glass sphinx Nov 22, 2023, 11:20 PM

#

good

#

no braingarbage

distant lodge Nov 22, 2023, 11:20 PM

#

same

delicate rain Nov 22, 2023, 11:20 PM

#

I forgot 50% of Vulkan

#

Maybe more

glass sphinx Nov 22, 2023, 11:20 PM

#

very good\

delicate rain Nov 22, 2023, 11:20 PM

#

Gone

distant lodge Nov 22, 2023, 11:20 PM

#

all the *Layout stuff is happily gone

#

goodbye

delicate rain Nov 22, 2023, 11:20 PM

#

Replaced by daxa knowledge

#

And tears

distant lodge Nov 22, 2023, 11:21 PM

#

all these vulkan structures lost to time, like tears in the rain

delicate rain Nov 22, 2023, 11:21 PM

#

If Patrick (or me) actually rewrites task graph there will be so many bugs

#

I will not pull master for a month at least

distant lodge Nov 22, 2023, 11:22 PM

#

oh yeah I am worried about looking at my frame graph again

wicked notch Nov 22, 2023, 11:22 PM

#

there is only so many neurons in the human brain

#

it's either vulkan or your abstraction

#

can't have both

distant lodge Nov 22, 2023, 11:23 PM

#

I really need to look into timeline semaphores though, I initially wrote them off

glass sphinx Nov 22, 2023, 11:24 PM

#

delicate rain If Patrick (or me) actually rewrites task graph there will be so many bugs

i am just [utting makeup on

#

i actually start to be backwards compatible + deprecation now

#

too many people shitting on me when i break things now

delicate rain Nov 22, 2023, 11:24 PM

#

No ☹️

#

I hate deprecating

#

We just remove

glass sphinx Nov 22, 2023, 11:25 PM

#

#

I will use this against you

delicate rain Nov 22, 2023, 11:25 PM

#

I only shit on you because I'm tilted from my bugs and you are near

glass sphinx Nov 22, 2023, 11:25 PM

#

froge_evil

delicate rain Nov 22, 2023, 11:25 PM

#

I always pull by accident in the middle of something and then have to spend 2 years updating

#

I hate C API though

#

Wish we could deprecate that

wicked notch Nov 22, 2023, 11:25 PM

#

https://tenor.com/view/thanos-just-the-snap-avengers-infinity-war-gif-12393235

Tenor

#

saky be like:

delicate rain Nov 22, 2023, 11:26 PM

#

SmallString gone bye bye

#

Actual 50 bugs in 200 loc

glass sphinx Nov 22, 2023, 11:27 PM

#

you are my continuous integration

#

https://tenor.com/view/lemon-demon-robot-jones-cabinet-man-neil-cicierega-half-human-half-machine-gif-17315975

Tenor

#

you ^

wicked notch Nov 22, 2023, 11:27 PM

#

~~human slavery~~
continuous integration

glass sphinx Nov 22, 2023, 11:27 PM

#

froghorror

wicked notch Nov 24, 2023, 12:14 AM

#

The ShaRT is finally complete

#

resource/state tracking is cancer

#

Now it's time for the even funnier part

#

edge cases bleakekw

frank sail Nov 24, 2023, 12:34 AM

#

can you remind me why you ShaRTed yourself

#

(I forgot what it means)

wicked notch Nov 24, 2023, 11:19 AM

#

The ShaRT or shader resource table is a big ass descriptor set that keeps track of resources bound to it

#

it's very useful because I can turn my brain off and get OpenGL level of ease when binding resources

wicked notch Nov 24, 2023, 11:36 AM

#

template <uint64 N, typename T>
struct _descriptor_table_t {
    struct write_info_t {
        using content_type = std::conditional_t<std::is_same_v<T, arc_ptr<buffer_t<uint8>>>, buffer_info_t, image_info_t>;

        uint32 binding = -1;
        uint32 offset = -1;
        content_type content = {};
    };
    std::array<T, N> resources = {};
    std::array<uint64, N / 64> used = {};
    std::vector<write_info_t> updates;
};

_descriptor_table_t<max_descriptors_in_binding, arc_ptr<buffer_t<uint8>>> _buffer_table;
_descriptor_table_t<max_descriptors_in_binding, arc_ptr<sampler_t>> _sampler_table;
_descriptor_table_t<max_descriptors_in_binding, arc_ptr<image_t>> _image_table;
_descriptor_table_t<max_descriptors_in_binding, arc_ptr<texture_t>> _texture_table;
arc_ptr<descriptor_set_t> _set;```

glass sphinx Nov 24, 2023, 1:25 PM

#

i like the lower case naming on first glance

#

something unusual

#

looks tempting

#

arc_ptr is also a sexy name

distant lodge Nov 24, 2023, 2:22 PM

#

intel only

wispy spear Nov 24, 2023, 2:29 PM

#

the underscore isms for types is ugly af

#

but i like the lowercase_t stuff too

#

i suppose _foo_t is supposed to be "internal" only

#

still ugly 😄

distant lodge Nov 24, 2023, 2:31 PM

#

yet it has an internal type that doesn't have an underscore

wispy spear Nov 24, 2023, 2:31 PM

#

what is arc supposed to be?

wicked notch Nov 24, 2023, 3:31 PM

#

atomic reference counted

wicked notch Nov 24, 2023, 3:39 PM

#

distant lodge yet it has an internal type that doesn't have an underscore

because it's not internal relative to the parent type

#

but it's probably better if all children of an internal type are classified as internal too

#

fo readability

wispy spear Nov 30, 2023, 12:06 PM

#

its weird

#

the more i look at iris, the more i realise thats exactly how i write c# code

#

besides the self and (auto -> returntype) nonsense its quite neat too how you write code : )

wicked notch Nov 30, 2023, 12:09 PM

#

I steal concepts from other langs bleakekw

frank sail Nov 30, 2023, 12:09 PM

#

says the rust hater

wicked notch Nov 30, 2023, 12:09 PM

#

not me hating rust while integrating half of their features KEKW

distant lodge Nov 30, 2023, 1:56 PM

#

EEE but awesome

wispy spear Dec 2, 2023, 9:22 PM

#

lvstri my man

#

how can i use your buffer_t in conjunction with std::make_unique

#

your buffer is using a neat factory method to create the thing... buffer_t::create which returns a buffer_t... but i would like to
std::unique_ptr<buffer_t> _vertexBuffer and _vertexBuffer = std::make_unique(buffer_t::create(......));

#

@wicked notch

wicked notch Dec 2, 2023, 9:26 PM

#

You can just return buffer and C++ will take care of everything for you

#

I think buffer needs to be movable though

#

for ptr = std::make_unique<T>(T::create()); to work

wispy spear Dec 2, 2023, 9:26 PM

#

hmmf

wicked notch Dec 2, 2023, 9:27 PM

#

I return arc_ptr<buffer_t> if you notice

#

just removing the arc_ptr part will let you use make_unique, assuming your buffer is movable

wispy spear Dec 2, 2023, 9:27 PM

#

i use Iris' bufferisms, not irisvk

wicked notch Dec 2, 2023, 9:28 PM

#

oh

#

I forgor how those work bleakekw

wispy spear Dec 2, 2023, 9:28 PM

#

: >

wicked notch Dec 2, 2023, 9:28 PM

#

let me check one sec

wispy spear Dec 2, 2023, 9:28 PM

#

shall i peek at irisvks'?

wicked notch Dec 2, 2023, 9:30 PM

#

ok ye

#

my buffer_t in GL Iris should work fine with make_unique I think

#

did you try it and it doesn't work?

