Vkguide 2 thread | Graphics Programming | Page 3

hallow berry Dec 27, 2023, 11:11 AM

#

"Old VkGuide"

#

next to "VkGuide"

kind ferry Dec 27, 2023, 11:21 AM

#

i dont think i can do that without breaking the hyperlinks

#

not sure

hallow berry Dec 27, 2023, 11:32 AM

#

ah

#

i forgot about that

#

well 🙂

#

keep all the links

#

they dont have to appear anywhere on the site, but people can navigate to them, and when you open an old link

#

you will just put a gif showing some cool fitting meme and a link to vkguide2 home

#

like indiana jones finding the holy grail or something

#

also a little text explaining that the old stuff is now history and that you worked on v2

#

im partially serious and partially jk

kind ferry Dec 27, 2023, 12:05 PM

#

#

so what im thinking is to have the 2 versions under their own tag, vkguid 1 and 2. Then remove the "chapter" sections

#

no more than 1 level of nesting

#

alternatively just replace the chapters and tag the older ones as (OLD) 0. initial setup

hallow berry Dec 27, 2023, 12:19 PM

#

can you reshuffle the menu items without breaking links?

#

so that the new stuff appears as if its vkg1

#

and the old one gets an entry like Vkguide 2 now, in that screensho

ancient portal Dec 27, 2023, 12:24 PM

#

Another way is to do URL redirection

kind ferry Dec 27, 2023, 12:26 PM

#

#

good attempt

hallow berry Dec 27, 2023, 12:27 PM

#

url redirection might require access to the webserver

#

hmm i thought vkguide.dev was an url pointing to github.io 🙂

kind ferry Dec 27, 2023, 12:29 PM

#

it kinda is

#

its hosted through github

#

but it uses custom domain

#

im not retarded enough to have an important site on a github.io without its own domain

kind ferry Dec 27, 2023, 12:45 PM

#

#

its fixed

#

it sucks :/

#

also i need to update the introduction text to comment that you should go with the older version if you cant run the 1.3 stuff

ancient portal Dec 27, 2023, 12:46 PM

#

So the menu is autogenerated by url structure and there is no way to change that?

kind ferry Dec 27, 2023, 12:46 PM

#

kinda

#

its autogenerated from folder structure of the files

#

i can actually fix the URL of a given file

#

but it seems having 2 levels is a bit wonky

#

probably a code issue with the system

#

i can stuff the (OLD) part into essentially how it worked with the vkguide 2 beta where its embedded into another chapter and it is double nested for the articles

kind ferry Dec 27, 2023, 1:31 PM

#

ok it seems i can fix the navigation

#

by setting up grand-parent links correctly

#

damn jekyll is straight up stupid

#

you need to set the grand-parent links manually

#

it cant figure out parent of a parent by itself

kind ferry Dec 27, 2023, 2:00 PM

#

ok ive updated the site with a bunch of legacy warnings and legacy vkguide chapter and so on

#

how do you think its now?

hallow berry Dec 27, 2023, 2:37 PM

#

👁️ 👃 👁️

#

looks much better now

#

is it possible however, to rename "new_" to something "v2" perhaps"

#

because when next vulkan thing comes out, and warrants a renewal. you got old, new and then new new? 🙂

#

and one more thing but thats just me...

#

when you open each chapter you get this nice indented list of topics

#

#

is it possible to avoid word wrap in 4)

#

that Textures and Engine Architecture kind of ruins the main level

#

or is it possible to indent "Architecture" to be on the same level as the rest of the words

#

also also, dear imgui and multi threading from Extra chapter should be part of the main tutorial

#

the latter perhaps even combined with texture or model loading or a successor of it

kind ferry Dec 27, 2023, 2:45 PM

#

dear imgui is part of the tutorial

#

chapter 3

#

sorry chapter 2

#

it does imgui before drawing hello-triangle

#

multithreading isnt

#

because its just not needed, and that article is more of an overview, not a tutorial

hallow berry Dec 27, 2023, 2:47 PM

#

ah

#

then perhaps the dear imgui in Extra is not needed anymore

kind ferry Dec 27, 2023, 2:47 PM

#

#

i leave it there because its still a useful stand-alone article

hallow berry Dec 27, 2023, 2:47 PM

#

man i cant read

#

and links out there point to it perhaps

kind ferry Dec 27, 2023, 2:48 PM

#

yes

#

i have not renamed any hyperlinked thing, its all on the same place

#

just some of the extra chapters (and the first gpu driven rendering) mention that its for vkguide 1

hallow berry Dec 27, 2023, 2:50 PM

#

#

hehe

kind ferry Dec 27, 2023, 2:50 PM

#

janky ass code

hallow berry Dec 27, 2023, 2:50 PM

#

i clicked through from the top (from inside the articles)

#

and noticed that the legacy chapters open up too in the menu

kind ferry Dec 27, 2023, 2:50 PM

#

nah thats just jank code on the generator

#

it compares the stuff by name

hallow berry Dec 27, 2023, 2:50 PM

#

fook

#

did i tell you that i can reference code in files in mkdocs too?

kind ferry Dec 27, 2023, 2:51 PM

#

seems to be because the chapter name is the same

#

0: initial setup

hallow berry Dec 27, 2023, 2:51 PM

#

yeah

kind ferry Dec 27, 2023, 2:51 PM

#

on both

hallow berry Dec 27, 2023, 2:51 PM

#

if you use the new_ prefix for the main articles, perhaps use it for all subarticles too

#

links are also all the same

kind ferry Dec 27, 2023, 2:52 PM

#

pushed a fix to the 0: initial setup issue

#

@hallow berry nah i dont mind having the links be like that on the articles

#

the individual articles arent using the permanent URL that fixes it

#

so they come from folder

hallow berry Dec 27, 2023, 2:56 PM

#

is it just me or does ctrl+f5 not work in ff anymore

#

the fix you pushed, also pushed the next links underneath the comment section

#

kind ferry Dec 27, 2023, 2:58 PM

#

wtf its on all chapters

#

more jank this is supposed o be setup correctly

hallow berry Dec 27, 2023, 2:58 PM

#

hihi

kind ferry Dec 27, 2023, 2:59 PM

#

ah i see why

#

its been like that forever

#

#

the comment system is plugged in with a macro but the codegen generates the child articles below the normal content as a footer

hallow berry Dec 27, 2023, 3:01 PM

#

🇼 🇹🇫

#

ok i probably spoke to soon, i have no idea how weird it would be if i integrated that comment chat thing into my thing

#

but all that logic there looks weird

kind ferry Dec 27, 2023, 3:02 PM

#

i can probably do some hackery to force it down

#

i can also just not use the automated footer thing

#

and write it manually

#

which works fine too

#

what do you think of forcing the child list to be on top?

#

its an easy fix there and makes sense, gonnat ry it

#

#

insanely jank

#

gonna check in jekyll docs if i can automate the comment thing as a page property

#

and embed them in the correct place

#

oh thats actually pretty easy

hallow berry Dec 27, 2023, 3:12 PM

#

kind ferry

im ok with that

kind ferry Dec 27, 2023, 3:12 PM

#

ive modified the layout system

hallow berry Dec 27, 2023, 3:12 PM

#

let me check

kind ferry Dec 27, 2023, 3:12 PM

#

now you set the comment chapter in page properties

#

and it embeds them in the correct place

#

currently running ghpages refresh

hallow berry Dec 27, 2023, 3:13 PM

#

let me know when its done

kind ferry Dec 27, 2023, 3:21 PM

#

#

jekyll is going full retard

#

that one works

#

hallow berry Dec 27, 2023, 3:22 PM

#

ill look

kind ferry Dec 27, 2023, 3:23 PM

#

the styling engine is essentially creating the exact same string so why is it not working

hallow berry Dec 27, 2023, 3:23 PM

#

kind ferry Dec 27, 2023, 3:23 PM

#

wait what if im colliding on the name

hallow berry Dec 27, 2023, 3:24 PM

#

hmm i have another idea...

#

lets assume someone clicked on an old link... chapter 3 memory transfer

#

would it work if, when you opened it now, that it simply opens a new page, with a hint that chapter 3 memory transfer has now moved to legacy/3/memorytransfer

kind ferry Dec 27, 2023, 3:25 PM

#

the old links are all being kept

#

it would just point to the legacy one

hallow berry Dec 27, 2023, 3:25 PM

#

that means all old links can be prefixed ... ah ok

kind ferry Dec 27, 2023, 3:27 PM

#

ok fuck this im embedding the html

#

lets see now

hallow berry Dec 27, 2023, 3:29 PM

#

toc is below comments still

#

Building project looks somewhat correct for some reason

#

i also explicitly disabled cache, just in case

kind ferry Dec 27, 2023, 3:30 PM

#

#

fixed

#

gonna implement it across the tutorial

hallow berry Dec 27, 2023, 3:32 PM

#

while at it

#

can you make a little more space between the Next chapter links and the comments section

#

like a <p>

kind ferry Dec 27, 2023, 3:37 PM

#

done, but for the chapters with children

#

now everything has the new comments

hallow berry Dec 27, 2023, 3:39 PM

#

#

space looks good though : >

#

seems to be just that one, others seem to work

kind ferry Dec 27, 2023, 3:40 PM

#

parser broke FUCK

hallow berry Dec 27, 2023, 3:42 PM

#

yep but works

#

i went through all chapters and articles in the main quest

#

next links work there

#

its only in building_project

kind ferry Dec 27, 2023, 3:46 PM

#

ye its because it is the absolute last line and i had a by-one error

#

cpp kinda sucks at parsing

#

i should flip this parser and codegen system into rust

hallow berry Dec 27, 2023, 3:46 PM

#

lol

#

you should just use a framework which does all that nonsense already

kind ferry Dec 27, 2023, 3:46 PM

#

no

hallow berry Dec 27, 2023, 3:46 PM

#

against NIH? 😄

kind ferry Dec 27, 2023, 3:46 PM

#

the parser is like 300 locs

#

and it has the features i need to build the tutorial

hallow berry Dec 27, 2023, 3:46 PM

#

exactly

#

like what features?

#

reading code from underlying source files?

kind ferry Dec 27, 2023, 3:47 PM

#

copypasting tags from source code and setting up a few macros

hallow berry Dec 27, 2023, 3:47 PM

#

mkdocs has this syntax

#

```cmake title="lib/CMakeLists.txt"
--8<-- "lib/CMakeLists.txt:3:5"
```

#

cmake for the code hilighter.... title for the caption of the box... that 8< thing is the include trigger thingy... and then the path with :startlinenumber:linecount

#

example https://deccer.github.io/OpenGL-Getting-Started/01-basics/01-basic-window/

#

all the code schnippets are made with this thing

kind ferry Dec 27, 2023, 3:55 PM

#

that kinda sucks as it uses lines

#

i use text tags

hallow berry Dec 27, 2023, 3:56 PM

#

it sucks, when you modify code, then you have to update the lines ofc

#

and i wouldnt use it to docs my actual engine project because of that

kind ferry Dec 27, 2023, 3:57 PM

#

#

thats kinda how mine works

#

i dont support nesting

#

but i havent needed it

hallow berry Dec 27, 2023, 3:58 PM

#

ah it feels like we spokea bout that already some time ago now that i see that snippet again

kind ferry Dec 27, 2023, 4:00 PM

#

ye

#

im still using the same system

#

just fixing some bugs/etc

hallow berry Dec 27, 2023, 4:00 PM

#

ok then ignore what i said, and i need to get some gingko pills or so to help my brain

kind ferry Dec 27, 2023, 4:00 PM

#

in some cases in the code i use macros to have 2 versions of a given function so i can tag it in different ways for tutorial stages

#

not all the code in the guide is from tags, a few is copypasted manually or edited

hallow berry Dec 27, 2023, 4:01 PM

#

heh

low minnow Dec 28, 2023, 5:32 PM

#

chapter 5 vkloader.cpp still uses std::cerr instead of fmt::print

kind ferry Dec 28, 2023, 5:45 PM

#

ill fix

low minnow Dec 28, 2023, 5:49 PM

#

already done

#

see pr

kind ferry Dec 28, 2023, 5:58 PM

#

i see so just error function and the generate indices flag

#

the cmake things im not sure

#

but ill pick those small thingies in the loader

low minnow Dec 28, 2023, 6:27 PM

#

ive pushed a few other things

civic mantle Dec 28, 2023, 6:45 PM

#

The cmake things are to only copy runtime dlls on platforms where that makes sense - i'm not sure if the logic used to decide when to do that is 'the best' but it makes sense. Linux & macOS have rpaths in their linker making the need to copy dlls unnecessary

hallow berry Dec 28, 2023, 6:47 PM

#

there should be custom dll copy cmake isms in there already

civic mantle Dec 28, 2023, 6:51 PM

#

I mean that the cmake code to copy dll's is modified by the PR sean made so that it only happens on platforms where it makes sense - but i'm not 100% that the logic for 'which platforms need it' is the best. Except, I don't want to investigate cmake internals nor work through what the logic does exactly.

