's the little things that keep us going

big blocks of green give me the good feels haha

[sweet nimbus]

track changes is so insanely useful damn

[mystic lark]

well, unless there are a lot of changes

[sweet nimbus]

this was worth paying $0 for the free trial

[mystic lark]

I hope you don't mind I accepted both of ours changes before you got to see them

and I just left Demongods suggestions

[sweet nimbus]

yeah that's fine

I'm reviewing his changes and accepting them rn

[mystic lark]

good good

[sweet nimbus]

man is the last cascade really 65x65 km

[mystic lark]

2^15, I checked it

[sweet nimbus]

isn't it 2^16 if the first one is 2x2 meters

[mystic lark]

lmao

yeah

[sweet nimbus]

that's kinda crazy

UE5:

By default, clipmap levels 6 through 22 are allocated virtual shadow map page tables. This means that the default settings have the most detailed clipmap covering 64 cm (2^6 cm) from the camera position, and the broadest clipmap covering about 40 kilometers (2^22 cm).

[mystic lark]

6 cm

WHY?

no way you need 128 texels for 6 cm

[sweet nimbus]

erm, 16k texels

because the whole clipmap covers 6cm 😄

[mystic lark]

I guess the culling just kills everything so they are esentially free

but ehhh

[sweet nimbus]

and that's level 6 still

levels 0-5 are even smaller

[mystic lark]

Unreal needs to be stopped lmao

[sweet nimbus]

someone rejected my metadata edit 😠

[mystic lark]

metadata edit?

[sweet nimbus]

I think metadata > meta-information

[mystic lark]

oh lmao

I think I overwrote it back

sorry

[sweet nimbus]

I'll un-overwrite it

[mystic lark]

🔫

I think it was an accident, as I like meta-information better lol

[sweet nimbus]

metadata is an actual word though 😄

[acoustic bobcat]

Perfect for my purposes

[sweet nimbus]

you can trivially add more cascades to make the number even more absurd

[acoustic bobcat]

You can barely even see 65km in the atmosphere unless the conditions are pristine lol

[sweet nimbus]

what about non-visible wavelengths

radio imaging

[acoustic bobcat]

I have done it

But raytracing is necessary

[sweet nimbus]

I know we do it for astrophotography, but I'm guessing the resolution of stuff on earth would be shite

[acoustic bobcat]

Hmm to leak project images or not to leak project images

Nahh maybe later

It's not my project to leak

[sweet nimbus]

yeah don't do it lmao

[acoustic bobcat]

It's not for work is more side project stuff

[sweet nimbus]

even though I'm guessing you have something epically relevant

[acoustic bobcat]

Yea

[sweet nimbus]

I'll just imagine you sent the relevant pic and I replied with

[acoustic bobcat]

I'll post it when I can

For now enjoy my college personal project that is also relevant

https://i.imgur.com/QOuUcih.mp4

[sweet nimbus]

ah yes, ultrasound

I think it's a food baby

[humble shadow]

My mom had the same when I came up

[mystic lark]

why is it so satisfying to watch the scan like update?

[vestal snow]

Do we have any assets that let us test the last cascade

wasn't there a landscape someone found or made a while ago

[acoustic bobcat]

My time to shine

[mystic lark]

Ye Jaker had the terrain

[vestal snow]

[sweet nimbus]

I can put it on gdrive

[acoustic bobcat]

Btw this ran on a single thread of my shitty laptop

Well, one tracing thread and one video thread

[sweet nimbus]

do you support ktx textures

or do you want unprocessed gltf

[vestal snow]

I should but not yet 😦

so unprocessed or glb would be better

[sweet nimbus]

hmm I'm not sure where I put it

I have the processed one but that won't do

perhaps you could just load it without textures?

make it grayscale or smth

[hazy steppe]

i read "ye jaker has the brain"

[acoustic bobcat]

🇩🇪 pilled

[sweet nimbus]

well I do have the raw heightmap for the terrain if you want to turn that into a mesh in blender

I'm uploading the raw heightmap and the ktx compressed gltf asset to the shared gdrive

ok they're up

[vestal snow]

Thanks sounds good, that'll be nice to have

[sweet nimbus]

it's not too difficult to turn the heightmap into a mesh in blender. I looked up a tutorial on using a texture to influence vertex displacement, then applied it to a subdivided plane mesh

[acoustic bobcat]

Plane mesh > subdivision (regular mode, not the one that rounds things) > displace modifier > image texture for displacement

[sweet nimbus]

you should also add a skirt to the mesh

[mystic lark]

That is also the truth

[sweet nimbus]

he forgot "worm" at the end of that sentence

[mystic lark]

btw I added listing and three paragraphs to the drawing section but my brain is officialy fried now

I just put garbage words into sentences now lmao

I'll add bibtex file and fill it with references and that will be it for me today