wispy spear Dec 2, 2023, 9:31 PM

#

i am getting a template dump of isms

#

let me see if i can copy it out somehow

#

[build] /home/deccer/Personal/Code/Projects/lessGravity/OpenSpacePP/src/GameClient/GameApplication.cpp:52:121: error: no match for ‘operator=’ (operand types are ‘std::unique_ptr<Buffer>’ and ‘Buffer’)
[build]    52 |     _vertexBuffer = Buffer::Create("Buffer_Vertices_Triangle", GL_ARRAY_BUFFER, SizeInBytes(_vertices), GL_MAP_WRITE_BIT);
[build]       |                                                                                                                         ^
[build] In file included from /usr/include/c++/13.2.1/memory:78,
[build]                  from /home/deccer/Personal/Code/Projects/lessGravity/OpenSpacePP/src/Engine/Include/Engine/Application.hpp:4,
[build]                  from /home/deccer/Personal/Code/Projects/lessGravity/OpenSpacePP/src/GameClient/Include/GameClient/GameApplication.hpp:3,
[build]                  from /home/deccer/Personal/Code/Projects/lessGravity/OpenSpacePP/src/GameClient/GameApplication.cpp:1:
[build] /usr/include/c++/13.2.1/bits/unique_ptr.h:430:9: note: candidate: ‘template<class _Up, class _Ep> constexpr typename std::enable_if<std::__and_<std::__and_<std::is_convertible<typename std::unique_ptr<_Up, _Ep>::pointer, typename std::__uniq_ptr_impl<_Tp, _Dp>::pointer>, std::__not_<std::is_array<_Up> > >, std::is_assignable<_Dp&, _Ep&&> >::value, std::unique_ptr<_Tp, _Dp>&>::type std::unique_ptr<_Tp, _Dp>::operator=(std::unique_ptr<_Up, _Ep>&&) [with _Ep = _Up; _Tp = Buffer; _Dp = std::default_delete<Buffer>]’
[build]   430 |         operator=(unique_ptr<_Up, _Ep>&& __u) noexcept
[build]       |         ^~~~~~~~
[build] /usr/include/c++/13.2.1/bits/unique_ptr.h:430:9: note:   template argument deduction/substitution failed:
[build] /home/deccer/Personal/Code/Projects/lessGravity/OpenSpacePP/src/GameClient/GameApplication.cpp:52:121: note:   ‘Buffer’ is not derived from ‘std::unique_ptr<_Tp, _Dp>’
[build]    52 |     _vertexBuffer = Buffer::Create("Buffer_Vertices_Triangle", GL_ARRAY_BUFFER, SizeInBytes(_vertices), GL_MAP_WRITE_BIT);
[build]       |                                                                                                                         ^
[build] /usr/include/c++/13.2.1/bits/unique_ptr.h:414:19: note: candidate: ‘constexpr std::unique_ptr<_Tp, _Dp>& std::unique_ptr<_Tp, _Dp>::operator=(std::unique_ptr<_Tp, _Dp>&&) [with _Tp = Buffer; _Dp = std::default_delete<Buffer>]’
[build]   414 |       unique_ptr& operator=(unique_ptr&&) = default;
[build]       |                   ^~~~~~~~
[build] /usr/include/c++/13.2.1/bits/unique_ptr.h:414:29: note:   no known conversion for argument 1 from ‘Buffer’ to ‘std::unique_ptr<Buffer>&&’
[build]   414 |       unique_ptr& operator=(unique_ptr&&) = default;
[build]       |                             ^~~~~~~~~~~~
[build] /usr/include/c++/13.2.1/bits/unique_ptr.h:440:7: note: candidate: ‘constexpr std::unique_ptr<_Tp, _Dp>& std::unique_ptr<_Tp, _Dp>::operator=(std::nullptr_t) [with _Tp = Buffer; _Dp = std::default_delete<Buffer>; std::nullptr_t = std::nullptr_t]’
[build]   440 |       operator=(nullptr_t) noexcept
[build]       |       ^~~~~~~~
[build] /usr/include/c++/13.2.1/bits/unique_ptr.h:440:17: note:   no known conversion for argument 1 from ‘Buffer’ to ‘std::nullptr_t’
[build]   440 |       operator=(nullptr_t) noexcept
[build]       |                 ^~~~~~~~~

#

i apologise for this wall ofisms

#

i cant take a screenshot of the popup in vscode, when i do the momenti press enter to capture the popup disappears ;C

wicked notch Dec 2, 2023, 9:33 PM

#

I don't see le make_unique though, did you forgor?

wispy spear Dec 2, 2023, 9:33 PM

#

one sec

#

    _vertexBuffer = Buffer::Create("Buffer_Vertices_Triangle", GL_ARRAY_BUFFER, SizeInBytes(_vertices), GL_MAP_WRITE_BIT);

#

left out the make unique because you said so earlier

wicked notch Dec 2, 2023, 9:34 PM

#

Oh my bad yeah you need the make_unique here

#

try this: _vertexBuffer = std::make_unique<Buffer>(Buffer::Create("Buffer_Vertices_Triangle", GL_ARRAY_BUFFER, SizeInBytes(_vertices), GL_MAP_WRITE_BIT));

wispy spear Dec 2, 2023, 9:35 PM

#

yeah i just did that and ofc it works :3

#

sorry for the nonsense : )

wispy spear Dec 2, 2023, 10:04 PM

#

https://github.com/deccer/OpenSpacePP/blob/main/src/GameClient/GameApplication.cpp#L52-L57 works ❤️

wispy spear Dec 5, 2023, 1:15 AM

#

struct custom_hash_unique_object_representation {
    using is_avalanching = void;

    [[nodiscard]] auto operator()(point const& f) const noexcept -> uint64_t {
        static_assert(std::has_unique_object_representations_v<point>);
        return ankerl::unordered_dense::detail::wyhash::hash(&f, sizeof(f));
    }
};

#

is ankerl using those usings as comments? there is no use of is_avalanching nor void

#

ive seen you have it all over the place in irisvk too

#

ah he explains it

#

i was just to impatient

#

weird thing either way

frank sail Dec 5, 2023, 1:25 AM

#

Now you can never complain about my hashisms again

wispy spear Dec 5, 2023, 1:27 AM

#

: )

#

yours has a lot of ...voodoo in it

#

i admire you guys for understanding all that c++ nonsense, when to use what and how : )

finite yacht Dec 5, 2023, 1:30 AM

#

i am also more on the c# side but I imagine its like learning OpenGL

wispy spear Dec 5, 2023, 1:30 AM

#

i would say its 100 times more complicated

#

opengl has no template nonsense, no variadics, no this-and-that-elision

frank sail Dec 5, 2023, 1:31 AM

#

But at the end of the day it's just a programming language and we're all familiar with those already

wispy spear Dec 5, 2023, 1:32 AM

#

yep

#

just like c# is "hard" for noobs who just started 5mins ago

frank sail Dec 5, 2023, 1:32 AM

#

Learning OpenGL without knowing another API is hard af

wispy spear Dec 5, 2023, 1:35 AM

#

when i have

struct A
{
    XXX x;
};

struct B
{
    YYY y;
};

struct C
{
    A a;
    A aa;
    std::span<const B> bs;    
};

#

my understanding is thati have to implement operator== for all of them first

#

if i want to put them in a set-like-container

frank sail Dec 5, 2023, 1:36 AM

#

you can do = default; on them

wispy spear Dec 5, 2023, 1:36 AM

#

yeah for trivial types?

frank sail Dec 5, 2023, 1:36 AM

#

if the members support comparison already

wispy spear Dec 5, 2023, 1:37 AM

#

is that span covered by that as well?

#

struct C
{
    A a;
    A aa;
    std::span<const B> bs;

    bool operator==(const C& other) const
    {
        return a == other.a &&
               aa == other.aa &&
               bs == other.bs;
    }
};

#

hmm i believe i saw some std::funky_v thing which can detect whether an array is different, not sure where i did

frank sail Dec 5, 2023, 1:39 AM

#

apparently std::span doesn't implement comparison

#

https://stackoverflow.com/questions/60633668/why-does-stdspan-lack-the-comparison-operators

#

but you could use this
https://en.cppreference.com/w/cpp/algorithm/ranges/equal

#

should be as shrimple as && std::ranges::equal(bs, other.bs)

wispy spear Dec 5, 2023, 1:41 AM

#

nice

#

and now i need to impl a hash function per type too, if i dont want to solve it by 1 template

wicked notch Dec 5, 2023, 1:43 AM

#

you can also default <=> for easy mode comparisons

frank sail Dec 5, 2023, 1:43 AM

#

https://brevzin.github.io/c++/2020/03/30/span-comparisons/

wispy spear Dec 5, 2023, 1:43 AM

#

yeah jakerino mentioned it, for trivial types i can just = default

frank sail Dec 5, 2023, 1:44 AM

#

<=> is different

wispy spear Dec 5, 2023, 1:44 AM

#

ah its something else

frank sail Dec 5, 2023, 1:44 AM

#

it also does the < and > operators at the same time

wispy spear Dec 5, 2023, 1:44 AM

#

:S

frank sail Dec 5, 2023, 1:44 AM

#

it's the "spaceship" operator (your favorite)

wispy spear Dec 5, 2023, 1:44 AM

#

: )

frank sail Dec 5, 2023, 1:44 AM

#

https://brevzin.github.io/c++/2020/03/30/span-comparisons/

wispy spear Dec 5, 2023, 1:45 AM

#

ah this one also mentions span_ext

wicked notch Dec 5, 2023, 1:46 AM

#

remember to write a specialization for equal_to as well

#

unless

#

no nvm you can avoid that

wispy spear Dec 5, 2023, 1:46 AM

#

specialization means template?

wicked notch Dec 5, 2023, 1:47 AM

#

ye but you can avoid that if you just default/define == no prob

wispy spear Dec 5, 2023, 1:47 AM

#

i want to try solve this in a naive fashion first

#

no unreadable templates for now if possible heh

wicked notch Dec 5, 2023, 1:50 AM

#

you have ugly makros to solve unreadable templates though smart

wispy spear Dec 5, 2023, 1:52 AM

#

hehe

wispy spear Dec 5, 2023, 1:53 AM

#

wicked notch you have ugly makros to solve unreadable templates though <:smart:59186497729658...

https://tenor.com/view/dredd-i-knew-youd-say-that-stallone-gif-23229845

Tenor

#

i suppose what im trying right now will also not work

#

make hash functions for all of those

struct A
{
    XXX x;
};

struct B
{
    YYY y;
};

struct C
{
    A a;
    A aa;
    std::span<const B> bs;    
};