low minnow Dec 28, 2023, 6:53 PM

#

no the if I check is the one that gives exactly the information about this

hallow berry Dec 28, 2023, 6:54 PM

#

uh i totally didnt notice that thomas is writing too, i thought its all sean xD

low minnow Dec 28, 2023, 6:56 PM

#

see this: https://discourse.cmake.org/t/add-interface-library-but-i-want-to-copy-my-dll-in-the-same-cmake/9319/2

CMake Discourse

Add interface library but I want to copy my .dll in the same CMake

It looks like you’re trying to add a pre-compiled shared library. I suggest you use the IMPORTED keyword instead of INTERFACE. For example: add_library(Foo SHARED IMPORTED) set_target_properties(Foo PROPERTIES IMPORTED_LOCATION "${CMAKE_CURRENT_SOURCE_DIR}/bin/libfoo.dll" IMPORTED_IMPLIB "${CMAKE_CURRENT_SOURCE_DIR}/lib/libfoo.lib"/${CM...

civic mantle Dec 28, 2023, 7:05 PM

#

$<TARGET_RUNTIME_DLLS:chapter_0>
From the docs:
On non-DLL platforms, this expression always evaluates to an empty string.
https://cmake.org/cmake/help/latest/manual/cmake-generator-expressions.7.html#genex:TARGET_RUNTIME_DLLS

The generator expression is empty when there isn't anything to copy. Relying on the suffix used is weird and may break in situations where the compiler is on a platform where copying is required (windows) but is producing linux style binaries (like mingw). Ask me know I know about mingw shenanigans

low minnow Dec 28, 2023, 7:08 PM

#

yes but that is a generator expression so I can't use it in an if

#

and if that command is run that will be an empty string and then cmake -E copy will fail because theres no source specified

civic mantle Dec 28, 2023, 7:08 PM

#

~~I know~~ I see that now, I'm trying to see if there is a better way

low minnow Dec 28, 2023, 7:08 PM

#

so it will error while building

civic mantle Dec 28, 2023, 7:10 PM

#

COMMAND ${CMAKE_COMMAND} -E copy $<TARGET_FILE:exe> $<TARGET_RUNTIME_DLLS:exe> $<TARGET_FILE_DIR:exe>

Apparently this could work around it

#

In the meantime, here's my workaround: COMMAND ${CMAKE_COMMAND} -E copy $<TARGET_FILE:exe> $<TARGET_RUNTIME_DLLS:exe> $<TARGET_FILE_DIR:exe>. This also copies the target to its own directory. This is wasteful and silly but makes sure that if the *DLLS is an empty string it's still a valid command.

https://stackoverflow.com/questions/10671916/how-to-copy-dll-files-into-the-same-folder-as-the-executable-using-cmake

lapis isle Dec 28, 2023, 7:14 PM

#

why not do something like

if (NOT ("$<TARGET_RUNTIME_DLLS:exe>" STREQUAL ""))
  # copy the DLLs
endif()

civic mantle Dec 28, 2023, 7:14 PM

#

lapis isle why not do something like ```cmake if (NOT ("$<TARGET_RUNTIME_DLLS:exe>" STREQUA...

cmake

#

generator expressions run at 'configure time' which is after all the if statements are processed.
(in a nutshell, its more complicated than that)

#

@low minnow Why not just use if( WIN32)? For the most part, this project is simple usage and knowing what the hell CMAKE_IMPORT_LIBRARY_SUFFIX means is a big ask over what WIN32 means

low minnow Dec 28, 2023, 7:18 PM

#

no idea

#

i just found that post on the cmake forums and decided it looked good

kind ferry Dec 28, 2023, 7:20 PM

#

whats the error you get when compiling on mac?

civic mantle Dec 28, 2023, 7:20 PM

#

(this is why bad cmake self propagates - enough of it is written online that people find examples that do work but because it is an arcane incantation rather than sensible language the actually bad code is intermingled with just fine code)
Not trying to cast shade, I too reach for online examples and samples since there is way to much to learn to wrangle cmake.

kind ferry Dec 28, 2023, 7:20 PM

#

does the code even run on mac? being 1.3

civic mantle Dec 28, 2023, 7:21 PM

#

does it build on linux? The copy command probably fails in the exact same way. "didn't specify destination" or some error like that.

low minnow Dec 28, 2023, 7:21 PM

#

kind ferry does the code even run on mac? being 1.3

no as I mentioned in the PR

#

but I think this would be the same error on Linux

kind ferry Dec 28, 2023, 7:22 PM

#

does it actually error

low minnow Dec 28, 2023, 7:22 PM

#

because cmake -E copy with the runtime dlls fail when there are no runtime dlls

#

so the build fails

kind ferry Dec 28, 2023, 7:22 PM

#

target_runtime_dlls would be empty so no copy command

low minnow Dec 28, 2023, 7:22 PM

#

no the copy command triggers build failure

kind ferry Dec 28, 2023, 7:22 PM

#

hmm

#

i guess the if(win) is the best approach

low minnow Dec 28, 2023, 7:23 PM

#

shall I change it or remove that commit

civic mantle Dec 28, 2023, 7:23 PM

#

doesn't error out for me on linux

low minnow Dec 28, 2023, 7:23 PM

#

with the if or without?

civic mantle Dec 28, 2023, 7:24 PM

#

just on all-chapters branch - so no if statement

low minnow Dec 28, 2023, 7:24 PM

#

hmm

#

so then if(WIN32) wouldnt work because there are runtime dlls it needs to copy on linux?

civic mantle Dec 28, 2023, 7:25 PM

#

ignore me, the all-chapters doesn't do the dll copying

#

Yep, linux build fails with error due to copying (and some bad code things that I think the PR fixes as well)

ninja -C build
ninja: Entering directory `build'
[376/393] Linking CXX executable /home/cdgiessen/personal/temp-vulkan-guide-vblanco/bin/chapter_0
FAILED: /home/cdgiessen/personal/temp-vulkan-guide-vblanco/bin/chapter_0 
...
CMake Error: cmake version 3.27.4
Usage: /usr/bin/cmake -E <command> [arguments...]
Available commands:

#

Yeah so if if(CMAKE_IMPORT_LIBRARY_SUFFIX) was replaced with if(WIN32) I believe it'll do all the things we want and be a bit clearer about why.

Its not like this project is dealing with cross compilation, mingw, or any other weird build setup (unlike the Vulkan-Loader grrrrr)

kind ferry Dec 28, 2023, 7:30 PM

#

@civic mantle all-chapters-2

#

the other is for older vkguide

#

sorry

#

all-chapters-1.3-wip

low minnow Dec 28, 2023, 7:31 PM

#

civic mantle Yeah so if `if(CMAKE_IMPORT_LIBRARY_SUFFIX)` was replaced with `if(WIN32)` I bel...

ok ill change it

civic mantle Dec 28, 2023, 7:32 PM

#

so wait, which is the 'this is the vkguide 2.0 all-chapters' branch?

low minnow Dec 28, 2023, 7:32 PM

#

1.3-wip

kind ferry Dec 28, 2023, 7:32 PM

#

one sec ill upload all-chapters-2 to be clearer and easier to find

#

up it is

low minnow Dec 28, 2023, 7:34 PM

#

should I re-target the PR?

kind ferry Dec 28, 2023, 7:37 PM

#

i wont directly merge it anyway

#

need to do things like re-run codegen scripts and so on

#

so dont bother ill manually add it

kind ferry Dec 28, 2023, 8:34 PM

#

there is one part of vkguide 2 im not 100% sure with, which is the descriptors

#

they are under-explained compared to vkguide 1

#

of course there is just a lot less need for them

#

better abstractions and later it will just go bindless

#

but what do you think?

lapis isle Dec 28, 2023, 8:46 PM

#

what chapter do you go over descriptors in

#

https://vkguide.dev/docs/new_chapter_4/descriptor_abstractions/
I found this chapter but I can't find the part of chapter 2 you mention

kind ferry Dec 28, 2023, 8:47 PM

#

they are used and started in the 2 Vulkan Shaders articles in chapter 2

lapis isle Dec 28, 2023, 8:49 PM

#

#

this is like the minimum of info, and the - Code section seems to jump right into the descriptor pool implementation

#

I would want at least a small Descriptors subheader rather than just saying "In vulkan, shader IO is handled through descriptor sets"

kind ferry Dec 28, 2023, 8:54 PM

#

i have some bigger paragraphs of explanation i can copypasta from the older one

#

https://vkguide.dev/docs/chapter-4/descriptors/ the one here

Vulkan Guide

Descriptor Sets

Practical guide to vulkan graphics programming

#

but i kinda worry its just a really huge wall of text to throw to people chapter 2. on vkguide 1 this was on chapter 4

lapis isle Dec 28, 2023, 8:56 PM

#

yeah it's kinda tough to mitigate the frontloading of vulkan

#

but you're just doing stuff in a different order (which I prefer btw)

lapis isle Dec 28, 2023, 8:57 PM

#

lapis isle

my concern is that this is the main place to put the "theory" section, esp. since you have a separate section for the code

#

but I guess it's hard to describe what descriptors are when you don't let the user create buffers and images yet

kind ferry Dec 28, 2023, 8:57 PM

#

ye

#

i could copy that article and put it after the first compute shader draw

#

to kinda explain in detail what it does

#

or try to embd those paragraphs while building the abstractions

#

currently they do have parts of those but.. more

#

i think the one to add is probably the diagram and the first paragraph of the descriptor set mental model

lapis isle Dec 28, 2023, 8:58 PM

#

or alternatively, shorten the section on pipeline layout, cut out any details about descriptor sets

#

and leave that all to later

kind ferry Dec 28, 2023, 9:01 PM

#

thats kinda critical tho

#

ill think of it

#

i think doing the descriptor sets with the small abstractions from the start make them so much clearer

#

everyone gets ultra confused with them

#

but if you have these builders and its like 'yeh, build layout, use layout to create descriptor set, write it' its much clearer

lapis isle Dec 28, 2023, 9:04 PM

#

maybe just bringing the diagram to the shader theory chapter is enough

#

and doing a short discussion on it

lapis isle Dec 28, 2023, 9:05 PM

#

kind ferry i have some bigger paragraphs of explanation i can copypasta from the older one

I'm actually liking this idea a lot now that I reread it

kind ferry Dec 28, 2023, 9:09 PM

#

i could also copy the entire chapter before the second part of descriptor stuff

#

which is later in the tutorial

#

where the allocator v2 (this time it reallocs pools) and descriptor-writer is implemented

kind ferry Dec 28, 2023, 10:05 PM

#

ive updated that article a bit

#

moved some of the wording from the older descriptor article into it

#

but small amount

#

https://vkguide.dev/docs/new_chapter_2/vulkan_shader_drawing/

Vulkan Guide

Vulkan Shaders

Practical guide to vulkan graphics programming

lapis isle Dec 28, 2023, 10:30 PM

#

I like it a lot

#

I think it makes everything fit together more

kind ferry Dec 29, 2023, 9:20 AM

#

is there a KTX lib/header to read them?

kind ferry Dec 29, 2023, 9:38 AM

#

im thinking to use it to load the filtered cubemaps from that gltf environments repo

#

acts as both fancy hdr image loading and mipmapped loading

outer sigil Dec 29, 2023, 9:39 AM

#

did anyone with no experience tried to go thru this guide? looks good so I might try 😄

lapis isle Dec 29, 2023, 10:34 AM

#

https://github.com/libktx/ktx
yeah there's khronos' official one

kind ferry Dec 29, 2023, 10:47 AM

#

#

you kinda got mistaken by the naming there

lapis isle Dec 29, 2023, 10:55 AM

#

lmao rip

#

https://github.com/KhronosGroup/KTX-Software

#

this is the one

hallow berry Dec 29, 2023, 11:15 AM

#

using libktx is just ~10loc

kind ferry Dec 29, 2023, 11:34 AM

#

oh so it can also bake textures to ktx

#

with mipmap gen and everything

#

and with a direct upload-to-vulkan in the lib 🤔

#

pretty hax thats awesome

hallow berry Dec 29, 2023, 11:43 AM

#

ive only fiddled with loading ktx2

#

more or less

tex = ktx2_loadfrommemory(imagebytesfromgltf)
if ktx2_encodingrequired(tex)
{ 
    ktx2_reencodeit(tex);
}
uploadtogl/vk(vulkanformattoglformat(tex.format), tex.data);
ktx2_free(tex);

low minnow Dec 30, 2023, 9:03 PM

#

@kind ferry you might want to consider switching to a static commit for fastgltf in vkguide

#

im about to change a bunch of breaking things

kind ferry Dec 30, 2023, 9:36 PM

#

do you have release tag?