[sweet nimbus]

thank you brotha

[mystic lark]

I think not much remains no?

the results and conclusion

[sweet nimbus]

and sampling

[mystic lark]

and the sampling part

yeah

[sweet nimbus]

sampling is trivial

well unless we talk about filtering

[mystic lark]

I really like the wrapping image Stephano did

really good job

[sweet nimbus]

hmm which one

what's it called

[mystic lark]

das

Wrap caching

.png

[sweet nimbus]

ah

beautiful pics

[mystic lark]

jaja

[sweet nimbus]

I love the color in all of these

the depth analysis diagram is kinda goth though 😳 hehe

[vestal snow]

oh yeah now that you mention it

it's like pure black and white lol

[sweet nimbus]

it's totally fine lol I was just commenting that it has less color than the others

[mystic lark]

I added all citations (except for the unreal one, I will need to add that one manually probably) but for some reason it is not linking these together, even though the syntax suggestions do give the correct ones

so I don't really know what I'm doing wrong

maybe we should ask the editors about this, it is really weird, as this always worked for me

I really need to eep now, I will be back tomorrow evening (in 15-16 hours or so)

[sweet nimbus]

alright

huh how do we actually use this bib file

do we need to include something in the main article

[mystic lark]

oh thats what I forgor hahaha

[sweet nimbus]

let me look at another article 🕵️

[mystic lark]

I just copied it

[sweet nimbus]

ah they just added a \bibliography thing at the end

[mystic lark]

yeah I just added that

still does not work

[sweet nimbus]

I see

[mystic lark]

okay there we go

[sweet nimbus]

I made some of it show up

[mystic lark]

citing works now

[sweet nimbus]

oh did you change something too lol

[mystic lark]

yeah we need the \bibilographystyle

[sweet nimbus]

lmao the diagram randomly in the bibliography

[mystic lark]

inorder for them to work

hahaha yeah we will need to fix that

also we have too little references

we need to add more

if you want to add them you can follow my style/ask copilot to generate the structure for you and just fill in the data

[sweet nimbus]

alright

[mystic lark]

or you can just find the paper on google scholar, click cite and then select bibtex on the bottom of the popup window

and just copy that

[sweet nimbus]

btw [!htb] doesn't work for the figure

[mystic lark]

yeah ! only works sometimes

you can yell at it but it won't listen

I suggest fixing this when we copy the article over to the official OL

they might have additional style settings and such, breaking everything...

okay 🛌 now, gn

[sweet nimbus]

🇬🇳

I got the thing to work. I had to use [H] (not [h]) and move the \usepackage{float} thing before the article start so it wouldn't error

does every reference need an inline citation? that would be a strange requirement probably

[mystic lark]

What do you mean by "every reference"?

[sweet nimbus]

everything in the bibliography

[acoustic bobcat]

Even the strongest enforcement of figure placement is at best a gentle suggestion lol

[acoustic bobcat]

Everything in the bib should be inline cited at least once

[sweet nimbus]

yeah it looks like latex just removes them from the bib if they aren't referenced

sad

[mystic lark]

Well it makes sense 😅 we can't have shadow references

[sweet nimbus]

lol I want them

[mystic lark]

We can include some of the clipmap papers, they also use the toroidal or 2D wrap mapping

We don't have to talk about clipmaps for thay

I can read the paper Wiemmer, I talked about it with him when I was at TU Wien so I'm actually pretty interested in it

[sweet nimbus]

alright

[mystic lark]

And then I'll add a reference to it in the introduction

[sweet nimbus]

I'll reference the UE5 thing too

[mystic lark]

You will probably need an @online or @misc reference

[sweet nimbus]

ye

[mystic lark]

We can also mention pcss and optimal bias papers in sampling

That's another like 2-3 references

[sweet nimbus]

yep

I think I'm good now. you can sleep 😄

[mystic lark]

I'll be back in 4 hours

Only on phone unfortunately

[sweet nimbus]

aight, sleep tight

btw I put our names in alphabetical order (not sure how else they are supposed to be put)

hmm is it fine to put two inline citations next to each other

I see it all the time on wikipedia

[vestal snow]

I'll try to do a full readthrough soon

what's the timeline like now? Like current/next steps

[sweet nimbus]

finish the draft today and get the main content in

it can be rough

then we keep refining it until the section editor takes a look

btw I literally cannot find ue5 docs for nanite

to cite

[vestal snow]

Ok makes sense

[sweet nimbus]

actually I can cite this perhaps
https://advances.realtimerendering.com/s2021/Karis_Nanite_SIGGRAPH_Advances_2021_final.pdf

[vestal snow]

Maybe lvstri has the links

[sweet nimbus]

not sure the proper way to cite a siggraph presentation, but I'll give it a shot

idk if this is even relevant
https://www.siggraph.org/preparing-your-content/citing-siggraph-publications/