#

struct HashFunctions
{
  std::size_t operator() (const XXX& xxx) const
  {
    auto h1 = std::hash<whateverX1is>()(xxx.x1);
    return h1;
  }

  std::size_t operator() (const A& a) const
  {
    auto h1 = std::hash<XXX>()(a.x);
    return h1;
  }

  std::size_t operator() (const YYY& yyy) const
  ... and so forth... for all types involved
};

frank sail Dec 5, 2023, 2:09 AM

#

yeah you basically need a hash function for everything

#

the only nice thing is that each thing only needs to call the hasher on each of its members, as long as those members (and its members, etc.) have a hash function

wispy spear Dec 5, 2023, 2:11 AM

#

😛

#

#

the plain xor should also be wrong, usually you multiply by some prime number and xor, but thats not important here right now

frank sail Dec 5, 2023, 2:13 AM

#

look up a hash combiner

#

I copied some variadic template thing from SO

wispy spear Dec 5, 2023, 2:14 AM

#

i saw yours

frank sail Dec 5, 2023, 2:14 AM

#

I wish I better understood how it worked exactly

#

like why they use giant prime numbers instead of just xoring everything like u

wispy spear Dec 5, 2023, 2:15 AM

#

avoid collisions would be my guess

frank sail Dec 5, 2023, 2:18 AM

#

ah it probably rotates the bits "randomly"

#

if you hash two of the same value with your thing, you'll end up with 0 when they get xor'd

#

but a smarter hasher will have a seed that gets randomized before it combines one hash with another

wispy spear Dec 5, 2023, 2:21 AM

#

that makes sense

#

ah you also only use it for the VertexArrayCache

#

not for the framebuffercache

frank sail Dec 5, 2023, 2:23 AM

#

I made the framebuffer cache a vector for some reason

wispy spear Dec 5, 2023, 2:23 AM

#

perhaps one of the two existed before the other

frank sail Dec 5, 2023, 2:23 AM

#

could do the same for the vao cache

wispy spear Dec 5, 2023, 2:33 AM

#

not great, not terrible

#

    size_t GetFramebufferInfoHash(const FramebufferInfo& framebufferInfo)
    {
        size_t hashValue = {};

        auto part1 = std::make_tuple(framebufferInfo.DepthAttachment.Attachment._id, framebufferInfo.DepthAttachment.ClearOperation);
        Hash::CombineHash(hashValue, Hash::Hash<decltype(part1)>{}(part1));

        auto part2 = std::make_tuple(framebufferInfo.StencilAttachment.Attachment._id, framebufferInfo.StencilAttachment.ClearOperation);
        Hash::CombineHash(hashValue, Hash::Hash<decltype(part2)>{}(part2));

        for (const auto& colorAttachment : framebufferInfo.ColorAttachments)
        {
            auto part3 = std::make_tuple(colorAttachment.Attachment._id, colorAttachment.ClearOperation);
            Hash::CombineHash(hashValue, Hash::Hash<decltype(part3)>{}(part3));
        }

        return hashValue;
    }

frank sail Dec 5, 2023, 2:34 AM

#

btw I'm not sure if caching framebuffers is worth it

#

compared to just having one framebuffer that you respec a bunch

wispy spear Dec 5, 2023, 2:34 AM

#

hmm

#

id still need to cache those specs

#

that i know when to respec the single fbo

#

or just respec every frame?

frank sail Dec 5, 2023, 2:36 AM

#

how often are you binding the same set of render targets

wispy spear Dec 5, 2023, 2:36 AM

#

0 per frame

frank sail Dec 5, 2023, 2:36 AM

#

frog_thinkk

#

ok so respecing is fine

wispy spear Dec 5, 2023, 2:36 AM

#

yeah

frank sail Dec 5, 2023, 2:36 AM

#

and you do it every time you bind a new set of render targets

wispy spear Dec 5, 2023, 2:37 AM

#

yes

frank sail Dec 5, 2023, 2:37 AM

#

no need to hash it up

#

(probs)

wispy spear Dec 5, 2023, 2:37 AM

#

absolutely

#

we were much younger a year ago when we wrote fwog/enginekit : >

frank sail Dec 5, 2023, 2:37 AM

#

hehe

#

caching makes me feel secure sometimes

#

even if it's definitely premature

wispy spear Dec 5, 2023, 2:38 AM

#

yep, right now im not caching anything at all

#

and that will most likely be fine too

#

bindpipeline will enable/disable state, bindasubo/bind-ass-sbo/bindimage/texture will ust run per frame, then setting the fbo can run per frame too

#

i think i wanted to use the cache as impl detail

#

to hide teh fbo creation

#

instead of
_fbo1 = createfbo(attachments...)
_fbo2 = createfbo(otherattachments...);

while (true)
{
beginrender(_fbo1);
bindpipelione(_pipeloine1);
...
endrender();
...
}

#

have
...
beginrender({ ... describe fbo info here ... })
...

finite yacht Dec 5, 2023, 2:46 AM

#

you can do exactly that when re-modifying a single fbo

wispy spear Dec 5, 2023, 2:48 AM

#

hmm absolutely true

#

haha

#

you guys are the best ducks

twin bough Dec 11, 2023, 6:17 AM

#

quack

frank sail Dec 11, 2023, 11:28 PM

#

https://www.jcgt.org/published/0012/02/01/

wispy spear Dec 11, 2023, 11:28 PM

#

fresh off the press

wicked notch Dec 11, 2023, 11:39 PM

#

for a fraction of a second

#

I'd hoped to see some epic algorithm that generated meshlets on the fly like we do in frogfood

#

rip

frank sail Dec 12, 2023, 12:20 AM

#

the paper links this repo
https://github.com/Senbyo/meshletmaker

#

I looked at a random source file and saw
#define TINYOBJLOADER_IMPLEMENTATION
so I'm guessing it's not supposed to be used as a library

wispy spear Dec 12, 2023, 12:21 AM

#

or the author doesnt know how to make a library (yet)

#

1:21:21.212

frank sail Dec 12, 2023, 12:22 AM

#

KEKW
https://github.com/Senbyo/meshletmaker/blob/main/core/idxBufCleaner.cpp

wispy spear Dec 12, 2023, 12:22 AM

#

😄

frank sail Dec 12, 2023, 12:23 AM

#

a bunch of empty files too

wispy spear Dec 12, 2023, 12:23 AM

#

cleaning indicles isnt important enough

frank sail Dec 12, 2023, 12:23 AM

#

std::vector<uint32_t> AreaWeightedTriangleList(const std::vector<Triangle*>& triangles, const Vertex* vertexBuffer) {

frank sail Dec 12, 2023, 12:31 AM

#

frank sail https://www.jcgt.org/published/0012/02/01/

do they normally not review the language in the paper to make sure it uses correct spelling and grammar

wispy spear Dec 12, 2023, 12:32 AM

#

id hope somebody does

frank sail Dec 12, 2023, 12:32 AM

#

I'm on page 1 and I've found several issues already 😄

wispy spear Dec 12, 2023, 12:45 AM

#

oof

wicked notch Dec 20, 2023, 12:01 PM

#

revive thyself!

#

ok so big update, the shader resource table is now fully functional

#

also big update, it's winter break!

#

I have one week of time before death

wispy spear Dec 20, 2023, 12:06 PM

#

ill get bob the necromancer, after you have passed, no worries

#

you cant escape VSM/Nanite 2.0 🙂

wicked notch Dec 20, 2023, 2:25 PM

#

my C++ brain also can't understand rust's modules

#

I have no idea how they work

wicked notch Dec 20, 2023, 7:57 PM

#

Also, I think I'm going to do a slight rewrite of IrisVk

#

there's a lot of clutter and stuff from the past

wispy spear Dec 20, 2023, 7:59 PM

#

https://tenor.com/view/awkward-im-out-byee-ok-gif-15304701432226785572

Tenor

wicked notch Dec 23, 2023, 11:40 PM

#

man

#

Eradicating renderpasses feels good

#

dynamic rendering ftw

distant lodge Dec 24, 2023, 12:39 AM

#

time for me to integrate renderpass2 (double the renderpasses to put the world into equilibrium)

pale horizon Dec 24, 2023, 10:52 PM

#

Reject render pass, return to immediate drawing (GPUs are fast) KEKW

frank sail Dec 25, 2023, 1:19 AM

#

vkQueueSubmit for every triangle bleakekw

wicked notch Dec 25, 2023, 11:06 AM

#

frogmas

wicked notch Dec 25, 2023, 12:30 PM

#

I have a question for my boy potrick

#

Would you mind explaining why this line here works: https://github.com/Ipotrick/Daxa/blob/75f4b85da63af1ad01de0934d479a27c8307ed6f/src/impl_swapchain.cpp#L179-L183?

#

I have the same thing in my code, except I wait for the next target value

#

auto SyncHostTimeline::WaitForNextTimelineValue() noexcept -> uint64 {
    const auto target = static_cast<int64>(_hostTimelineValue) - static_cast<int64>(_maxTimelineDifference);
    _deviceTimelineSemaphore->Wait(std::max<int64>(target + 1, 0));
    return _hostTimelineValue % _maxTimelineDifference;
}```

#

My reasoning is:
First FIF: wait for timeline 0 -> signals timeline 1
Second FIF: wait for timeline 0 -> signals for timeline 2
First FIF Again: wait for timeline 1 -> signals for timeline 3
...