#

also its embedded as code in the project. no submodules

low minnow Dec 30, 2023, 9:40 PM

#

yes v0.6.1 is the last

lusty hound Dec 31, 2023, 9:54 PM

#

so uh how followable is vkguide without vkb? gotten through half of the first chapter of vulkan-tutorial

lapis isle Dec 31, 2023, 9:58 PM

#

vulkan-tutorial will cover the missing gap of figuring out how to initialize vulkan

#

but honestly that's just your own time you're wasting, vulkan initialization is not interesting

lusty hound Dec 31, 2023, 10:04 PM

#

i see i'll probably switch over to it then since it's getting pretty hard ot follow vulkan-tutorial when i effectively gutted it out and replaced with dynamic rendering and sync2

hallow berry Dec 31, 2023, 10:06 PM

#

vkguide covers the initialization part

#

by using vbkootstrap

kind ferry Dec 31, 2023, 10:08 PM

#

bootstrap is used to find queue, create device, and create swapchain

#

thats the first few chapters of tutorial

lusty hound Dec 31, 2023, 10:12 PM

#

hmmhm well im doing this in rust and there isn't a vkb equivalent for now, but i gotten through the initalization of vulkan-tutorial so i think i should be fine for the rest of vkguide, no?

hallow berry Dec 31, 2023, 10:26 PM

#

yes

#

idk about rust, im sure it has if you look

civic mantle Dec 31, 2023, 10:42 PM

#

lusty hound hmmhm well im doing this in rust and there isn't a vkb equivalent for now, but i...

sounds like someone should poke me enough into writing a rust version of vkb...

lusty hound Dec 31, 2023, 10:43 PM

#

civic mantle sounds like someone should poke me enough into writing a rust version of vkb...

boy oh boy do i have a great idea for you

civic mantle Dec 31, 2023, 10:43 PM

#

😛

kind ferry Dec 31, 2023, 11:13 PM

#

@lusty hound definitely. once yoy get to hello triangle, you have all you need

lusty hound Dec 31, 2023, 11:13 PM

#

okokok thank you

kind ferry Dec 31, 2023, 11:15 PM

#

there is also a dude who did vkguide on Ig

#

zig

#

and had to implement the start. you can copy what he did

#

its one of the top posts in zig reddit from last month

lusty hound Jan 1, 2024, 5:17 AM

#

How necessary is going through the compute chapter of the tutorial is...? I know it's dumb KEKW but given I gotten to Hello, Triangle it does feel like taking a lot of steps back

hallow berry Jan 1, 2024, 10:03 AM

#

go through it

#

dont be that lazy copy pasta guy

kind ferry Jan 2, 2024, 12:10 PM

#

gonna add a page on descriptor indexing for current end of chapter 5. it will probably be moved around later, but for now a lot of people really want it to be explained

#

for descriptor indexing, do you use combined sampler-image? or separated

#

with current style where im using the combined ones, its a simpler code change to do that, but obviously if a texture is accessed in 2 ways it would take 2 slots. On the other hand i doubt thats going to be a common case

drifting maple Jan 2, 2024, 12:31 PM

#

I think combined image shrimplers are good for starting out but they pose serious usability problems

#

wanna switch out a sampler for a texture, well you gotta call vkUpdateDescriptorSets

#

maybe change LOD bias, another call to vkUpdateDescriptorSets

kind ferry Jan 2, 2024, 12:35 PM

#

im going to write the descriptor array every frame

#

once

#

im thinking of putting it on its own descriptor set so that i have this generic "all textures" set

#

to use across the entire engine

#

on novuscore we do that and its fine, i think it costs like 0.something ms to update, which is a fine per-frame cost to pay

drifting maple Jan 2, 2024, 12:41 PM

#

descriptor buffer can't come any sooner

kind ferry Jan 2, 2024, 12:44 PM

#

does renderdoc already support them?

crimson timber Jan 2, 2024, 1:25 PM

#

kind ferry im thinking of putting it on its own descriptor set so that i have this generic ...

yes

#

i dont like that you upload the whole thing every frame. That partly negates the whole point of bindless

kind ferry Jan 2, 2024, 1:26 PM

#

its perfectly fine, i have some sort of texture manager, and it reallocs every frame. Of course, could always cache it between frames

crimson timber Jan 2, 2024, 1:27 PM

#

why do you not simply update only changing slots inside the array

kind ferry Jan 2, 2024, 1:27 PM

#

we do that in novuscore and 0 issues, as long as its really once per frame

#

couse i remember it was like 0.2 or something ms

crimson timber Jan 2, 2024, 1:27 PM

#

why waste 0.2ms

kind ferry Jan 2, 2024, 1:27 PM

#

because its a tutorial and doing proper caching complicates things

crimson timber Jan 2, 2024, 1:28 PM

#

hmm

kind ferry Jan 2, 2024, 1:28 PM

#

and 0.2 cost to enable bindless textures is.. perfectly fine

#

this would be done exactly once through this texture manager object thingy

crimson timber Jan 2, 2024, 1:28 PM

#

im sceptical that this is simpler

kind ferry Jan 2, 2024, 1:28 PM

#

keep in mind if you are doing texture streaming you would be modifying it almost every frame

#

resizes not that often, but updates sure

crimson timber Jan 2, 2024, 1:29 PM

#

it doesnt matter

#

simply send the slot to be reused on destruction

#

they have the same lifetime

#

the slot and the resource

kind ferry Jan 2, 2024, 1:30 PM

#

yeye, im thinking of having the texture manager be some sort of slotmap with stable indexing

#

so it keeps holes and reuses them

crimson timber Jan 2, 2024, 1:30 PM

#

yes

#

but why copy every frame then

drifting maple Jan 2, 2024, 1:30 PM

#

to be fair, since potrick showed me the shader resource table™️ (a stupid simple free list allocator of descriptors KEKW ), I agree that any other method of managing descriptor sets feels weird

kind ferry Jan 2, 2024, 1:30 PM

#

🤔

#

i can see your point

#

so just init the descriptor array as fucking-huge

#

and then do the individual writes to it?

drifting maple Jan 2, 2024, 1:31 PM

#

ye

crimson timber Jan 2, 2024, 1:31 PM

#

yea babyy

kind ferry Jan 2, 2024, 1:31 PM

#

i would still need 2 due to frames on flight

drifting maple Jan 2, 2024, 1:31 PM

#

no

crimson timber Jan 2, 2024, 1:31 PM

#

no

drifting maple Jan 2, 2024, 1:31 PM

#

you can update unused slots and defer destruction of existing resources

kind ferry Jan 2, 2024, 1:31 PM

#

oh

#

was thinking it wouldnt be allowed to modify that descriptor array while its used

#

ok that makes sense

#

what about the question of combined sampler vs separate?

drifting maple Jan 2, 2024, 1:32 PM

#

I like separate, they just make more sense, but perhaps combined samplers are better for beginners

kind ferry Jan 2, 2024, 1:33 PM

#

combined sampler has the bonus that it maps directly to what i already have

crimson timber Jan 2, 2024, 1:33 PM

#

i cant answer that i only use separate

kind ferry Jan 2, 2024, 1:33 PM

#

but i remember the hardware vendors didnt like it much

drifting maple Jan 2, 2024, 1:33 PM

#

NV still recommends combined

crimson timber Jan 2, 2024, 1:33 PM

#

but only nv

drifting maple Jan 2, 2024, 1:33 PM

#

(for some reason)

crimson timber Jan 2, 2024, 1:35 PM

#

they can pack a texture and sampler pointer into one 32bit int

#

and use that for sampling directly

#

so a combined sampler for them is just that already packed for use

kind ferry Jan 2, 2024, 1:36 PM

#

combined it is then

#

i find it much easier to think about

crimson timber Jan 2, 2024, 1:36 PM

#

meh

#

imo combined is a complication

#

https://github.com/Ipotrick/Daxa/blob/78811a374bdce71d4b4a02e10a6549278649931f/src/impl_gpu_resources.cpp#L37

GitHub

Daxa/src/impl_gpu_resources.cpp at 78811a374bdce71d4b4a02e10a654927...

Daxa is a convenient, simple and modern gpu abstraction built on vulkan - Ipotrick/Daxa

#

thats how i do it (if you care 💅). Should have everything desceiptor related inside

drifting maple Jan 2, 2024, 1:40 PM

#

Roast my implementation as well
https://github.com/LVSTRI/Retina/blob/master/src/Retina/Graphics/Resources/ShaderResourceTable.cpp
https://github.com/LVSTRI/Retina/blob/master/include/Retina/Graphics/Resources/ShaderResourceTable.hpp

lapis isle Jan 2, 2024, 2:08 PM

#

yeah I use separate, also I am currently trashing my current implementation because it sucks in several ways:

relies on robustness ext for null descriptors
has frames-in-flight descriptor sets and writing logic built around that
is built on top of DescriptorAllocator abstractions instead of just owning its own precisely sized descriptor pool
has shit for storage buffer indexing (now useless since I use BDA)

#

is the most relevant point for vkguide imo because I built my descriptor indexing on top of the abstractions I got from vkguide without really thinking about whether it made sense

#

but all 4 are real mistakes I made from just going off on my own, so they might be worth covering

lusty hound Jan 4, 2024, 7:17 PM

#

uh I'm having trouble understanding wtf PoolSizeRatio is doing

#

why would we need a ratio larger than 1...?

kind ferry Jan 4, 2024, 7:55 PM

#

its for the case where 1 set will have multiple descriptors

#

imagine a layout where its 4 images

#

1 set has 4 images

#

so thats a ratio 4 on COMBINED_IMAGE_SAMPLER

lusty hound Jan 4, 2024, 7:58 PM

#

ohhh

#

alright

#

thank you

hallow berry Jan 6, 2024, 1:59 PM

#

@kind ferry are you resetting the comments? you mentioned you would, somewhere

#

now that vkguide2 is the default

#

there are also some unsigned int vs uint32_t inconsistencies, also only noticed that now

#

init_commands ... for (int i = 0; i < FRAME_OVERLAP... i should also be uint32_t, since FRAME_OVERLAP is a unsigned int ... warnings suck

hallow berry Jan 6, 2024, 3:19 PM

#

in Draw Loop

#

this sentence doesnt seem to fit

#

The timeout of the WaitFences call is of 1 second. It’s using nanoseconds for the wait time. If you call the function with 0 as the timeout, you can use it to know if the GPU is still executing the command or not.

#

no beginner will understand what that means, and you also dont explain it any further

#

Id remove it

#

Or pick it up in some appendix again where "you can use it to know if gpu is still executing ...or not" is explained ideally with an example usecase

#

get_current_frame() is not defined anywhere

#

i know its just a return _frames[_frameNumber % FRAME_OVERLAP]; but still

#

im also not a fan of naming commandBuffer cmd, cmd sounds like just command

#

i assume its for the sake of keeping the code visible in that little code box... although other code boxes are not wrapping any code either

#

anyway thats just me nitpicking

#

This is optional, but we might get a small speedup from our command encoding if we can tell the drivers that this buffer will only be submitted and executed once. We are only doing 1 submit per frame before the command buffer is reset, so this is perfectly good for us.
``` this should probably be picked up later again as well

#

where you have to deal with other scenarios... ill just leave it here, maybe this will be handled further down the road : )

#

This means we are going to need a way to transition images as part of a command buffer instead of using a renderpass, so we are going to add it as a function on vk_images.h
``` casually thrown in "instead of using a renderpass"

#

that probably needs to be explained further

#

typo: We will be doing a pipeline barrier, using the syncronization 2 feature/extension which synchronization

hallow berry Jan 6, 2024, 4:11 PM

#

all fields in FrameData are not initialized : (

kind ferry Jan 6, 2024, 4:21 PM

#

but they are 🤔 i manually went through the entire tutorial copypasting the code snippets as it worked

#

might be shown a bit later

hallow berry Jan 6, 2024, 4:26 PM

#

im also stumbling over that waitfence 1s timeout thing again, i dont remember what i did in the old version, my vkguide1 code also just showes 1s

kind ferry Jan 6, 2024, 4:28 PM

#

that code is unchanged from vkguide 1

#

a lot of the first chapter is copypasta-d

#

where it goes different is on chapter 2 forward where its an almost total rewrite

hallow berry Jan 6, 2024, 4:29 PM

#

thats what im doing too : )

#

just removed all vk bits from my main branch, and follow vguide2

hallow berry Jan 6, 2024, 4:30 PM

#

kind ferry but they are 🤔 i manually went through the entire tutorial copypasting the cod...

not in the docs

kind ferry Jan 6, 2024, 4:32 PM

#

what do you mean with in the docs

hallow berry Jan 6, 2024, 4:34 PM

#

in the guide, sorry

#

and please dont do int a,b,c; 😦

kind ferry Jan 6, 2024, 4:35 PM

#

but that stuff is initialized right after when it builds the init_sync_structures

hallow berry Jan 6, 2024, 4:35 PM

#

c++ good practice should apply here too

#

tutorial or not

#

noobs copy paste stuff blindly

kind ferry Jan 6, 2024, 4:36 PM

#

nope

#

its a tutorial it needs to be almost codegolfed

hallow berry Jan 6, 2024, 4:37 PM

#

noooooooooooooooo

kind ferry Jan 6, 2024, 4:37 PM

#

there is no need to waste space on initializers when the code is already hardcoded to init it

hallow berry Jan 6, 2024, 4:37 PM

#

not cool

#

noobs will forget to call InitializeXXX

#

and then those objects have random values instead of 0, pointing them right at "you forgot to call InitializeXXX"

#

also again these underscores, which indicate private visibilitlity, when they are not really?