[sweet nimbus]

hmm just learned that d3d11 has tiled resources

[sweet nimbus]

so previously (on discord) I used stable page addressing to refer to what our article now calls wraparound addressing

the article also describes stable page addressing as how we move the light matrices, but that isn't describing an addressing scheme

I think we should just drop the "stable page addressing" term in the article

[vestal snow]

Stable page addressing was fully replaced by wraparound addressing right? term wise at least

[sweet nimbus]

well not in the article

stable page addressing is just introduced and never mentioned again lol

but what it describes is how the light matrix position is quantized basically

[vestal snow]

oh lol

handwaved

projectile aliasing lol

autocorrect in overleaf is pretty funny

[sweet nimbus]

I fixed one of the listings by simply removing the newline between \begin{lstlisting} and its parameters

[vestal snow]

I have question(s) about the selection heuristic

in the second one are we using depth buffer value directly as d?

also the article is coming along really nicely

[sweet nimbus]

I think we still unproject it to get the world space distance

tbh I think both methods could be simplified to ultimately not need any unprojection or funny math, but what we have now works 🤷

[vestal snow]

ok so is it like

depth -> world coord, d = dist(world coord, camera location)?

[sweet nimbus]

uhm I haven't implemented it myself

[vestal snow]

no problem

During the readthrough I think that was the main part where I realized I don't follow what it's saying

[sweet nimbus]

add a comment if you haven't

[vestal snow]

Oh and the stable page addressing part. I see what you mean where it's briefly mentioned then we forget about it lol

alright I'll add one

[sweet nimbus]

yeah it's just "btw we need to move our matrices in page-size increments or caching won't work". we don't need a special term for that

[mystic lark]

I think in papers, usually the first one is the main author followed by arbitrary order

[mystic lark]

It is yeah

[sweet nimbus]

good morning sleepyhead

[mystic lark]

Yeah you take the distance of the fragment from the camera

[mystic lark]

Halloo

[acoustic bobcat]

Author order is usually:
Most contribution, 2nd, 3rd, ...Nth, principal investigator

First author is the one who did the most work or was responsible for the project, last author is the "most important" by seniority and organized/funded the student, and in between is just by decreasing order of contribution

There's no PI here really so you can either order yourselves by contribution or just say fuckit and make it alphabetical or something

[mystic lark]

Merging them is also fine, I was really struggling with explaining that so I tried to split it into smaller parts

[vestal snow]

thanks

For the cascades are they all at the same origin or do they kind of float a bit since world page sizes differ?

[mystic lark]

They float a bit, like you said

[acoustic bobcat]

Is there any particular good time for me to take another pass over the article

Otherwise I'll do it in 30 min or so before I go to bed

[sweet nimbus]

hmm maybe tomorrow if the section editor hasn't gotten to it first

when was the last time you looked?

[acoustic bobcat]

This morning, you saw all my comments already I think

[sweet nimbus]

alright

I don't think the article has changed too drastically since you last looked. I think it will be worthy of review once the last couple sections are filled in

[acoustic bobcat]

alright

I can try to look before work tomorrow

[sweet nimbus]

I wonder if they will notice if screenshots come from different renderers

[wooden jolt]

it's fiiiine

there is a lot of green and yellow btw

[humble shadow]

You don't get it, it's an advantage: proof this method works in multiple engines/implementations

[sweet nimbus]

hmm I have a cursed problem where I get self-shadowing, but only on concave triangle edges 😩

[hazy steppe]

those are not NaNs?

[wooden jolt]

could be

[sweet nimbus]

hmm

let me see

[wooden jolt]

btw jaker what do you mean by perspective aliasing

I know what projective aliasing is

[sweet nimbus]

camera goes close and you can see individual shadow texels

it certainly seems as though the black pixels are caused by the shadow

[wooden jolt]

it should be explained I fink

[sweet nimbus]

it is

[wooden jolt]

uh

oh yeah, I'm blind

ok so time to do some writing

[sweet nimbus]

can you do the sampling section? my brain died

[wooden jolt]

ye I saw the sampling section is very lonely

[sweet nimbus]

and/or review the big green section saky wrote just before that

while I take screenshots

[sweet nimbus]

fortunately fsr2 smooths over this

[wooden jolt]

AA my beloved

does it happen if you get close as well?

[sweet nimbus]

yeah but since it's on triangle edges it becomes less noticeable

btw does this look like an interesting enough scene (bistro is in the center)

[hazy steppe]

High Fov player showing off FPS scene in bistro

[sweet nimbus]

the plan is to somehow show that it works in far and near detail

[hazy steppe]

this is super cool

perchance put a smol circle around bistor

for the actual blind people

[sweet nimbus]

here is bistor

[wooden jolt]

iirc the terrain doesn't have many tringles does it, but it's fine

it looks good, that's what matters