#

Here _hostTimelineValue starts out as 0 and _maxTimelineDifference is 2

glass sphinx Dec 25, 2023, 1:24 PM

#

hmm your reasoning makes sense rn

#

i dont remember the details

#

When I'm back home in 2 days i ll take another look

wicked notch Dec 25, 2023, 1:25 PM

#

sure thing, merry christmas potrick frogeheart

glass sphinx Dec 25, 2023, 1:25 PM

#

merry Christmas lvstri 🌲

delicate rain Dec 25, 2023, 1:48 PM

#

wicked notch My reasoning is: First FIF: wait for timeline 0 -> signals timeline 1 Second FIF...

If you do this you only have 1 frame in flight no?

#

As frame 0 is your worked on frame, frame 1 is your frame in flight, and frame 2 waits for frame 0 to finish - so to become the frame in flight instead of frame 1

wicked notch Dec 25, 2023, 1:50 PM

#

That's what happens though isn't it?

#

0 waits NONE (or 0) signals 1
1 waits NONE (or 0) signals 2
2 waits 1 signals 3
3 waits 2 signals 4
...

delicate rain Dec 25, 2023, 1:52 PM

#

Well in Daxa (with 2 frames in flight), 2 would not wait for 1 but wait for 0 no?

#

As the previous signal value is 2 - 2fif = 0

wicked notch Dec 25, 2023, 1:53 PM

#

so 2 frames in flight means that the CPU can prepare 3 frames before waiting right?

delicate rain Dec 25, 2023, 1:53 PM

#

Yes

wicked notch Dec 25, 2023, 1:53 PM

#

gotcha

delicate rain Dec 25, 2023, 1:53 PM

#

(I was also confused about that, before Patrick explained)

wicked notch Dec 25, 2023, 1:53 PM

#

I defined frames in flight as the number of frames that the CPU can prepare

delicate rain Dec 25, 2023, 1:53 PM

#

Yeah I had the same reasoning in mind

#

But apparently it's not like that

wicked notch Dec 25, 2023, 1:54 PM

#

so when you make buffers that are updated from the CPU you allocate (FRAMES_IN_FLIGHT + 1) * capacity instead of just FIF * capacity?

delicate rain Dec 25, 2023, 1:57 PM

#

We should

#

Now thinking about it I'm not sure I do in my projects hehe

#

Because I, until recently, thought the same and thought FIF = CPU buffered frames

glass sphinx Dec 25, 2023, 3:59 PM

#

we have many comments in the cpp for swapchain

#

without them i wouldnt be able to reunderstand

#

sadly they only seem to be enough for me to understand. Probably cause i wrote them for me froge_bleak

wicked notch Dec 25, 2023, 4:23 PM

#

llvm's cmake tries to run sh on windows if you don't use MSVC

#

How can people be so big brained to make LLVM and then proceed to make the most unhinged, retarded and stupid ass cmake build script ever known to man

wicked notch Dec 25, 2023, 4:42 PM

#

ihatewindowsihatewindowsihatewindowsihatewindows

#

Completely random errors

#

llvm cmake can't find python with MSVC

#

made sure it was in the path and all

distant lodge Dec 25, 2023, 4:54 PM

#

perfect timing, you're getting your brain in a twist with FIF and I'm finally upgrading to timeline semaphores

#

the +1 is quite weird though

wicked notch Dec 25, 2023, 5:00 PM

#

it comes from my misunderstanding for what a FIF is

#

I thought that N FIF meant "The CPU can prepare N frames before waiting"

distant lodge Dec 25, 2023, 5:01 PM

#

that's my understanding, and I'm pretty sure that's how my code works, e.g. 0 FIF would mean nothing happens

#

1 FIF means that, if my CPU takes less time to submit+present than my GPU takes to execute, I will have to wait on my sema every frame

wicked notch Dec 25, 2023, 5:02 PM

#

ye, if you define FIF like that then you need to either target the next timeline value, or decrease the max timeline difference by one

distant lodge Dec 25, 2023, 5:03 PM

#

yeah, that's what I do, doing that difference thing seems weirdly complicated. I literally just do

WaitSemaphores(frameIndex);
Submit(signal = frameIndex + 1);
++frameIndex;

#

and I use frameIndex % FIF for the binary semas on my swapchain

wicked notch Dec 25, 2023, 5:04 PM

#

you need to wait for 0 twice though

#

that's what the difference thing does

distant lodge Dec 25, 2023, 5:05 PM

#

nah, frame 0 waits for 0 which is a no-op since that's the start of the semaphore, frame 1 waits for 1 because of the ++frameIndex, and the last submit signals frameIndex + 1, so for frame 0 that'd be 1

#

wait hold on let me think about this some more...

wicked notch Dec 25, 2023, 5:06 PM

#

wicked notch That's what happens though isn't it?

you likely want this behavior

#

for 2 FIF

distant lodge Dec 25, 2023, 5:06 PM

#

yeah right shit

#

I was just happy that switching from fences to timeline semas ran with no validation errors so I didn't think about it lol

delicate rain Dec 25, 2023, 5:12 PM

#

distant lodge yeah, that's what I do, doing that difference thing seems weirdly complicated. I...

How would this work for more fif

#

You never take the number of fif into consideration

distant lodge Dec 25, 2023, 5:29 PM

#

lmao yep, time to rethink it (and probably make it work the same way as yours)

#

I guess it's just

signal(frame) = frame + 1
wait(frame) = max(frame - (FIF - 1), 0)

wicked notch Dec 25, 2023, 5:43 PM

#

ye

distant lodge Dec 25, 2023, 5:43 PM

#

and I'm guessing you don't do FIF - 1 hence the discrepancy

wicked notch Dec 25, 2023, 5:44 PM

#

I do max(frame - FIF + 1, 0) which is the same thing

distant lodge Dec 25, 2023, 5:45 PM

#

yeah, I meant that the - 1 is the reason for the discrepancy in thinking that FIF frames is FIF + 1 frames prepared before waiting

wicked notch Dec 25, 2023, 6:08 PM

#

How does a human being remove vcpkg's cmake integration

#

vcpkg integrate remove says there's no integration

#

but running cmake fatally fails with the message that vcpkg is missing

wispy spear Dec 25, 2023, 6:11 PM

#

is there some FindVcpkg nonsense maybe?

wicked notch Dec 25, 2023, 6:13 PM

#

there's a singular cmake variable in a cache somewhere

#

I have no idea where this "global cache" is

wicked notch Dec 26, 2023, 1:39 PM

#

I have fixed everything

#

#include <Macros.hlsl>

struct SVertexOutput {
    Location(0) float4 Position : SV_Position;
    Location(1) float4 Color : Color;
};

SVertexOutput VSMain(uint vertexId : SV_VertexID) {
    const float3 trianglePositions[] = {
        float3( 0.5,  0.5, 0.0),
        float3(-0.5,  0.5, 0.0),
        float3( 0.0, -0.5, 0.0)
    };
    const float3 triangleColors[] = {
        float3(1.0, 0.0, 0.0),
        float3(0.0, 1.0, 0.0),
        float3(0.0, 0.0, 1.0)
    };
    SVertexOutput output;
    output.Position = float4(trianglePositions[vertexId], 1.0);
    output.Color = float4(triangleColors[vertexId], 1.0);
    return output;
}

struct SPixelOutput {
    Location(0) float4 Color : SV_Target0;
};

SPixelOutput PSMain(SVertexOutput input) {
    SPixelOutput output;
    output.Color = input.Color;
    return output;
}

le tringle with le HLSL

#

let's see how far I can push DXC before it gives up on me

#

Also, with new IrisVk the tringle sample is only 100 LOC

wicked notch Dec 26, 2023, 10:07 PM

#

struct SCamera {
    float4x4 Projection;
    float4x4 View;
    float4x4 ProjView;
    float4 Position;
};
[[Binding(0, 0)]] StructuredBuffer<SCamera> b_cameraBuffer[];``` this syntax is honestly pretty great

#

who said HLSL buffers suck

wispy spear Dec 26, 2023, 10:08 PM

#

bindless when

wicked notch Dec 26, 2023, 10:08 PM

#

no bda in hlsl

#

agonyfrog

wispy spear Dec 26, 2023, 10:08 PM

#

;C

wicked notch Dec 26, 2023, 10:08 PM

#

at least I can do shader resource table

#

but meh

wispy spear Dec 26, 2023, 10:09 PM

#

also funny how you went PascalCase : )

wicked notch Dec 26, 2023, 10:09 PM

#

I have

#

I have been persuaded by Microsoft and the C# gang

wispy spear Dec 26, 2023, 10:09 PM

#

hehe

#

is bda coming?

wicked notch Dec 26, 2023, 10:09 PM

#

idk, devsh is working on it

wispy spear Dec 26, 2023, 10:09 PM

#

ah

wicked notch Dec 26, 2023, 10:10 PM

#

he's bullying DXC devs full time

wispy spear Dec 26, 2023, 10:10 PM

#

aaah i remember some conversation about it

twin bough Dec 26, 2023, 10:10 PM

#

add bda to hlsl pls 🥺

wicked notch Dec 27, 2023, 2:32 PM

#

I love dxc

#

ModuleProcessed("dxc-commit-hash: 93ad5b31");
ModuleProcessed("dxc-cl-option:  -E VSMain -T vs_6_6 -spirv -fspv-target-env=vulkan1.3 -Zpc -Zi -O0 -I D:/Dev/CLion/Retina/Shaders");


struct SCamera {
  float4x4 Projection : [[RowMajor, MatrixStride(16), Offset(0)]];
  float4x4 View : [[RowMajor, MatrixStride(16), Offset(64)]];
  float4x4 ProjView : [[RowMajor, MatrixStride(16), Offset(128)]];
  float4 Position : [[Offset(192)]];
}```