#

anyway

#

feels like im talking to a block of wood or so

kind ferry Jan 6, 2024, 4:47 PM

#

the underscores will be fixed

#

they were supposed to be for private stuff but stuff changed midway wich is a bit strange

#

btw regarding the beta comments, they will get nuked probably next week

hallow berry Jan 6, 2024, 7:55 PM

#

i have a bit more nitpickingisms

#

in the improving renderloop chapter

#

creating all the infos for the drawimage

#

you call those a little weirdly.. rimg_xxxx

#

why not drawImageCreateInfo and drawImageAllocatorInfo and drawImageViewCreateInfo

#

since they create things for he drawImage : )

#

perhaps it was meant to be called renderImage hence rimg... but idk

round jungle Jan 6, 2024, 7:58 PM

#

hallow berry and please dont do int a,b,c; 😦

What's wrong with that style of code?

hallow berry Jan 6, 2024, 7:59 PM

#

round jungle What's wrong with that style of code?

error prone

#

and picture adding *

round jungle Jan 6, 2024, 8:01 PM

#

Well the * goes on the variable not the data type, but I can see how noobs can be confused

hallow berry Jan 6, 2024, 8:02 PM

#

* goes on the type not on the variable

drifting maple Jan 6, 2024, 8:02 PM

#

round jungle Well the * goes on the variable not the data type, but I can see how noobs can b...

you are so wrong

hallow berry Jan 6, 2024, 8:02 PM

#

but anyway : )

#

we have #bikeshed-😇 for those kind of things

#

there is enough bad and unreadable code out there already

round jungle Jan 6, 2024, 8:08 PM

#

okay just need a yes or no, don't want to #bikeshed-😇, am I wrong on code-style or how pointer declaration works

hallow berry Jan 6, 2024, 8:11 PM

#

theres 3 styles, or 3 types of people

#

1) type* var;
2) type * var;
3) type *var;

#

imho 1) is the one and only true way of declaring a pointer type

hallow berry Jan 6, 2024, 8:31 PM

#

more foggy stuff imho

#

in New draw loop

Vulkan has 2 main ways of copying one image to another. you can use VkCmdCopyImage or VkCmdBlitImage. CopyImage is faster, but its much more restricted, for example the resolution on both images must match. Meanwhile, blit image lets you copy images of different formats and different sizes into one another. You have a source rectangle and a target rectangle, and the system copies it into its position. Those two functions are useful when setting up the engine, but later its best to ignore them and write your own version that can do extra logic on a fullscreen fragment shader.

#

the comment about "but later its best to ignore them and write your own version that can do extra logic on a fullscreen fragment shader" feels out of place

#

what does that mean? or how is THAT particular thing implemented... perhaps yet again another appendix

#

hmm more stuff.

#

you added a deletion queue into FrameData, and just mention where to call it, but you never push anything into it anywhere

#

just a tiny sentence saying that we should push into it, whenever we create something. Idk perhaps you can put it in one of those /!\ ATTENTION /!\ boxes so that people dont read over it

kind ferry Jan 6, 2024, 8:51 PM

#

the deletion queue stuff is for later

#

its mostly because i implement the deletion queue so might as well also add it to framedata early

#

you are right with the copyimage paragraph there, could use some rewording

#

how to implement that thing is something i will do on chapter 6. the compute shader that copies to swapchain is the one that will do exposure/tonemapping

hallow berry Jan 6, 2024, 8:54 PM

#

the deletionqueue.flush on the framedata is also kaputt... or im too stupid to read

kind ferry Jan 6, 2024, 8:54 PM

#

not really? its actually used later for allocating per-frame uniforms

hallow berry Jan 6, 2024, 8:54 PM

#

you are flusing it in Draw() after vkWaitForFences

#

but there is no code yet which recreates those

#

in new renderloop

kind ferry Jan 6, 2024, 8:55 PM

#

what do you mean re-create

hallow berry Jan 6, 2024, 8:55 PM

#

when you flush the deletion queue it invokes vkDestroyXXX

kind ferry Jan 6, 2024, 8:55 PM

#

ye

#

but the frame fences arent put into the per-frame destruction queue

#

the only thing that is added to the per-frame destruction queue is some camera uniform buffer later