#

me: make the default column major please

#

DXC: oh yeah sure (he doesn't know lmao)

wicked notch Dec 28, 2023, 12:14 AM

#

@glass sphinx

#ifndef RETINA_HLSL_BINDINGS_HEADER
#define RETINA_HLSL_BINDINGS_HEADER

#include <Macros.hlsl>

#define RETINA_MASTER_ADDRESS_BINDING [[Binding(0, 0)]]

RETINA_MASTER_ADDRESS_BINDING ByteAddressBuffer b_masterBuffer[];
RETINA_MASTER_ADDRESS_BINDING RWByteAddressBuffer b_rwMasterBuffer[];

#define DeclareAliasedStructuredBuffer(type, name)               \
    RETINA_MASTER_ADDRESS_BINDING StructuredBuffer<type> name[];

#define DeclareAliasedRWStructuredBuffer(type, name)               \
    RETINA_MASTER_ADDRESS_BINDING RWStructuredBuffer<type> name[];

#endif
``` here's my epic fix for HLSL not having BDA

#

Is there any better way?

glass sphinx Dec 28, 2023, 12:26 AM

#

wicked notch <@238694102437330944> ```glsl #ifndef RETINA_HLSL_BINDINGS_HEADER #define RETINA...

i do exactly the same bleakekw

wicked notch Dec 28, 2023, 12:26 AM

#

rip

glass sphinx Dec 28, 2023, 12:26 AM

#

i think i ll kick out structured buffer tho

wicked notch Dec 28, 2023, 12:27 AM

#

why?

glass sphinx Dec 28, 2023, 12:27 AM

#

byteaddressbuffer is the best of the bunch

wicked notch Dec 28, 2023, 12:27 AM

#

froge_bleak

#

is it?

glass sphinx Dec 28, 2023, 12:27 AM

#

i find its interface the best after having used it a lot at work

wicked notch Dec 28, 2023, 12:27 AM

#

you can only load uint's tho

glass sphinx Dec 28, 2023, 12:27 AM

#

ohohoh no

#

not true

#

it can load and stroe and atomic anything

wicked notch Dec 28, 2023, 12:27 AM

#

how

#

teach me

glass sphinx Dec 28, 2023, 12:27 AM

#

https://github.com/microsoft/DirectXShaderCompiler/wiki/ByteAddressBuffer-Load-Store-Additions

GitHub

ByteAddressBuffer Load Store Additions

This repo hosts the source for the DirectX Shader Compiler which is based on LLVM/Clang. - microsoft/DirectXShaderCompiler

wicked notch Dec 28, 2023, 12:27 AM

#

maybe the docs are outdated?

glass sphinx Dec 28, 2023, 12:28 AM

#

they are its a little sus with the new things

#

the only downside of byteaddressbuffer is that it generates a lot of loads

#

it doesnt understand alignment past 4 bytes

#

so if make an aligned load/store of lets say 16 bytes it will cut it up into 4 spirv loads

#

sad life

wicked notch Dec 28, 2023, 12:29 AM

#

damn

glass sphinx Dec 28, 2023, 12:29 AM

#

only for spirv supposedly

wicked notch Dec 28, 2023, 12:29 AM

#

is it that difficult to have good buffers

#

microsoft pls

glass sphinx Dec 28, 2023, 12:29 AM

#

in dx its doing aligned loads just fine

#

hlsl vulkan is kinda poop in that way

wispy spear Dec 28, 2023, 12:30 AM

#

need to put that on devsh's agenda

glass sphinx Dec 28, 2023, 12:30 AM

#

afaik its also not a technical limitation they just implemented the spirv support for byteaddressbuffer poorly

#

could be wrong on that one

wicked notch Dec 28, 2023, 12:31 AM

#

sigh

#

I'll stay with structuredbuffer for now

glass sphinx Dec 28, 2023, 12:31 AM

#

its fine TM

wicked notch Dec 28, 2023, 12:32 AM

#

thanks for the wiki page btw, sneaky little thing

#

lots of useful info there

glass sphinx Dec 28, 2023, 12:32 AM

#

tho dont be afraid from the syntax suggesting you load a whole stuct or store a whole struct

#

if you do this:

RWByteAddressBuffer b : register(u0);
struct Tester{ float4 vec; uint4 intvec; };
void test()
{
  Tester t = b.Load<Tester>(some_offset);
  t.vec[0] = 1.3f;
  b.Store(some_offset, T);
}

This will consistently be optimized to only store the first element of vec and do nothing else

distant lodge Dec 28, 2023, 12:35 AM

#

I saw some posts in #vulkan a while back about that generating different spirv than the equivalent GLSL, did they fix that or is it moreso that it converges at the driver level

glass sphinx Dec 28, 2023, 12:35 AM

#

distant lodge I saw some posts in <#318783283984990210> a while back about that generating dif...

that was about RawBufferLoad

#

and RawBufferWrite, the poopy bda integration for hlsl. ByteAddressBuffers do not have this problem.

wicked notch Dec 28, 2023, 12:35 AM

#

glass sphinx if you do this: ```hlsl RWByteAddressBuffer b : register(u0); struct Tester{ flo...

does the same hold for StructuredBuffer<T>

glass sphinx Dec 28, 2023, 12:35 AM

#

wicked notch does the same hold for StructuredBuffer<T>

yes

wicked notch Dec 28, 2023, 12:35 AM

#

nice

distant lodge Dec 28, 2023, 12:35 AM

#

oh so what you're saying is HLSL BDA can give you perf landmines if you're porting from GLSL

#

fun

glass sphinx Dec 28, 2023, 12:36 AM

#

definetly yes

#

well depends on how you do it

#

but they just dont optimize it properly

#

sadge

wicked notch Dec 28, 2023, 12:36 AM

#

as I am seeing if you use StructuredBuffer<T> you're gonna be fine

glass sphinx Dec 28, 2023, 12:36 AM

#

BufferPtr in hlsl will safe us all

wicked notch Dec 28, 2023, 12:36 AM

#

ByteAddressBuffer is more tricky but it'll be fine as well

glass sphinx Dec 28, 2023, 12:37 AM

#

not for me tho. I use tons or non aligned offsets everywhere within my buffers

#

cant really use structured buffers for much in my engine

distant lodge Dec 28, 2023, 12:37 AM

#

maybe 2024 will be the year of the HLSL BDA

glass sphinx Dec 28, 2023, 12:37 AM

#

my meshes for example are all subacllocated within buffers

#

xD

wicked notch Dec 28, 2023, 12:37 AM

#

damn, I've yet to encounter a scenario where I need non-aligned load/store

glass sphinx Dec 28, 2023, 12:37 AM

#

froghorror

#

well i also have lots of buffers thyat have a counter in the first few byutes

#

i guess that is a common one you might encounter, no?

wicked notch Dec 28, 2023, 12:38 AM

#

oh

#

I guess that counts as unaligned yes

#

KEKW

#

yeah I hadn't thought about that shit

glass sphinx Dec 28, 2023, 12:38 AM

#

💀

distant lodge Dec 28, 2023, 12:38 AM

#

what is unaligned in this scenario

glass sphinx Dec 28, 2023, 12:39 AM

#

well structured buffer requires alignment to the whole struct size

#

so you cant just put 4 bytes in the beginning

#

as your structs may have wild sizes

distant lodge Dec 28, 2023, 12:39 AM

#

so you can't just be a smartass and make it part of your struct?

#

and have the rest of the data as the inner part

glass sphinx Dec 28, 2023, 12:39 AM

#

huuuuuuuuuuuuuuuh

#

i guess so

#

yea that works but then you cant have unbound anything

wicked notch Dec 28, 2023, 12:40 AM

#

wait hold up how the hell do I translate to HLSL

layout (...) buffer X {
    uint x;
    T data[];
};```

#

???

glass sphinx Dec 28, 2023, 12:40 AM

#

yea cant do

wicked notch Dec 28, 2023, 12:40 AM

#

💀

distant lodge Dec 28, 2023, 12:40 AM

#

wow that sux

wicked notch Dec 28, 2023, 12:40 AM

#

alright we're pulling the plug on HLSL

#

it's over

wispy spear Dec 28, 2023, 12:41 AM

#

inb4 lvstri reinventing Cg

wicked notch Dec 28, 2023, 12:41 AM

#

nah I have to be quick

#

gotta make the game

wispy spear Dec 28, 2023, 12:41 AM

#

: )

distant lodge Dec 28, 2023, 12:41 AM

#

glsl is fine tbh, gotta give hlsl more time to git gud

glass sphinx Dec 28, 2023, 12:42 AM

#

🤠

wicked notch Dec 28, 2023, 12:42 AM

#

devsh can't bully dxc devs enough really

#

we need more bullying

wispy spear Dec 28, 2023, 12:42 AM

#

even potti wishes to go back to glsl but is stuck in d3d12

distant lodge Dec 28, 2023, 12:42 AM

#

so true

#

if they want me to stop using glsl they gotta earn it

wicked notch Dec 28, 2023, 1:13 PM

#

GLSL is back in

#

froge_bleak

wicked notch Dec 28, 2023, 2:03 PM

#

also it's kinda crazy how the vulkan sdk is so broken with other dependencies like shaderc or dxc lol