hallow berry Jan 6, 2024, 8:56 PM

#

ah i think that one sentence tricked me then

#

With the deletion queue set, now whenever we create new vulkan objects we can just add them into the queue.
``` this one, i spotted it after rereading this and the section before

#

and assumed i forgot something

#

ok, im just too dumb to read properly

kind ferry Jan 6, 2024, 8:57 PM

#

its for stuff like more draw image, descriptor pools, pipelines, etc

hallow berry Jan 6, 2024, 8:57 PM

#

yeah

#

that makes sense now

kind ferry Jan 6, 2024, 8:57 PM

#

to stop adding everything into the cleanup() function

hallow berry Jan 6, 2024, 8:57 PM

#

well

#

vk is verbose

#

i dont mind having a 50loc cleanup function 😄

kind ferry Jan 6, 2024, 9:04 PM

#

issue is that it would mean the entire tutorial keeps coming back to the cleanup function

#

deletion queue and push is just waaaaaaay easier

hallow berry Jan 6, 2024, 9:05 PM

#

deletionQueue makes sense

#

i guess the per frame stuff to be created/deleted is something i need to wrap my head around still

#

it feels wrong to create and destroy stuff - per frame - same for opengl where you can do that s well and its fine

kind ferry Jan 6, 2024, 9:13 PM

#

every engine ever will need to create and destroy stuff dynamically perframe

hallow berry Jan 6, 2024, 9:23 PM

#

and that has no impact?

#

compared to just update an ubo with camera data for instance

kind ferry Jan 6, 2024, 9:49 PM

#

of course it has an impact, but its much easier to code

#

and the impact is like a microsecond or two so its not like it matters

lapis isle Jan 7, 2024, 12:18 AM

#

btw what made you opt to have fences instead of timeline semaphores after all

kind ferry Jan 7, 2024, 9:22 AM

#

cant use timeline on swapchin

#

would need to explain 2 types and use only 1 (the old)

lusty hound Jan 7, 2024, 5:48 PM

#

Uh how devastating would it be if I skipped the dear imgui part?
Rn egui or imgui-rs integration is either non existent or requires me to rewrite the entire tutorial bleakekw

kind ferry Jan 7, 2024, 6:08 PM

#

its barely used

lusty hound Jan 7, 2024, 7:08 PM

#

okok thank you

lusty hound Jan 8, 2024, 2:48 AM

#

Hey uh question. When we run draw_background and draw_triangle, how do we ensure the correct order of drawing occurred?
As I;m just confused how we ensured the triangle draws overtop the draw image produced by compute

#

iirc vkSubmit only ensures order of submission but never order of completion?

fringe berry Jan 8, 2024, 2:55 AM

#

lusty hound Hey uh question. When we run draw_background and draw_triangle, how do we ensure...

The layout transition in between the two ensures that all writes done by the compute shader must be done before the fragment shader can do any writes

lusty hound Jan 8, 2024, 3:02 AM

#

AHHHH WAIT YEAH

#

Image transitions

#

Thank you

#

So the rasterzzer waits until the draw image transitions from general to color attachment i guess?

fringe berry Jan 8, 2024, 3:09 AM

#

well because vkguide has it simplified, everything waits until the image transition is done. even the vertex shader won't run until the transition is done

#

if you set it up optimally, the vertex shader, the rasterizer, and even the fragment shader can run while the transition happens; they just can't write to the image until the compute is done

lusty hound Jan 8, 2024, 3:17 AM

#

Huhh okay okay

junior sinew Jan 8, 2024, 8:47 AM

#

https://vkguide.dev/docs/new_chapter_1/vulkan_mainloop_code/

VkSemaphoreSubmitInfo vkinit::semaphore_submit_info(VkPipelineStageFlags2 stageMask, VkSemaphore semaphore)
{
    VkSemaphoreSubmitInfo submitInfo{};
    submitInfo.sType = VK_STRUCTURE_TYPE_SEMAPHORE_SUBMIT_INFO;
    submitInfo.pNext = nullptr;
    submitInfo.semaphore = semaphore;
    submitInfo.stageMask = stageMask;
    submitInfo.deviceIndex = 0;
    submitInfo.value = 1;

    return submitInfo;
}

https://registry.khronos.org/vulkan/specs/1.3-extensions/man/html/VkSemaphoreSubmitInfo.html

value is either the value used to signal semaphore or the value waited on by semaphore, if semaphore is a timeline semaphore. Otherwise it is ignored.

why is value being set? it doesn't look like you are using timeline semaphores

#

1 seems incorrect to use even if you were using them

kind ferry Jan 8, 2024, 12:17 PM

#

hm i actually dont know why its 1 there

#

the inits have been copypasted for a very long time so i dunno when that got added

hallow berry Jan 9, 2024, 10:26 PM

#

hmm

#

vkg2 half way in still feels like vkg1

#

hmm

#

why is dynamic rendering not mentioned in the navigation thing

#

its casually sprinkled into the imgui chapter

#

this deserves its own major chapter

#

another typo:

#

in the imgui chapter, first sub chapter "Immediate GPU commands" above the 2nd code box

we need to create those syncronization structures for immediate submit, so lets go into init_commands() function and hook the command part. typo in synchronization

junior sinew Jan 9, 2024, 11:19 PM

#

also "let's"

lusty hound Jan 10, 2024, 4:07 AM

#

hallow berry why is dynamic rendering not mentioned in the navigation thing

Yeah it does KEKW

#

It gets brought way too casually then slides into implementation

kind ferry Jan 10, 2024, 9:14 AM

#

it then gets implemented again in the rendering chapter

#

there also isnt much to explain, its kinda beyond trivial. The hard part is getting the images on their correct layouts for drawing, but beggining a renderpass is a very simple thing of just grabbing some render targets and doing begin rendering

#

renderpasses used to have their whole long explanation due to al they had to do

junior sinew Jan 22, 2024, 6:03 AM

#

https://vkguide.dev/docs/new_chapter_4/textures/

We start by allocating a staging buffer with enough space for the pixel data, on the CPU_TO_GPU memory type.
VMA_MEMORY_USAGE_CPU_TO_GPU is deprecated btw. You should use VMA_MEMORY_USAGE_AUTO_PREFER_HOST and then add VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT or VMA_ALLOCATION_CREATE_HOST_ACCESS_RANDOM_BIT to VmaAllocationCreateInfo::flags

kind ferry Jan 22, 2024, 8:40 AM

#

its more complicated, i prefer using the older vma then

#

its not like vma will stop working

#

the lib barely updates over the years and its basically done

kind ferry Jan 22, 2024, 9:03 AM

#

i understand that new workflow, but i kinda dont like it with the PC target the tutorial has

low minnow Jan 22, 2024, 9:04 AM

#

kind ferry its not like vma will stop working

they’re still marked as deprecated

junior sinew Jan 22, 2024, 9:05 AM

#

maybe put a disclaimer if you're going to tell newbies to use deprecated functionality

kind ferry Jan 22, 2024, 9:09 AM

#

good point

#

vma on vkguide codebase is embedded so it wont auto-update

#

if users want to then update it to latest for some reason then they will see the deprecation stuff

low minnow Jan 22, 2024, 9:33 AM

#

you explicitly using VMA 2 in vkguide sources?

junior sinew Jan 22, 2024, 9:34 AM

#

vma 2?

low minnow Jan 22, 2024, 10:05 AM

#

yeah

#

current version is 3

#

which introduced all of that automatic usage stuff with CPU_TO_GPU and whatnot

#

so I asked if vkguide still uses vma 2

#

see https://github.com/GPUOpen-LibrariesAndSDKs/VulkanMemoryAllocator/releases/tag/v3.0.0

junior sinew Jan 22, 2024, 10:07 AM

#

I see

low minnow Jan 22, 2024, 10:08 AM

#

oh wait no it couldn't use vma 2 as that didn't support bda

#

and iirc vkguide 2 heavily uses that?

kind ferry Jan 22, 2024, 10:45 AM

#

vkguide was using the vma from 2020

#

vkguide 2 was created with that version too

#

i just later went and updated to latest

low minnow Feb 12, 2024, 5:22 PM

#

question about fastgltf

#

the current code in the repository won't compile anymore if anyone uses main or the latest release

#

or at least clarify which version the guide uses

kind ferry Feb 12, 2024, 5:43 PM

#

thats why everything is vendored-in

#

but good point, i should add the versions of the stuff to the guide

kind ferry Feb 13, 2024, 5:03 PM

#

what if i use descriptor buffers for the bindless stuff?

#

im tinkering with them and seems the api is better for bindless usage as its much clearer. you allocate a region of memory to store N texture descriptors and go

#

do you think it would make sense? and whats the current support for them? does all hw that does bindless on pc also do descriptor buffer?

#

tho im not sure if one can combine a descriptor set and the buffer 🤔

lapis isle Feb 13, 2024, 5:22 PM

#

what's the status on renderdoc support for descriptor buffers? last I heard there wasn't any

crimson timber Feb 13, 2024, 5:22 PM

#

i dont see how they help at all for bindless

#

also no renderdoc big nono

kind ferry Feb 13, 2024, 5:57 PM

#

didnt it add them last version?

crimson timber Feb 13, 2024, 6:30 PM

#

i only know of mesh shader s beegin added

drifting maple Feb 13, 2024, 6:31 PM

#

it added shader debugging with BDA

#

(finally)

#

but no descriptor buffer in sight

#

to be honest idk if it's really that useful

#

but it's certainly easier

sage orbit Feb 15, 2024, 2:40 PM

#

~/work/repos_to_check/vulkan-guide/bin$ ./chapter_2
Error when building the compute shader 
Error when building the compute shader 
[ERROR: Validation]
Validation Error: [ VUID-VkPipelineShaderStageCreateInfo-module-parameter ] Object 0: handle = 0x55876070e2f0, type = VK_OBJECT_TYPE_INSTANCE; | MessageID = 0xfd71dc70 | vkCreateComputePipelines(): pCreateInfos[0].stage.module Invalid VkShaderModule Object 0x7ffccf08e830. The Vulkan spec states: If module is not VK_NULL_HANDLE, module must be a valid VkShaderModule handle (https://vulkan.lunarg.com/doc/view/1.3.275.0/linux/1.3-extensions/vkspec.html#VUID-VkPipelineShaderStageCreateInfo-module-parameter)
Segmentation fault (core dumped)

bleakekw thonk

#

Oh, it tries to load shader from ../../shaders, but the executable is in ./bin, so it should search for it in ../shaders - where do executables get built on Windows?

kind ferry Feb 15, 2024, 2:45 PM

#

/bin/Release/engine.exe

sage orbit Feb 15, 2024, 2:45 PM

#

Yeah, that's VS only...

#

Ninja, Make etc. will build in ./bin/chapter_2 etc.

kind ferry Feb 15, 2024, 2:49 PM

#

not really sure how to solve unless it does some kind of path searching at startup

sage orbit Feb 15, 2024, 2:49 PM

#

I'll try some stuff

#

What if you did this?

make_directory("${PROJECT_SOURCE_DIR}/bin/shaders")
foreach(GLSL ${GLSL_SOURCE_FILES})
  message(STATUS "BUILDING SHADER")
  get_filename_component(FILE_NAME ${GLSL} NAME)
  set(SPIRV "${PROJECT_SOURCE_DIR}/bin/shaders/${FILE_NAME}.spv")
  ...

And then in shaders instead of using "../../shaders", you just use "shaders"
(For VS it would need to be a bit more complex, though, to put it to "<Config>/shaders" instead)

#

Basically you'll get this structure

kind ferry Feb 15, 2024, 2:55 PM

#

no

#

because then i would also need to move the assets folder

sage orbit Feb 15, 2024, 2:56 PM

#

Oh, assets use "../../" too bleakekw

#

Well, you can symlink this directory to be present in bin

kind ferry Feb 15, 2024, 3:00 PM

#

lmao no

#

i think the best solution would be to have it autodetected at build time in a define

#

or alternatively, have it auto-search

#

so from .exe location, check up level until it finds both /bin and /assets

lapis isle Feb 15, 2024, 3:00 PM

#

yeah you can get it with a generator expression, I don't see why symlinks are too bad though

kind ferry Feb 15, 2024, 3:01 PM

#

platform support

#

and are megajank overall

#

plus they wouldnt be relative

#

so if you copy the folder and put it somewhere else it would break

#

which is about the biggest nono you can do on a game engine of this sort

sage orbit Feb 15, 2024, 3:02 PM

#

//< init_data
testMeshes = loadGltfMeshes(this,"..\\..\\assets\\basicmesh.glb").value();

froge_bleak

#

Idk if you want to support Linux or not. If not, I'm not going to bother and will fork temporarily to make it build, but won't PR the changes

kind ferry Feb 15, 2024, 3:03 PM

#

just fix that, its basically only in a couple places

#

hapens on windows too if one runs the cmake directly on VS

sage orbit Feb 15, 2024, 3:08 PM

#

if (WIN32)
  set (CMAKE_RUNTIME_OUTPUT_DIRECTORY "${PROJECT_SOURCE_DIR}/bin")
  set (CMAKE_LIBRARY_OUTPUT_DIRECTORY "${PROJECT_SOURCE_DIR}/bin")
else ()
  make_directory("${PROJECT_SOURCE_DIR}/bin/linux")
  set (CMAKE_RUNTIME_OUTPUT_DIRECTORY "${PROJECT_SOURCE_DIR}/bin/linux")
  set (CMAKE_LIBRARY_OUTPUT_DIRECTORY "${PROJECT_SOURCE_DIR}/bin/linux")
endif ()

#

Now ../../<dir> works KEKW

#

There are probably 1000s of better ways to fix this, but I won't bother, heh

lapis isle Feb 15, 2024, 3:12 PM

#

target_compile_definitions(vkguide PRIVATE VKGUIDE_RESOURCE_PATH=$<PATH:GET_RELATIVE_PART,$<TARGET_FILE_DIR:vkguide>>)

I think this works

#

assuming you can nest generator expressions like this

sage orbit Feb 15, 2024, 3:16 PM

#

Also had to do this (chapter 2):

VkPhysicalDeviceVulkan12Features features12{};
features12.bufferDeviceAddress = true;

And I'm still getting some validation errors, might try to fix later

#

E.g. (chapter 2,3 on destruction)

[ERROR: Validation]
Validation Error: [ VUID-vkDestroyPipeline-pipeline-parameter ] Object 0: handle = 0x5635dfa212a0, type = VK_OBJECT_TYPE_INSTANCE; | MessageID = 0xaeb3d1a6 | vkDestroyPipeline(): pipeline Invalid VkPipeline Object 0x8d2b9306125bd000. The Vulkan spec states: If pipeline is not VK_NULL_HANDLE, pipeline must be a valid VkPipeline handle (https://vulkan.lunarg.com/doc/view/1.3.275.0/linux/1.3-extensions/vkspec.html#VUID-vkDestroyPipeline-pipeline-parameter)

kind ferry Feb 17, 2024, 2:44 PM

#

chapter 5 with the refactors needed for texture cache and bindless becomes a huge mess to mantain :/

#

chapter 6 it is then

#

gonna make the code i have now into chapter 6 and clean up chapter 5 a bit

#

i think ill put the bindless texture stuff + some refactoring of the codebase as chapter 6 start parts

#

an important one is to optimize the vertex format and fix a few things so that the fancy brickadia maps and big stuff like bistro can be loaded without any issue

kind ferry Feb 17, 2024, 3:36 PM