#

they have absolute paths to PDBs in their cmake apparently

glass sphinx Dec 28, 2023, 3:01 PM

#

lmao

wispy spear Dec 28, 2023, 4:57 PM

#

can you not apply for some internship at lunarg or something

#

and finish your masters there

wicked notch Dec 29, 2023, 1:21 AM

#

I love templates

#

2AM template code is the best

wispy spear Dec 29, 2023, 11:16 AM

#

i dont approve of C prefix in Regina, but i understand why its there and therefore implicitly approve its use

wicked notch Dec 29, 2023, 1:31 PM

#

template <
    typename C,
    typename R = C::Resource,
    typename D = C::Descriptor,
    typename... Args
>
    requires (std::same_as<D, SBufferDescriptor>)
RETINA_NODISCARD auto AllocateResource(
    SConstructResourceInlineTag,
    Args&&... args
) noexcept -> C;

template <
    typename C,
    typename R = C::Resource,
    typename D = C::Descriptor,
    typename... Args
>
    requires (std::same_as<D, SBufferDescriptor>)
RETINA_NODISCARD auto AllocateResource(
    uint32 count,
    Args&&... args
) noexcept -> std::vector<C>;

template <
    typename C,
    typename R = C::Resource,
    typename D = C::Descriptor,
    typename... Args
>
    requires (std::same_as<D, SImageDescriptor>)
RETINA_NODISCARD auto AllocateResource(
    SConstructResourceInlineTag,
    EImageLayout layout,
    Args&&... args
) noexcept -> C;

template <
    typename C,
    typename R = C::Resource,
    typename D = C::Descriptor,
    typename... Args
>
    requires (std::same_as<D, SImageDescriptor>)
RETINA_NODISCARD auto AllocateResource(
    uint32 count,
    EImageLayout layout,
    Args&&... args
) noexcept -> std::vector<C>;``` This was the template madness of yesterday btw ![KEKW](https://cdn.discordapp.com/emojis/666849321462792234.webp?size=128 "KEKW")

wicked notch Dec 29, 2023, 1:31 PM

#

wispy spear i dont approve of C prefix in Regina, but i understand why its there and therefo...

ye types and functions being pascal case caused some issues

#

so I went with the Unreal Engine™️ code style

wispy spear Dec 29, 2023, 3:50 PM

#

R = C

#

D = C

#

😄

#

can you write out C, D and R? Descriptor, Resource, Cock reads better

wicked notch Dec 29, 2023, 10:21 PM

#

layout (push_constant) uniform UPushConstant {
    uint u_inputImageId;
};

RetinaDeclareStorageImage(restrict readonly image2D, u_image2DBlock);
#define inputImage RetinaGetStorageImage(u_image2DBlock, u_inputImageId)

vec3 Tonemap(in vec3 color) {
    const float luminance = GetLuminance(color);
    const vec3 reinhard = color / (color + vec3(1.0));
    return Saturate(mix(color / (luminance + 1.0), reinhard, reinhard));
}

void main() {
    o_output = vec4(ToNonLinearFromLinear(Tonemap(vec3(imageLoad(inputImage, ivec2(gl_FragCoord.xy))))), 1.0);
}

#

beautiful

pale horizon Dec 29, 2023, 10:21 PM

#

nice Lisp program

wicked notch Dec 29, 2023, 10:22 PM

#

@glass sphinx how do you manage daxa::BufferId lifetimes

glass sphinx Dec 30, 2023, 7:58 AM

#

manually

#

the c ghetto way

glass sphinx Dec 30, 2023, 8:28 AM

#

fr tho: for daya to dictate the lifetime managememt for buffers and images aside from manual free and delete is sus because they have top variee lifetime requirements

#

and refcounting is too slow for this many objects for a general solution imo

runic surge Dec 30, 2023, 8:47 AM

#

bleakekw

#

I’m literally ref counting it all

wicked notch Dec 30, 2023, 11:37 AM

#

glass sphinx and refcounting is too slow for this many objects for a general solution imo

fair

wicked notch Dec 30, 2023, 8:23 PM

#

le tringle

glass sphinx Dec 30, 2023, 8:25 PM

#

runic surge I’m literally ref counting it all

its a good solution too

wispy spear Dec 30, 2023, 8:25 PM

#

rendered out of thin air i assume?

glass sphinx Dec 30, 2023, 8:25 PM

#

ref coubting gives you a lot of benefits that many ppl ignore/dont know

runic surge Dec 30, 2023, 8:31 PM

#

yeah like i can just fire and forget bleakekw

#

and i don't what to worry about manually cleaning it all up later but that can also be a shot in the foot if i somehow mess up and do circular ref counts

glass sphinx Dec 30, 2023, 8:33 PM

#

also lors of multithreadding guarantees

#

if you pass a rer counted handle to a function via ref you can know 100% the object wont be destroyed while running the function

#

which is a big problem with manual management

runic surge Dec 30, 2023, 8:35 PM

#

yeah i kinda just accepted that i won't probably be making 1000+ vulkan objects in a single frame so performance is just neglible

glass sphinx Dec 30, 2023, 8:35 PM

#

in daxa i avoided these peoblems with lots of hacks and a big id system

runic surge Dec 30, 2023, 8:35 PM

#

bleakekw

glass sphinx Dec 30, 2023, 8:35 PM

#

ref counting would have been muuuuch simpler

runic surge Dec 30, 2023, 8:35 PM

#

huh big id system?

glass sphinx Dec 30, 2023, 8:36 PM

#

hard to explain

#

index + version like entt

#

but it is also mt save and such via some other shingus

runic surge Dec 30, 2023, 8:36 PM

#

KEKW i don't even know what entt does

glass sphinx Dec 30, 2023, 8:36 PM

#

very fast tho

#

https://tenor.com/view/oh-dear-uhoh-uh-oops-gif-22478082

Tenor

#

entt is great

#

its the usual ecs id tho

runic surge Dec 30, 2023, 8:38 PM

#

mhm

glass sphinx Dec 30, 2023, 8:38 PM

#

the extra daxa shingus

#

magic

runic surge Dec 30, 2023, 8:38 PM

#

magik

#

❇️

#

my favorite magic engine (daxa)

glass sphinx Dec 30, 2023, 8:39 PM

#

https://tenor.com/view/finding-dory-big-eyes-gif-11607952788069142933

Tenor

#

notice the daya purple in the eyes

#

i payed disney for it

wicked notch Dec 30, 2023, 8:54 PM

#

1280 bytes for a BLAS

#

a single triangle BLAS

#

with compaction

#

wtf nvidia

glass sphinx Dec 30, 2023, 9:10 PM

#

they have a fixed minimal overhead

#

i had like 80 in and it was the same size

runic surge Dec 31, 2023, 6:24 AM

#

wicked notch 1280 bytes for a BLAS

Iirc nvidia dos and donts says that it is not worth it compacting really small meshes such as particles (which only take up 1-2 triangles)

wicked notch Dec 31, 2023, 1:23 PM

#

struct SAccelerationStructureGeometryInstance {
    glm::mat3x4 Transform = {};
    uint32 ObjectIndex : 24 = 0;
    uint32 Mask : 8 = 0xff;
    uint32 ShaderBindingTableOffset : 24 = 0;
    EAccelerationStructureGeometryInstanceFlag Flags : 8 = {};
    uint64 AccelerationStructureAddress = 0;
};``` why bitfields

#

why

#

why would you do this to me

wicked notch Dec 31, 2023, 2:25 PM

#

refresh me on matrix memory layout

#

if the transform matrix is row major then the last column contains the transform?

frank sail Dec 31, 2023, 2:28 PM

#

how the matrix is displayed there shouldn't depend on the majorness since that only defines the memory layout

#

tl;dr that column should be the translation

wicked notch Dec 31, 2023, 2:29 PM

#

True

#

I guess this helps more

auto ToNativeTransformMatrix(const glm::mat4& transform) noexcept -> glm::mat3x4 {
    RETINA_PROFILE_SCOPED();
    auto result = glm::mat3x4();
    std::memcpy(&result, AsConstPtr(glm::transpose(transform)), sizeof(glm::mat3x4));
    return result;
}```

#

I think this is correct

#

maybe

frank sail Dec 31, 2023, 2:30 PM

#

another way to remember is that the rotation+scale are expressed by a 3x3, and the translation is a 1x3

#

and with a non-square matrix you can tell by looking

wicked notch Jan 1, 2024, 12:42 AM

#

bistro be looking a little wonky

#

And I have to go out in 5 minutes

wispy spear Jan 1, 2024, 12:43 AM

#

solo un pochino

wicked notch Jan 2, 2024, 12:40 AM

#

what laziness does to a mf:

if (u_frameIndex > 0) {
    const vec3 previousLighting = vec3(imageLoad(g_accumulationImage, ivec2(gl_LaunchIDEXT.xy)));
    incomingLight = mix(previousLighting, incomingLight, 1.0 / float(u_frameIndex + 1));
}
imageStore(g_accumulationImage, ivec2(gl_LaunchIDEXT.xy), vec4(incomingLight, 1.0));
imageStore(g_inputImage, ivec2(gl_LaunchIDEXT.xy), vec4(incomingLight, 1.0));```

runic surge Jan 2, 2024, 12:59 AM

#

good enough

wicked notch Jan 2, 2024, 1:53 AM

#

Alright

#

with this RT sample new Iris (codename Retina) is back to the same functionality level of old Iris

#

...but since RT is fucking addictive I'll be messing around for a day or two bleakekw

delicate rain Jan 2, 2024, 1:57 AM

#

Or a year or few 🥹

glass sphinx Jan 2, 2024, 10:54 AM

#

delicate rain Or a year or few 🥹

ptsd

wicked notch Jan 2, 2024, 8:42 PM

#

it truly is, addictive

wispy spear Jan 2, 2024, 8:47 PM

#

that does look neat

#

is this real time still?

wicked notch Jan 2, 2024, 8:47 PM

#

maybe, if 40ms counts as realtime

#

wanna try it on your 4070?

wispy spear Jan 2, 2024, 8:48 PM

#

sure : >

#

Regina still?

#

i have the bistro gltf without the interior

wicked notch Jan 2, 2024, 8:49 PM

#

ye let me 🅱️ush

wispy spear Jan 2, 2024, 9:07 PM

#

are you pushing bistor too? 😛 i do have it here already

wicked notch Jan 2, 2024, 9:08 PM

#

mayhaps

#

pushing incoming in 60 seconds

wispy spear Jan 2, 2024, 9:09 PM

#

github will not be happy about it

#

should probably be on gitignore

wicked notch Jan 2, 2024, 9:12 PM

#

alright 🅱️ushed

#

if you want textures you should put bistro through gltfpack

#

Otherwise https://drive.google.com/file/d/1zgDmlXBoRdkNDbmGDYJO2IKGTYvpQQ7O/view?usp=sharing

wispy spear Jan 2, 2024, 9:14 PM

#

fuck

#

can you provide the gltf pack params necessary for this postprocess?

wicked notch Jan 2, 2024, 9:15 PM

#

sure

#

gltfpack -v -noq -kn -km -ke -ac -tc -tq 10

wispy spear Jan 2, 2024, 9:15 PM

#

a klingon love poem

#

sorry for calling Retina Regina, no idea why i thought it was the latter : (

#

noice gltfpack uses all the cores, thx zeux

#

wicked notch Jan 2, 2024, 9:26 PM

#

why is everything so dark froghorror

wispy spear Jan 2, 2024, 9:26 PM

#

it feels smooth though

#

wicked notch Jan 2, 2024, 9:28 PM

#

can you try commenting out the section that grabs the textures

wispy spear Jan 2, 2024, 9:28 PM

#

sure

wicked notch Jan 2, 2024, 9:28 PM

#

it's in the rchit shader

wispy spear Jan 2, 2024, 9:31 PM

#

#

i just commented line 230 out

wicked notch Jan 2, 2024, 9:31 PM

#

something's wrong with how I shrimple textures maybe

wispy spear Jan 2, 2024, 9:31 PM

#

gpu runs at 144W

#

do you want a little video fly through too?

#

or is that image enough?

twin bough Jan 2, 2024, 9:32 PM

#

I'll test this too

wicked notch Jan 2, 2024, 9:33 PM

#

revert back to where it was and comment this section only https://github.com/LVSTRI/Retina/blob/master/Shaders/RayTracing/Main.rchit.glsl#L106-L116

wispy spear Jan 2, 2024, 9:33 PM

#

one sec

wispy spear Jan 2, 2024, 9:33 PM

#

twin bough I'll test this too

dont forget to add -i bistro.gltf -o bistro.glb 😛 for the gltfpack line

#

wicked notch Jan 2, 2024, 9:36 PM

#

alright so normal mapping is borked somehow

#

epic

wispy spear Jan 2, 2024, 9:37 PM

#

#

something is tanking my pc atm

wicked notch Jan 2, 2024, 9:37 PM

#

ah shit

#

I know why it's borked

wispy spear Jan 2, 2024, 9:37 PM

#

when i alt tabbed the first time from the running retina 😄

wicked notch Jan 2, 2024, 9:37 PM

#

it's because your the original gltfpack doesn't do normal maps properly

#

frick

wispy spear Jan 2, 2024, 9:37 PM

#

oh @ normals

#

200 out of 215W

wicked notch Jan 2, 2024, 9:38 PM

#

wicked notch Otherwise <https://drive.google.com/file/d/1zgDmlXBoRdkNDbmGDYJO2IKGTYvpQQ7O/vie...

if you use this bistro it should be good with normal mapping too

#

I think

wispy spear Jan 2, 2024, 9:38 PM

#

alright

wicked notch Jan 2, 2024, 9:38 PM

#

wispy spear 200 out of 215W

RT makes the GPU sweat yes

wispy spear Jan 2, 2024, 9:38 PM

#

yeah and much less vram usage too it seems just 1.5gb

#

interesting my bistro.glb is 2xx mb

#

yours is 3xx mb

wicked notch Jan 2, 2024, 9:39 PM

#

it has interior too I think

wispy spear Jan 2, 2024, 9:39 PM

#

ah

#

i can feel the warmth creeping up from under the desk xD

delicate rain Jan 2, 2024, 9:42 PM

#

It's addictive to look at too!! Looks awesome 😍

wispy spear Jan 2, 2024, 9:42 PM

#

#

hmm looks rather flat

#

i believe i commented the rchit in again and rebuilt everything

wicked notch Jan 2, 2024, 9:44 PM

#

it compiles shaders at runtime so there's no need to rebuild

wispy spear Jan 2, 2024, 9:44 PM

#

ah

#

can you check that the road is suppoed to look like that?

wicked notch Jan 2, 2024, 9:45 PM

#

I think it is

wispy spear Jan 2, 2024, 9:45 PM

#

ok

wicked notch Jan 2, 2024, 9:45 PM

#

I think my cosine weighted distribution is wrong

wispy spear Jan 2, 2024, 9:45 PM

#

sun angle might just be in a weird angle 🙂

wicked notch Jan 2, 2024, 9:45 PM

#

sun position is in the ray generation shader

#

I hardcoded it because I wanted to see shadows asap KEKW

wispy spear Jan 2, 2024, 9:46 PM

#

you should set the spawn point outside though

#

first time i thought it dowsnt work because black screen 😛

wicked notch Jan 2, 2024, 9:46 PM

#

now that I'm done bringing this new iris to full functionality I'll make the actual engine

#

first things first I gotta read what elias daler shared with me

wispy spear Jan 2, 2024, 9:47 PM

#

: )

#

boop me if you need it somehow tested

pale horizon Jan 2, 2024, 10:48 PM

#

Iris/Retina confirmed 99% stolen from me

wicked notch Jan 2, 2024, 10:48 PM

#

true and real

pale horizon Jan 2, 2024, 10:52 PM

#

Don’t read ideas.md if you don’t want the game spoiled KEKW

wicked notch Jan 4, 2024, 4:28 PM

#

It's quite incredible that the FBX importer/exporter is so garbage that it can work at full speed even when there's a gigabillion page faults happening

#

(it has to load the entire FBX in memory so I'm currently using 170GB of virtual memory with a working set of 32GB KEKW )

#

Ah yes 178772237 page faults and counting

wicked notch Jan 4, 2024, 4:56 PM

#

one must imagine Unreal users happy

glass sphinx Jan 4, 2024, 4:57 PM

#

wicked notch one must imagine Unreal users happy

they dont have time to be happy, they are waiting for startup and or compliation

wicked notch Jan 4, 2024, 8:08 PM

#

ffffffffuuuuuuuck

#

Why in the fucking hell would you use int to store the number of triangles of a model

wispy spear Jan 4, 2024, 8:20 PM

#

break the model into chunks perhaps : (

wicked notch Jan 4, 2024, 8:25 PM

#

I'm still waiting for the export to be done

#

Only now I've noticed that FBX2glTF can handle only 2 billion triangles

#

if this fails I'm gonna be so pissed

wispy spear Jan 4, 2024, 8:26 PM

#

understandable, so would i

#

how long is it running already?

wicked notch Jan 4, 2024, 8:27 PM

#

wicked notch It's quite incredible that the FBX importer/exporter is so garbage that it can w...

since here 💀

wispy spear Jan 4, 2024, 8:27 PM

#

fuck

wicked notch Jan 4, 2024, 8:27 PM

#

btw we're up to 400 million page faults KEKW

wispy spear Jan 4, 2024, 8:27 PM

#

: >

#

whats the PF delta?

wicked notch Jan 4, 2024, 8:27 PM

#

Honestly Windows is handling this incredibly well

wispy spear Jan 4, 2024, 8:27 PM

#

(its another column you can enable)

wicked notch Jan 4, 2024, 8:28 PM

#

ye right now it's small

wispy spear Jan 4, 2024, 8:28 PM

#

~1k ?

wicked notch Jan 4, 2024, 8:28 PM

#

yes