#

but until i advance more with chapter 6, ill write a extra-chapter about how to set bindless textures as the basics are simple

kind ferry Feb 17, 2024, 5:09 PM

#

anyway if you want to see how it works the WIP c6 code is up and has bindless

#

need to play with it more

sage orbit Feb 17, 2024, 5:40 PM

#

By the way, if you do add_dependencies(chapter_2 Shaders) (and the same for all the chapters), you'll get automatic shader compilation when the glsl file changes (no need to run Shaders target manually every time)

kind ferry Feb 17, 2024, 5:43 PM

#

i had it before

#

but removed it for a reason

#

dont really remember exactly why

sage orbit Feb 17, 2024, 6:43 PM

#

Finished this: https://vkguide.dev/docs/new_chapter_2/vulkan_shader_code/

Getting validation errors (or I missed something - but I don't see where you free up this stuff and I'm getting the same when I compile code from the repo):

[ERROR: Validation]
Validation Error: [ VUID-vkDestroyDevice-device-05137 ] Object 0: handle = 0x9fde6b0000000014, type = VK_OBJECT_TYPE_DESCRIPTOR_SET_LAYOUT; | MessageID = 0x4872eaa0 | vkCreateDevice():  OBJ ERROR : For VkDevice 0x55a0f266aec0[], VkDescriptorSetLayout 0x9fde6b0000000014[] has not been destroyed. The Vulkan spec states: All child objects created on device must have been destroyed prior to destroying device (https://vulkan.lunarg.com/doc/view/1.3.275.0/linux/1.3-extensions/vkspec.html#VUID-vkDestroyDevice-device-05137)
[ERROR: Validation]
Validation Error: [ VUID-vkDestroyDevice-device-05137 ] Object 0: handle = 0xdd3a8a0000000015, type = VK_OBJECT_TYPE_DESCRIPTOR_SET; | MessageID = 0x4872eaa0 | vkCreateDevice():  OBJ ERROR : For VkDevice 0x55a0f266aec0[], VkDescriptorSet 0xdd3a8a0000000015[] has not been destroyed. The Vulkan spec states: All child objects created on device must have been destroyed prior to destroying device (https://vulkan.lunarg.com/doc/view/1.3.275.0/linux/1.3-extensions/vkspec.html#VUID-vkDestroyDevice-device-05137)
[ERROR: Validation]
Validation Error: [ VUID-vkDestroyDevice-device-05137 ] Object 0: handle = 0xd175b40000000013, type = VK_OBJECT_TYPE_DESCRIPTOR_POOL; | MessageID = 0x4872eaa0 | vkCreateDevice():  OBJ ERROR : For VkDevice 0x55a0f266aec0[], VkDescriptorPool 0xd175b40000000013[] has not been destroyed. The Vulkan spec states: All child objects created on device must have been destroyed prior to destroying device (https://vulkan.lunarg.com/doc/view/1.3.275.0/linux/1.3-extensions/vkspec.html#VUID-vkDestroyDevice-device-05137)

kind ferry Feb 17, 2024, 6:48 PM

#

a lot of the cleanuo code is emitted, no need to worry

sage orbit Feb 17, 2024, 6:50 PM

#

Well, you do it for everything else, so why not do it for all the objects? 😅
(Will train beginners to do all the cleanup needed and have 0 validation errors all the time)

kind ferry Feb 17, 2024, 6:52 PM

#

its mostly failung in intermediate steps

#

almost all of those are fixed at the end of each chapter

sage orbit Feb 17, 2024, 6:54 PM

#

I’ll see how next chapters look for me. But the end result of ch2 has like 10 validation errors 😅

sage orbit Feb 18, 2024, 10:59 AM

#

Also apparently in C++20, implicit capture of this via [=] is deprecated, this thing passed:
https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2018/p0806r2.html
For now the compilers only issue a warning, but it's a bit annoying.

sage orbit Feb 18, 2024, 12:03 PM

#

ImGui chapter will need some modifications since people really love breaking stuff in its Vulkan backend.

immediate_submit is now not needed - font texture is loaded automatically
Need to pass VkPipelineRenderingCreateInfo to ImGui_ImplVulkan_InitInfo
vkDestroyDescriptorPool on ImGui's pool should be called after ImGui_ImplVulkan_Shutdown, otherwise you get validation errors

Here's my commit where I added Dear ImGui successfully 😄
https://github.com/eliasdaler/edbr/commit/0c4cbdcf9fee7c922acc78e2c985c88dbe94d8cc

sage orbit Feb 18, 2024, 12:31 PM

#

Also minor, but maybe using ColorEdit would be more easy to play around with, it's just this:

auto glmToArr = [](const glm::vec4& v) { return std::array<float, 4>{v.x, v.y, v.z, v.w}; };
auto arrToGLM = [](const std::array<float, 4>& v) { return glm::vec4{v[0], v[1], v[2], v[3]}; };

ImGui::Begin("Gradient");

auto from = glmToArr(gradientConstants.data1);
if (ImGui::ColorEdit3("From", from.data())) {
    gradientConstants.data1 = arrToGLM(from);
}

auto to = glmToArr(gradientConstants.data2);
if (ImGui::ColorEdit3("To", to.data())) {
    gradientConstants.data2 = arrToGLM(to);
}

ImGui::End();

low minnow Feb 18, 2024, 12:35 PM

#

sage orbit ImGui chapter will need some modifications since people really love breaking stu...

yeah they had to add the VkPipelineRenderingCreateInfo because of stencil and depth formats iirc

#

I only added a VkFormat for the color attachment in the original dynamic rendering pr

civic mantle Feb 19, 2024, 6:25 PM

#

sage orbit Also apparently in C++20, implicit capture of `this` via `[=]` is deprecated, th...

ima be that guy to say that the best way to get things fixed (when they have an obvious mistake) is to make a PR.
Cause otherwise its a lot of work for the maintainer to go through the issues listed and understand/validate them.

sage orbit Feb 19, 2024, 6:37 PM

#

I’m throwing some of these things in so that vblanco can say if they’re worth fixing or not. I can fix it all, but don’t want to waste my time if PR is then rejected

junior sinew Feb 19, 2024, 7:13 PM

#

also it's way easier to complain on discord than to go through the process of making a pr

#

which is what you said kekkedsadge

sage orbit Feb 19, 2024, 11:21 PM

#

GPU driven chapter update when? 😅
(And how different will it be from the old one?)

fringe berry Mar 4, 2024, 7:41 PM

#

~~TIL vkguide really hates intel~~ (found the issue when talking with vblanco)

lusty hound Mar 21, 2024, 7:00 PM

#

hey stupid question
https://vkguide.dev/docs/new_chapter_2/vulkan_new_rendering/#deletion-queue

Vulkan Guide

Improving the render loop

Practical guide to vulkan graphics programming

#

why do we not call it a deletion stack 🤓

fringe berry Mar 21, 2024, 7:05 PM

#

stack would imply last-in-first-out

lusty hound Mar 21, 2024, 7:10 PM

#

oh wait im dumb LMAO thanks

ancient portal Mar 21, 2024, 7:12 PM

#

The code is ~~FIFO~~LIFO though

fringe berry Mar 21, 2024, 7:13 PM

#

yes, FIFO is correct; LIFO is not

ancient portal Mar 21, 2024, 7:16 PM

#

fringe berry yes, FIFO is correct; LIFO is not

Sorry. I wrote it wrong. The code call callback in reverse order. And IMO that's The desirable behavior (just like C++ destructors)

fringe berry Mar 21, 2024, 7:18 PM

#

...ah wait you're right. I guess stack could make sense 🤔

lusty hound Mar 21, 2024, 7:19 PM

#

AHHHH

#

okay i thought i was going insane

low minnow Mar 21, 2024, 7:20 PM

#

yeah a stack is LIFO or FILO whatever you call it

#

it's just because the first thing that gets created gets destroyed last

#

I guess if you only look at it at the end when it gets destructed its a queue of objects waiting to be destructured

#

but that's just terminology

sturdy ivy Apr 3, 2024, 12:53 PM

#

(please let me know if that's relevant, or should be moved to 'questions')

Hello, there is an issue that has been driving me crazy for a lot of time, and decided to solve, but I am having a hard time. At some point during Chapter 3 > Mesh Loading, you start getting a memory allocation error upon exiting the application:

Assertion failed: m_pMetadata->IsEmpty() && "Some allocations were not freed before destruction of this memory block!", file C:\Users\Evripidis\Desktop\vkguide\vulkan-renderer\third_party\vma\vk_mem_alloc.h, line 11847

However, at this point, it's easily solved by pushing the deletions of the buffers after loading:

testMeshes = loadGltfMeshes(this, "..\\..\\assets\\basicmesh.glb").value();

_mainDeletionQueue.push_function([&]() {
    for (auto& t : testMeshes) {
        destroy_buffer(t->meshBuffers.vertexBuffer);
        destroy_buffer(t->meshBuffers.indexBuffer);
    }
});

Later on with the more complex scene, you get this thing again. Apparently the clearAll(); inside the LoadedGLTF deconstructor either doesn't do the job, or is called at a bad timing.

Has anyone encountered and / or solved this, and would like to share their wisdom? I tried a lot to tackle it as practice, but it starts to get annoying.

sage orbit Apr 3, 2024, 1:41 PM

#

See this KEKW
#1167091099446288444 message

#

I'd recommend making some sort of global "mesh cache" which just goes through all the loaded meshes and removes them.
I don't use deletion queues in my engine anywhere

sturdy ivy Apr 3, 2024, 2:14 PM

#

sage orbit See this <:KEKW:666849321462792234> https://discord.com/channels/31859000788123...

I meaaaan, this particular cleanup code is not emitted bleakekw
Ok, so change logic is the way. But I was also wondering if someone solved that, following the guide itself, out of curiosity.

low minnow Apr 12, 2024, 5:00 PM

#

VK_CHECK(vkCreateSampler(_device, &createInfo, 0, &_depthSampler));
ahhhh who uses 0 instead of nullptr (this is in the compute based culling chapter)

fringe berry Apr 12, 2024, 5:03 PM

#

absolute heresy

kind ferry Apr 12, 2024, 5:16 PM

#

probably copypasted from somewhere lol

tawdry sable Apr 16, 2024, 2:24 PM

#

is the project layout and libraries sections supposed to still list Tiny Obj Loader or is it outdated?

#

yeah its completely mismatched with the building project section

tawdry sable Apr 27, 2024, 8:17 PM

#

https://vkguide.dev/docs/new_chapter_5/gltf_nodes/
In the third paragraph, weren't is spelled as "werent" and simplicity is spelled as "simpicity"

Vulkan Guide

GLTF Scene Nodes

Practical guide to vulkan graphics programming

junior sinew May 4, 2024, 10:37 AM

#

Might be less effort to just open a PR with typo fixes because there are a lot

lusty hound May 11, 2024, 10:31 PM

#

It's really not the end of the world, but does vkguide just stop providing vkinit function bodies?

#

seems like after chapter 3 there aren't function bodies

kind ferry May 12, 2024, 12:10 PM

#

yeh because they are kind of pointless to keep posting when you can look at the provided vkinit file

#

at the start its more important because their members have to be explained more

lusty hound May 12, 2024, 3:12 PM

#

alright alright thank you blanco

lusty hound May 21, 2024, 1:37 PM

#

Yo sorry if this has been answered in vkguide, but why do we not use the in built vertex input and instead opt into handling that ourselves?

kind ferry May 21, 2024, 2:06 PM

#

because there is no point of dealing with vertex input logic any more

#

since the nvidia 1000 series and for a while on AMD vertex input and just accessing vertices from ssbo works about the same

#

and in vkguide 2 we do buffer device adress for the ssbos so it makes even more sense to just remove vertex input as a concept completely

kind ferry May 23, 2024, 12:33 PM

#

i now realized vkguide blending is 100% wrong

#

i have source and destination in reverse rip

lusty hound May 23, 2024, 4:17 PM

#

KEKW

#

thanks i was about to start that section

#

Oh yeah for windowing resizing, wouldn't you also wanna resize your depth image if you're also resizing the draw image as well?

fringe berry May 23, 2024, 4:23 PM

#

kind ferry i now realized vkguide blending is 100% wrong

I saw a user in the vulkan discord with that problem a couple days ago and I forgot to mention, rip

kind ferry May 23, 2024, 4:31 PM

#

ye but only uses additive and additive hqppens to work anyway

#

the one that is megafucked is alphablend

fringe berry May 23, 2024, 4:41 PM

#

yeah it was an interesting problem to figure out 'cause it almost looked like a reverse depth test

lusty hound May 23, 2024, 5:01 PM

#

so uh

#

which ones have their dst and src mixed around?

fringe berry May 23, 2024, 5:02 PM

#

lusty hound which ones have their dst and src mixed around?

https://vkguide.dev/docs/new_chapter_3/blending/ see enable_blending_alphablend

lusty hound May 23, 2024, 5:02 PM

#

ahhhh

#

so additive works as normal

fringe berry May 23, 2024, 5:02 PM

#

ye

lusty hound May 23, 2024, 5:03 PM

#

bleakekw

fringe berry May 23, 2024, 5:07 PM

#

that definitely looks like a layout problem

lusty hound May 23, 2024, 5:07 PM

#

hmm ill look more into it

lusty hound May 23, 2024, 5:09 PM

#

fringe berry that definitely looks like a layout problem

wait do you mean layout like image layout or pipeline layout KEKW

fringe berry May 23, 2024, 5:11 PM

#

image layout

#

it's either a bad layout or some kind of sync issue

lusty hound May 23, 2024, 5:16 PM

#

could it also be some stupid descriptor binding issue

kind ferry May 23, 2024, 5:17 PM

#

yeah so the issue with the blending

#

is that SRC refers to the current pixel in the shader

#

and dst is the render target

#

i had them flipped

#

so yeah all of those DST_alpha need to be SRC_Alpha

lusty hound May 23, 2024, 5:18 PM

#

oh

kind ferry May 23, 2024, 5:19 PM

#

#

this should be the correct one

lusty hound May 23, 2024, 5:20 PM

#

huh so turned out that pixel bug was literally because of that KEKW

#

although the monkey isn't transparent hmmm

lusty hound May 23, 2024, 5:36 PM

#

LET SGOOOO

#

not exactly sure why removing the draw background step would've caused all this sync issue however

kind ferry May 23, 2024, 5:42 PM

#

because the draw background step is a clear

#

without the background step its probably loading uninitialized memory or similar

#

try changing the renderpass to be LOAD_OP_CLEAR

fringe berry May 23, 2024, 5:42 PM

#

ah yeah that'd do it, so you were color blending with undefined

lusty hound May 23, 2024, 5:54 PM

#

AHHH

#

that makes sense

low minnow May 25, 2024, 10:01 AM

#

quick question, why does the DeletionQueue use a std::deque?

We will be using that deque as a First In Last Out queue, so that when we flush the deletion queue, it first destroys the objects that were added into it last.
you can do the same with a std::vector. In fact, the LIFO container from the std is std::stack, which would be exactly what you want.

kind ferry May 25, 2024, 10:03 AM

#

std vector does a full reallocate when it grows

#

deque doesnt

low minnow May 25, 2024, 10:04 AM

#

does it use a linked list underneath?

#

how does that work

kind ferry May 25, 2024, 10:05 AM

#

its a linked list of memory blocks

#

each block can store up to N items (impl defined, and that N is pretty low)

low minnow May 25, 2024, 10:06 AM

#

well interesting

kind ferry May 28, 2024, 11:08 AM

#

https://gitlab.com/Stowy/jai-vulkan-sdl

GitLab

Stowy / Jai Vulkan SDL · GitLab

GitLab.com

#

this dude has a repo of the first chapter of vkguide on jai

#

no vkbootstrap so he grabbed the init code from vulkan-tut

lusty hound Jun 14, 2024, 6:22 PM

#

uh quick question

#

https://vkguide.dev/docs/new_chapter_4/new_drawloop/

Vulkan Guide

Meshes and Camera

Practical guide to vulkan graphics programming

#

So when we multiply view and projection matrices

#

did you get the order wrong?

#

sceneData.viewproj = sceneData.proj * sceneData.view;

#

Fairly certain it is view * proj and not the other way around

low minnow Jun 14, 2024, 6:24 PM

#

with glm that's correct, no

lusty hound Jun 14, 2024, 6:38 PM

#

AHH glam has it different

#

alright that makes sense now

drifting maple Jun 14, 2024, 7:08 PM

#

d3d brained smh

kind ferry Jun 19, 2024, 3:56 PM

#

planning on writing an "extra" article on my experience going from vkguide codebase into more of a game framework (adding logic/physics/etc). any things you think i should mention?

#

no real code, more of a writeup

fringe berry Jun 19, 2024, 4:23 PM

#

Asset management has always been a thorn in my side. Loading and unloading dynamically at runtime.

kind ferry Jun 19, 2024, 4:24 PM

#

then it wont help you

#

because i just hardcode load the gltfs at startup

#

and i have multiple stuff on each gltf so then i just draw what i need from them

kind ferry Jun 19, 2024, 4:49 PM

#

plan is to talk about how to hook a "game loop" into the engine class (through callbacks), pointing to the fix-your-timestep article. Then how to use the gltf loader to render game objects, and how to hook an ecs

#

also how to move the renderer into deferred with postfx

twilit siren Jun 19, 2024, 5:22 PM

#

I am curious how your game object data in your ecs is read by your rendering engine

#

Like if your ecs entity moves in world space how does that position component data make it so it updates where it is drawn

kind ferry Jun 19, 2024, 5:27 PM

#

that one is dumb af lmao

twilit siren Jun 19, 2024, 5:27 PM

#

Why

kind ferry Jun 19, 2024, 5:27 PM

#

i just have a RenderMeshComponent, which holds model matrix + pointer to Mesh (on the gltf object)

twilit siren Jun 19, 2024, 5:28 PM

#

Oh i see

kind ferry Jun 19, 2024, 5:28 PM

#

the renderer loops though those and calls mesh->draw(matrix)

twilit siren Jun 19, 2024, 5:29 PM

#

Yeah that’s what I had too

kind ferry Jun 19, 2024, 5:29 PM

#

the different object types all end up refreshing the RenderMeshComponent from their logic

twilit siren Jun 19, 2024, 5:29 PM

#

Next time I want a separation

#

I want more separation between rendering and game systems

kind ferry Jun 19, 2024, 5:30 PM

#

why, its not that tricky

#

your game systems eventually just store RenderMeshComponent somehow

twilit siren Jun 19, 2024, 5:30 PM

#

It got really messy

kind ferry Jun 19, 2024, 5:32 PM

#

why?

twilit siren Jun 19, 2024, 5:34 PM

#

It just felt like the wrong abstraction to me, I didn’t run into any trouble with it

kind ferry Jun 19, 2024, 5:34 PM

#

my renderer basically only cares about that render-mesh-comp

#

(dont have lights yet, but they would work in the same fashion)

jaunty halo Jun 27, 2024, 7:46 PM

#

interesting website bug, on chrome ios I can't open the menu (but on safari I can)

weak ravine Sep 15, 2024, 2:33 PM

#

no mention to function rebuild_swapchain is in vkguide website, but I see it in chapter 2's source: https://github.com/vblanco20-1/vulkan-guide/blob/all-chapters-2/chapter-2/vk_engine.cpp#L330
and I see that it is used here in draw https://github.com/vblanco20-1/vulkan-guide/blob/all-chapters-2/chapter-2/vk_engine.cpp#L182

that is similar to the chapter 3's resize code. so I guess this is some sort of leftover? not sure this is right channel for the feedback, saw this thread while using discord search about this. anyway, fyi

twilit siren Sep 15, 2024, 4:00 PM

#

The guide provides an email for feedback iirc but people have left it here also

kind ferry Oct 8, 2024, 1:42 PM

#

mentioning it here couse relevant, but im going to write an article on the sheer crimes metafor:refantasio does on its renderer

#

as sort of a combo of render pipeline article + perf article

#

i was thinking of adding it to vkguide but im not sure about it, what do you think?

#

its kinda relevant to vkguide audience given the nature of the article

lapis isle Oct 8, 2024, 1:54 PM

#

yeah seems interesting, frame breakdown blogs are always cool

kind ferry Oct 8, 2024, 2:03 PM

#

i also could use the space to kinda quickly demonstrate how one does a perf analysis

kind ferry Oct 8, 2024, 4:28 PM

#

draft is up https://vkguide.dev/docs/extra-chapter/graphics_analysis_metafor/

Vulkan Guide

Performance and frame analysis - Metaphor ReFantazio

Practical guide to vulkan graphics programming

kind ferry Oct 8, 2024, 4:45 PM

#

@lapis isle what do you think? looking for some feedback before i put it into social media

lapis isle Oct 8, 2024, 4:50 PM

#

still reading through, minus the obvious stuff like fixing the hyperlinks I think it's alright so far

#

also this date confused me lol

#

did you ever look at this with barriers on? the first thing I'd ask is what the data dependencies look like

kind ferry Oct 8, 2024, 4:52 PM

#

no barriers

#

renderdoc shows the barriers as separate compute passes

lapis isle Oct 8, 2024, 4:57 PM

#

huh

#

explain why a FST/FSQ would require more bandwidth than compute

kind ferry Oct 8, 2024, 4:59 PM

#

its about the layer part

#

a fullscreen quad is fine, 5 of them in a row isnt

lapis isle Oct 8, 2024, 4:59 PM

#

oh I think that should be in a separate sentence then

kind ferry Oct 8, 2024, 5:00 PM

#

might reword it a bit then

lapis isle Oct 8, 2024, 5:00 PM

#

cuz I am still wondering what that has to do with compute vs not compute

#

since it seems like it's not compute that would've helped, but merging the kernels into 1 shader?

kind ferry Oct 8, 2024, 5:00 PM

#

that goes a bit clearer at the end

#

where i comment on the perf of things

tidal timber Oct 8, 2024, 5:21 PM

#

small typo in "An stimate"

kind ferry Oct 8, 2024, 6:24 PM

#

fixed both things

#

i need more people to check it out and comment, what do you think @crimson timber (shameless ping) and @lilac oak

lilac oak Oct 8, 2024, 6:27 PM

#

kind ferry i need more people to check it out and comment, what do you think <@238694102437...

unfortunately I cannot offer proofreading service until Vulkanised 2025

fringe berry Oct 10, 2024, 4:15 PM

#

so yesterday you post a blog about the graphics behind the game and today my language learning servers post about the game

#

I think the universe wants me to look at whatever this game is

kind ferry Oct 10, 2024, 4:16 PM

#

you mean the jrpg?

#

there is a demo its pretty cool game

fringe berry Oct 10, 2024, 4:16 PM

#

ye

clever adder Oct 12, 2024, 12:11 PM

#

oh yeah is the article public now?

kind ferry Oct 12, 2024, 2:18 PM

#

been for a couple days

#

wonder if graphics weekly will pick it up because the tweet i did got zero following

kind ferry Oct 12, 2024, 2:45 PM

#

https://x.com/Mad_Triangles/status/1843964939125633110

Mad Triangles Games (@Mad_Triangles) on X

Baffled with the performance of Metaphor: ReFantazio, i went to profile it on #steamdeck and analyze its performance and techniques to see whats going on with Renderdoc and Radeon Profiler. Wrote about it on https://t.co/6YUMQXDkRG #gamedev #vulkan

#

should have baited much harder

kind ferry Oct 15, 2024, 4:42 PM

#

not on the weekly graphics thingy

kind ferry Oct 15, 2024, 4:59 PM

#

seems like this kinda confirms tthat you need to be retweeted enough times to get there

#

reddit/others isnt enough

lapis isle Oct 15, 2024, 5:05 PM

#

have you tried submitting it directly to him

kind ferry Oct 15, 2024, 5:06 PM

#

how?

lapis isle Oct 15, 2024, 5:08 PM

#

idk he's probably got a contact on his site

#

#

doxed

narrow plover Nov 16, 2024, 11:38 PM

#

Hey, I'm kind of new to this. If I had a suggestion/question about a bug I encountered (and sort of fixed?) would I post it in here or would I use GitHub Issues

kind ferry Nov 17, 2024, 4:46 PM

#

in what way

twilit siren Nov 17, 2024, 6:28 PM

#

I read through vkguide quite a bit

#

I like the use of fmt and the guide for sending up imgui with dynamic rendering

#

I'm going to do those things

#

I think how I'll build my renderer is to go through the very outdated vulkan tutorial, do it, get it to work, and then replace the render pass with dynamic rendering, then once that works, replace the code to use shader objects and after that works replace the code with the descriptor indexing extension to get bindless

#

while at each stage looking at how vkguide does it

#

I'm also using SDL and vma as vkguide

#

it's like a blend

#

oh and I like the window resize part of vkguide

#

the reason for using the outdated vulkan tutorial instead of vkguide as the template though is to avoid the engine abstraction at the start

#

idk

#

I just need to start rendering things

narrow plover Nov 17, 2024, 8:15 PM

#

kind ferry in what way

I'm following the tutorial, im at chapter 5.3 (GLTF Scene Nodes and Textures). When I attempt to load structure.glb, my program produces a black screen, and crashes.

The model is being loaded but MeshNode::Draw is failing. It works when I modify it, but I don't think I'm supposed to do this...

void MeshNode::draw(const glm::mat4& top_matrix, DrawContext& ctx)
{
// START OF MODIFICATION: When I add this, most models load
// without it, only a few models will load.
    if (mesh.get() == nullptr)
    {
        return;
    }
// END OF MODIFICATION
    glm::mat4 node_matrix = top_matrix * world_transform;

    for (auto& s : mesh->surfaces)
    {
        RenderObject def;
        def.index_count           = s.count;
        def.first_index           = s.start_index;
        def.index_buffer          = mesh->mesh_buffers.index_buffer.buffer;
        def.material              = &s.material->data;
        def.bounds                = s.bounds;
        def.transform             = node_matrix;
        def.vertex_buffer_address = mesh->mesh_buffers.vertex_buffer_address;

        if (s.material->data.pass_type == MaterialPass::Transparent)
        {
            ctx.transparent_surfaces.push_back(def);
        }
        else
        {
            ctx.opaque_surfaces.push_back(def);
        }
    }

    // recurse down
    Node::draw(top_matrix, ctx);
}

#

I suspect its an issue with my load_gltf method, which i mostly copied from the tutorial

kind ferry Nov 17, 2024, 8:16 PM

#

that makes little sense, might be a problem with the gltf loading

#

because a mesh node is only created when there is a mesh

#

so a meshnode missing a mesh means something went wrong on the loadgltf function

narrow plover Nov 17, 2024, 8:19 PM

#

kind ferry so a meshnode missing a mesh means something went wrong on the loadgltf function

i had to make some changes to it because im using a newer version of fastgltf.
for instance

   fastgltf::GltfDataBuffer data;
    data.loadFromFile(filePath);

    fastgltf::Asset gltf;

    std::filesystem::path path = filePath;

    auto type = fastgltf::determineGltfFileType(&data);

was changed to

    auto data = fastgltf::GltfDataBuffer::FromPath(file_path);
    if (data.get_if() == nullptr)
    {
        fmt::println("Failed to load glTF: {}", file_path);
        return {};
    }

    fastgltf::Asset gltf;

    std::filesystem::path path = file_path;

    auto type = fastgltf::determineGltfFileType(data.get());

#

i think this is relevant to the issue but i could be totally wrong. sorry if im not understanding the problem

kind ferry Nov 17, 2024, 8:21 PM

#

its later

#

when you create the meshnode itself

#

you must give it a mesh, if it doesnt have a mesh then you should be creating a Node, not a MeshNode

narrow plover Nov 17, 2024, 8:25 PM

#

kind ferry you must give it a mesh, if it doesnt have a mesh then you should be creating a ...

Thank you so much, you were exactly right!

#

Now it works! I just need to figure out why the texture loading isnt working now, and then ill be done (think something is wrong with my std::visit...)

#

gltf failed to load texture tile_tech_panels_color_albedo
gltf failed to load texture trim_misc_1_albedo
gltf failed to load texture tile_rivet_panels_albedo
gltf failed to load texture set_asphalt_albedo
gltf failed to load texture tile_anti_slip_albedo
gltf failed to load texture tile_steel_albedo
gltf failed to load texture tile_painted_gun_metal_albedo
gltf failed to load texture trim_misc_2_albedo
gltf failed to load texture tile_tech_panels_albedo
gltf failed to load texture floor_grate_wholes_albedo
gltf failed to load texture column_normal
gltf failed to load texture column_albedo
gltf failed to load texture tile_metal_pillars_albedo
gltf failed to load texture light_shafts

#

i set a breakpoint but it is never reached

kind ferry Nov 17, 2024, 8:31 PM

#

check the paths and which of the visits is triggered

#

the gltf might do something like relative paths that you need to change

#

i use embedded textures as they work better almost always

narrow plover Nov 17, 2024, 8:32 PM

#

im just using the provided structure.glb

#

and as for the visits, im using Visual Studio's breakpoints and they dont seem to trigger? I might be placing them wrong

#

narrow plover Nov 17, 2024, 8:50 PM

#

looks like [&](fastgltf::sources::BufferView& view) is triggered...

#

i wonder if it has anything to do with me doing

reinterpret_cast<const unsigned char*>(vector.bytes.data())```
(same thing happens when i do)
```cpp
(unsigned char*)vector.bytes.data()```

#

because i wanted to change from std::byte * to a stbi_uc *

twilit siren Nov 17, 2024, 9:07 PM

#

follow the guide and once you get it working change stuff

#

then you don't blame the guide, but know it's something with your own code

narrow plover Nov 17, 2024, 9:09 PM

#

twilit siren follow the guide and once you get it working change stuff

i have the models all loaded im just trying to figure out whats wrong with my implementation of load_image()
im not trying to blame the guide, sorry if it came off that way

narrow plover Nov 17, 2024, 9:26 PM

#

oooh im so close. i know its the inner visit for BufferView that messes up. now i need to figure out why buffer.data isnt of type fastgltf::sources::Vector. time to consult the docs once again

#

I THINK I GOT IT WORKING. All thanks to matt328's issue https://github.com/vblanco20-1/vulkan-guide/issues/100

#

it was all because i used a newer version of fastgltf. lesson learned.

low minnow Nov 17, 2024, 10:49 PM

#

ah yeah Array vs Vector

#

I changed this in version 0.8 iirc

narrow plover Nov 19, 2024, 5:53 PM

#

I just ran into an issue where an image was being loaded, but not added to the images map in LoadedGLTF.

I loaded two images with identical names (both ""). Because of this, they cannot co-exist in the map.

Original code:

    for (fastgltf::Image& image : gltf.images)
    {
        std::optional<AllocatedImage> img = load_image(engine, gltf, image);

        if (img.has_value())
        {
            images.push_back(*img);
            file.images[image.name.c_str()] = *img; // image names used as keys
        }
        else
        {
            // use checkerboard texture in case of failure
            images.push_back(engine->_error_checkerboard_image);
            std::cout << "gltf failed to load texture " << image.name << std::endl;
        }
    }

Because of this, the clear_all() method could not delete all the loaded images via the map, and VMA would then complain:

UNFREED ALLOCATION; Offset: 4194304; Size: 4194304; UserData: 0000000000000000; Name: vma_empty; Type: IMAGE_OPTIMAL; Usage: 7
UNFREED ALLOCATION; Offset: 0; Size: 4194304; UserData: 0000000000000000; Name: vma_empty; Type: IMAGE_OPTIMAL; Usage: 7```

I created a workaround by adding a counter to the loop, so every image would have a unique name:
```CPP
    int ic = 0; // image counter
    for (fastgltf::Image& image : gltf.images)
    {
        std::optional<AllocatedImage> img = load_image(engine, gltf, image);

        if (img.has_value())
        {
            images.push_back(*img);
            file.images[(image.name + (char)ic).c_str()] = *img;
        }
        else
        {
            // use checkerboard texture in case of failure
            images.push_back(engine->_error_checkerboard_image);
            std::cout << "gltf failed to load texture " << image.name << std::endl;
        }
        ic++;
    }```

#

Did I miss something in the tutorial? Or is it normal for some GLB images to be nameless? This doesn't happen for all GLBs

kind ferry Nov 19, 2024, 5:56 PM

#

its not normal

#

first time i ever see this happening

narrow plover Nov 19, 2024, 5:56 PM

#

it was happening with ~~the provided structure.glb, and~~ a few others i tested

kind ferry Nov 19, 2024, 5:56 PM

#

like generally the images will have their path/etc as name

#

🤔

#

did something change then

#

i dont remember having any vma issues on the code when i made it

narrow plover Nov 19, 2024, 5:58 PM

#

its really weird, it seems to only happen for GLBs with at least 1 transparent texture

#

but im not certain

#

actually hold on

#

i think the structure.glb is ok, but some of the ones i downloaded from Khronos' repo cause the issue

narrow plover Nov 19, 2024, 6:01 PM

#

kind ferry i dont remember having any vma issues on the code when i made it

dont worry about your structure.glb or structure_mat.glb, they're ok

#

i misspoke

#

I was having the issue with DragonAttenuation.glb and AlphaBlendModeTest.glb https://github.com/KhronosGroup/glTF-Sample-Models/tree/main/2.0
and a sponza.glb I downloaded from elsewhere.

The counter seems to fix my issues but I wonder why the names are blank...

kind ferry Nov 19, 2024, 6:04 PM

#

alphablendmodetest has no names it seems

#

it only has the uri for loading from the jpeg path

narrow plover Nov 19, 2024, 6:05 PM

#

so if i wanted to load it, and properly add its textures to the images map, what should i do?

#

the counter works but the names are just 0,1,2,3,...

#

is it the responsibility of the author of the GLB/glTF file?

kind ferry Nov 19, 2024, 6:13 PM

#

adding the ID into the name works fine i think

#

it doesnt even really have to be a map

low minnow Nov 19, 2024, 7:17 PM

#

narrow plover I was having the issue with ``DragonAttenuation.glb`` and ``AlphaBlendModeTest.g...

The counter seems to fix my issues but I wonder why the names are blank
Simple, the glTF didn't specify any names

#

names are not something you should rely on at all

#

this code is bad in vkguide as well as in your app

#

so either you:

Assign images a unique name
Use another criteria to differentiate, like for example indices or a content hash

narrow plover Nov 19, 2024, 7:23 PM

#

low minnow names are not something you should rely on at all

thank you very much for the advice. im going to stick with the number counter for now, but later on ill probably just use a different data structure

low minnow Nov 19, 2024, 7:25 PM

#

just to quote the spec here:

Whereas indices are used for internal glTF references, optional names are used for application-specific uses such as display. Any top-level glTF object MAY have a name string property for this purpose. These property values are not guaranteed to be unique as they are intended to contain values created when the asset was authored.

#

this should probably also be fixed in vkguide

#

which reminds me that I already have a PR open that fixes a few issues

narrow plover Nov 19, 2024, 7:27 PM

#

thank you for your contributions to vkguide, and thanks for fastgltf

kind ferry Nov 19, 2024, 8:07 PM

#

@low minnow you mean the fastgltf fixes pr from a year ago?

low minnow Nov 19, 2024, 8:08 PM

#

ye its still open

#

idk if they have been addressed or not

kind ferry Nov 19, 2024, 8:38 PM

#

i think i fixed a couple

narrow plover Nov 21, 2024, 8:55 PM

#

@kind ferry I've completed chapter 5 recently and I just want to say: thank you for creating and maintaining Vkguide. It was relatively straight-forward and very informative. You saved me so much time!

If I wanted to contribute to the future versions of the guide (e.g. helping with grammar, bugfixes, etc.), would it be best to make a issue on GitHub, or should it be mentioned here?

kind ferry Nov 21, 2024, 9:51 PM

#

both

#

here allows discussing it a bit

narrow plover Nov 22, 2024, 6:17 PM

#

what is the purpose of

struct GLTFMaterial
{
    MaterialInstance data;
};```
in ``vk_loader.h``

why wouldn't I just use ``MaterialInstance`` directly?

kind ferry Nov 22, 2024, 7:09 PM

#

placeholder to then have more stuff later, particularly for different passes of that same material

#

narrow plover Nov 22, 2024, 9:23 PM

#

ah makes sense, ill label it as such.

#

any plans for MSAA in the next version?

kind ferry Nov 22, 2024, 9:28 PM

#

msaa is just kinda trivial to add

#

if anything it would just be a short extra chapter

twilit siren Nov 24, 2024, 10:24 AM

#

I got to the monke

kind ferry Nov 24, 2024, 10:33 AM

#

hell yeah

twilit siren Nov 24, 2024, 10:04 PM

#

so I am using shader objects instead of pipelines, I think what I'll just do though is write something similar to the vkguide pipeline builder but instead of it building a pipeline it just basically records a bunch of commands into the command buffer

pulsar egret Dec 9, 2024, 2:59 AM

#

Was going over https://vkguide.dev/docs/new_chapter_1/vulkan_mainloop_code/, and I found this line a bit strange, not sure if a typo or a joke that is going over my head (cuz its related to multi gpu I think?):

command_buffer_submit_info only needs the command buffer handle. We dont need anthing else, and we can leave the deviceMask at 0 as we are not it.

twilit siren Dec 15, 2024, 2:40 AM

#

https://github.com/vblanco20-1/vulkan-guide/blob/master/LICENSE.txt#L3 <--- is this who I should put as the copyright owner in my own License file, I am not sure if that is correct

GitHub

vulkan-guide/LICENSE.txt at master · vblanco20-1/vulkan-guide

Introductory guide to vulkan. Contribute to vblanco20-1/vulkan-guide development by creating an account on GitHub.

kind ferry Dec 15, 2024, 9:21 AM

#

thats because im using a open source theme

#

the MIT license there is from the theme, not the website

#

tho its a bit unclear i def should update that part

twilit siren Dec 15, 2024, 9:36 AM

#

yes I put my own name on my licenses

low minnow Feb 27, 2025, 1:02 AM

#

https://github.com/spnda/fastgltf/issues/90

GitHub

Destination Type needs to have the same AccessorType as accessor. ·...

Hello, I've been trying to implement fastgltf following vkguide for the most part, an issue I have run into is that the older version of fastgltf provided with the vkguide repo doesn't work...

#

this is a bug within the vkguide code because it unconditionally loads vertex colors as glm::vec4 which is plainly wrong

#

            // load vertex colors
            auto colors = p.findAttribute("COLOR_0");
            if (colors != p.attributes.end()) {

                fastgltf::iterateAccessorWithIndex<glm::vec4>(gltf, gltf.accessors[(*colors).second],
                    [&](glm::vec4 v, size_t index) {
                        vertices[initial_vtx + index].color = v;
                    });
            }

#

this should be used instead:

            // load vertex colors
            auto colors = p.findAttribute("COLOR_0");
            if (colors != p.attributes.end()) {
              if (colors->second.type == AccessorType::Vec4) {
                fastgltf::iterateAccessorWithIndex<glm::vec4>(gltf, gltf.accessors[(*colors).second],
                    [&](glm::vec4 v, size_t index) {
                        vertices[initial_vtx + index].color = v;
                    });
              } else if (colors->second.type == AccessorType::Vec3) {
                fastgltf::iterateAccessorWithIndex<glm::vec3>(gltf, gltf.accessors[(*colors).second],
                    [&](glm::vec3 v, size_t index) {
                        vertices[initial_vtx + index].color = glm::vec4(v, 1.f);
                    });
              }
            }

kind ferry Feb 27, 2025, 9:00 AM

#

blender used vec4s so i just never tested it

#

i kinda explain in the guide that its meant to work with gltfs from blender and just barely

#

i cant really use the space to fix all the stuff that is hardcoded/wrong in the loader part

kind ferry Aug 25, 2025, 1:20 PM

#

https://vkguide.dev/docs/ascendant/ascendant_geometry/ The article on the geometry pipeline of project ascendant is up. Not fully sold on how it ended but oh well

Vulkan Guide

High-performance voxel and mesh rendering

Practical guide to vulkan graphics programming

#

next one will be about framegraphs and gbuffers

civic mantle Sep 15, 2025, 2:06 AM

#

@kind ferry Dunno if you saw it, but I added a nifty feature to vk-bootstrap that lets you query why physical device creation failed, EG it'll generate a list of failed conditions like missing extensions & features. https://github.com/charles-lunarg/vk-bootstrap/pull/404
I'm more than happy to add it to the guide as a PR for ya, just wanting to give you a heads up.

kind ferry Sep 16, 2025, 4:49 PM

#

@civic mantle its a common issue people have, so if you can PR the change, then ill merge

#

you need to edit the source text on the new_guide folder, because the folder at /docs/ is codegen-d to insert the code snippets

#

you might need to hook the snippets too

civic mantle Sep 16, 2025, 4:53 PM

#

sounds good.
I'll add in the "dump reasons why it failed" to the snippets as there is no point otherwise

kind ferry Nov 12, 2025, 3:26 PM

#

Resurrecting this thing, im going to begin writing chapter 6, which is going to be PBR + lighting shaders

#

recomended test scenes? i have the gltf spaceship thingy from godot but its not very well configured in regards to shaders

#

also, do you know whats the best way of doing PBR formulas. @lilac oak i think knew about it. Im going to be doing forward+

lilac oak Nov 12, 2025, 3:29 PM

#

WASSUP

lilac oak Nov 12, 2025, 3:29 PM

#

kind ferry also, do you know whats the best way of doing PBR formulas. <@497820759096295429...

wdym PBR formulas ?

kind ferry Nov 12, 2025, 3:29 PM

#

the correct BRDF math

#

i remember you mentioned learnopengl was dogshit

lilac oak Nov 12, 2025, 3:30 PM

#

mmmm

kind ferry Nov 12, 2025, 3:30 PM

#

maybe i could go with the Filament ones, but afaik those are also fairly approximate

lilac oak Nov 12, 2025, 3:30 PM

#

son

#

knock yourself the fuck out

#

https://github.com/Devsh-Graphics-Programming/Nabla/tree/master/include/nbl/builtin/hlsl/bxdf

kind ferry Nov 12, 2025, 3:30 PM

#

given im making a new chapter i want to use high end formulas or whatever is the best ones to use atm

lilac oak Nov 12, 2025, 3:30 PM

#

I just merged this BxDF shit

#

all correct

#

all unit tested

kind ferry Nov 12, 2025, 3:30 PM

#

i was thinking maybe grab the khronos PBR material standard, with the multilyaer fuckery

lilac oak Nov 12, 2025, 3:30 PM

#

https://github.com/Devsh-Graphics-Programming/Nabla-Examples-and-Tests/pull/165/files

kind ferry Nov 12, 2025, 3:30 PM

#

and try to get that to work math-wise

lilac oak Nov 12, 2025, 3:31 PM

#

for non-RT rendering you just want my BxDF's eval() method

kind ferry Nov 12, 2025, 3:32 PM

#

do you happen to know good materials/models/etc i can use for validation ?

lilac oak Nov 12, 2025, 3:32 PM

#

if you want to PT or generate IBL (preintegrated envmap for every reflection vector) you can use generate and quotient_and_pdf

kind ferry Nov 12, 2025, 3:32 PM

#

im also thinking making this depth prepass-d forwardplus with msaa

#

so in a way, the chapter would be like
first part, fix material system and implement the better PBR models
next, implement bruteforce based pointlights and spotlights
then, implementing clustered rendering. I was thinking cluster so it affects transparent objects and i can build clusters from cpu for simplicity

#

and that would be the chapter

lilac oak Nov 12, 2025, 3:35 PM

#

lilac oak if you want to PT or generate IBL (preintegrated envmap for every reflection vec...

interaction.V = IBL_dir;
for (uint i=0; i<SampleCount; i++)
{
   float2 xi = scramble(lowDiscrepancySequence(i),IBL_uv);
   L = bxdf.sample(xi);
   IBLcolor += (bxdf.quotient_and_pdf(L).value-IBLColor)/float(i+1); // online average
}

#

pseudocode how you'd do IBL with my BxDF lib

#

C++20 concepts in shaders go BRRR

kind ferry Nov 12, 2025, 3:36 PM

#

interesting

#

maybe i should also do the chapter on slang

lilac oak Nov 12, 2025, 3:36 PM

#

kind ferry maybe i should also do the chapter on slang

prolly

#

organise your shit

lilac oak Nov 12, 2025, 3:39 PM

#

lilac oak <https://github.com/Devsh-Graphics-Programming/Nabla/tree/master/include/nbl/bui...

you can rewrite this to Slang I think

#

like my weird concepts are just Slang Interfaces

lilac oak Nov 12, 2025, 3:39 PM

#

kind ferry so in a way, the chapter would be like first part, fix material system and imple...

bruteforce pointlights and spotlights aren't really bruteforce

#

ah you mean the culling

kind ferry Nov 12, 2025, 3:40 PM

#

ye, just for progression

lilac oak Nov 12, 2025, 3:40 PM

#

tbf I don't like lights with limited ranges

kind ferry Nov 12, 2025, 3:40 PM

#

pass array of lights to the shader and go

#

neither do i

#

lights must always have a range limit, its nonsense if they dont

lilac oak Nov 12, 2025, 3:40 PM

#

nope 😛

#Vkguide 2 thread