#

bouncing between 500 and 6000

wispy spear Jan 4, 2024, 8:28 PM

#

so more doing IO then

wicked notch Jan 4, 2024, 8:28 PM

#

I/O to page files exclusively bleakekw

wispy spear Jan 4, 2024, 8:28 PM

#

: >

wicked notch Jan 4, 2024, 8:29 PM

#

Anyways, glad to see that Window's memory management systems are working properly

#

I went into this fully expecting to crash, but the system's been running smoothly for hours now

wispy spear Jan 4, 2024, 8:30 PM

#

knock on wood

#

windows professional?

wicked notch Jan 4, 2024, 8:31 PM

#

yeah

#

key was cheap

#

since Windows 10

wispy spear Jan 4, 2024, 8:31 PM

#

next time say something

wicked notch Jan 5, 2024, 5:25 PM

#

ah yes

#

allocating a 3.6GiB vertex buffer

#

for a single mesh

wispy spear Jan 5, 2024, 5:33 PM

#

UnsignedLongLongBuffer

wicked notch Jan 5, 2024, 5:35 PM

#

I think it's time to call the Darian

delicate rain Jan 5, 2024, 5:38 PM

#

Just download more VRAM from google

wispy spear Jan 5, 2024, 8:35 PM

#

lustri, you should blog about your stuff and progress

#

and let epic games hire you

wicked notch Jan 5, 2024, 11:39 PM

#

alright time to document FBX sdk adventures

#

numero uno

#

FbxStream is a class responsible for opening/reading/writing files

#

it has a pure virtual function that opens a file virtual bool Open(void* pStreamData) = 0;

#

now, which file is this function going to open

#

nobody fucking knows

wispy spear Jan 5, 2024, 11:42 PM

#

: D

wicked notch Jan 5, 2024, 11:42 PM

#

oh wait I get it

#

I'm supposed to put the filename in the constructor

#

and then Open just resets the file pointer to 0

#

this is great

wispy spear Jan 5, 2024, 11:43 PM

#

lol wat

#

who writes apis like that

wicked notch Jan 5, 2024, 11:43 PM

#

look at this poor excuse of an example

#

https://help.autodesk.com/view/FBX/2020/ENU/?guid=FBX_Developer_Help_cpp_ref_file_stream_2main_8cxx_example_html

#

FBXSDK_strcpy(mFileName, 30, "CustomStreamClass_temp.txt");
hmm yes very clear thank you

#

this tells me exactly what I have to do

wispy spear Jan 5, 2024, 11:44 PM

#

the heck

#

there are 2 open sauce variants of fbx

#

ufbx i believe is the one and openfbx the other

wicked notch Jan 5, 2024, 11:45 PM

#

I looked at them

#

they're not multithreaded and they can't handle big files

#

everyone seems to like 4GiB max

wispy spear Jan 5, 2024, 11:46 PM

#

ah

#

: (

#

and autodesks' fbx can handle more?

wicked notch Jan 5, 2024, 11:47 PM

#

apparently so

#

FBX2glTF did manage to export it

#

after I changed every single uint32_t in the codebase to a size_t

wispy spear Jan 5, 2024, 11:48 PM

#

...
        int offset = (int)pOffset;
        switch (pSeekPos)
        {
        case FbxFile::eBegin:
            fseek(mFile, offset, SEEK_SET);
...

wicked notch Jan 5, 2024, 11:48 PM

#

the slow part is surprisingly not the FBX SDK

wispy spear Jan 5, 2024, 11:48 PM

#

also just int offset

#

ah

wicked notch Jan 5, 2024, 11:48 PM

#

ye you can define custom streams

#

Like I'm doing

wispy spear Jan 5, 2024, 11:48 PM

#

ok

wicked notch Jan 5, 2024, 11:48 PM

#

I'm not sure if ufbx can too

#

maybe I'll try it

wispy spear Jan 5, 2024, 11:48 PM

#

writing a custom export plugin might probably still a better idea

wicked notch Jan 5, 2024, 11:49 PM

#

yes

wispy spear Jan 5, 2024, 11:49 PM

#

even if you just dump everything obj style into a txt file heh

wicked notch Jan 5, 2024, 11:49 PM

#

but figuring out unreal doesn't seem fun bleakekw

wispy spear Jan 5, 2024, 11:49 PM

#

yah : (

#

hmm maybe there is some blender bridge plugin

finite quartz Jan 6, 2024, 12:04 AM

#

FBX SDK has been cursed by every developer that had to integrate it. So don't forget the tradition.

wispy spear Jan 6, 2024, 12:05 AM

#

https://tenor.com/view/thomas-train-choo-cursed-gif-26158193

Tenor

wicked notch Jan 6, 2024, 12:21 AM

#

Alright

#

near 50% speedup already

#

thank god they put in zero effort, they make me look like a good programmer

delicate rain Jan 6, 2024, 12:22 AM

#

LVSTRI is the city scene you posted gltf or FBX?

wicked notch Jan 6, 2024, 12:23 AM

#

gltf

#

I'd never post anything FBX

#

it doesn't deserve taking my precious SSD space even

delicate rain Jan 6, 2024, 12:24 AM

#

And you say it's ~9 gigs?

wicked notch Jan 6, 2024, 12:24 AM

#

the FBX or the gltf?

delicate rain Jan 6, 2024, 12:25 AM

#

Gltf

wicked notch Jan 6, 2024, 12:25 AM

#

FBX is 11GiB, gltf is 500MiB of JSON + 7.5GiB of binary

delicate rain Jan 6, 2024, 12:25 AM

#

That might actually fit in my VRAM

wicked notch Jan 6, 2024, 12:25 AM

#

You need something slightly above 8GiB

#

unfortunately I have a garbage 3070

delicate rain Jan 6, 2024, 12:26 AM

#

Hmmmm bleakekw I have 1080

frank sail Jan 6, 2024, 12:26 AM

#

Just compress

wicked notch Jan 6, 2024, 12:26 AM

#

mate

delicate rain Jan 6, 2024, 12:26 AM

#

We do some index compression

wicked notch Jan 6, 2024, 12:26 AM

#

the index buffer is 3GiB

#

you can compress the index buffer?

delicate rain Jan 6, 2024, 12:26 AM

#

Palette compressing the indices to 8 bits if I understood correctly

wicked notch Jan 6, 2024, 12:26 AM

#

damn

delicate rain Jan 6, 2024, 12:26 AM

#

Microindices

#

Meshoptimizer does it iirc

#

But I have no binary gltf loading and no block compressed texture loading setup yet

#

But I wanna see if Patrick's culling can handle it

wicked notch Jan 6, 2024, 12:28 AM

#

it's not a glb

#

it can't be bleakekw

#

also it has no textures

#

It's 7GiB of pure geometry

delicate rain Jan 6, 2024, 12:29 AM

#

EASY

#

Gigachad model, I'll try tomorrow to load it

wicked notch Jan 6, 2024, 12:30 AM

#

I'm still waiting for Darian to make moana island into a gltf

#

(he ain't gonna do it KEKW )

delicate rain Jan 6, 2024, 12:30 AM

#

Bro I wanted to do that

#

And gave up

#

Its so much

wicked notch Jan 6, 2024, 12:31 AM

#

yeah..

delicate rain Jan 6, 2024, 12:31 AM

#

I opened part of it in blender but yeah

#

Out of core rt sure

#

But real time I gave up bleakekw

wicked notch Jan 6, 2024, 12:32 AM

#

I just gotta figure nanite out

#

I'm making good progress

frank sail Jan 6, 2024, 12:32 AM

#

How difficult would it be to modify gltf to support >4gb files

wicked notch Jan 6, 2024, 12:32 AM

#

zero effort

#

just find any exporter that uses 8 byte integers instead of 4

#

cgltf and fastgltf come to mind

frank sail Jan 6, 2024, 12:32 AM

#

@ sean

wicked notch Jan 6, 2024, 12:33 AM

#

it wouldn't be spec compliant

#

but fuck the spec

#

KEKW

delicate rain Jan 6, 2024, 12:33 AM

#

wicked notch I just gotta figure nanite out

Nanite stinks, I'm all tessellation and displacement team now

#

Patrickpilled

wicked notch Jan 6, 2024, 12:33 AM

#

I will prove you wrong

#

eventually

#

in the far future bleakekw

delicate rain Jan 6, 2024, 12:34 AM

#

I gladly will be, but geometry stuff too scary for me

wicked notch Jan 6, 2024, 12:34 AM

#

bro

#

I'm dying

#

but I won't let fear take over me

#

I just have to grow a brain big enough to store all the nanite knowledge

delicate rain Jan 6, 2024, 12:34 AM

#

Compute tessellation my beloved 🥹

#

Join the gang

wicked notch Jan 6, 2024, 12:36 AM

#

I'm too deep now

#

it's already over for me

delicate rain Jan 6, 2024, 12:37 AM

#

https://tenor.com/view/we-lost-him-charlie-hudson-hudson-and-rex-he-shook-us-off-lost-our-trail-gif-23973533

Tenor

twin musk Jan 6, 2024, 12:54 AM

#

wicked notch you can compress the index buffer?

retvrn to triangle strips

#

no need for a horrible index buffer using all your memory :^)

wicked notch Jan 7, 2024, 12:16 AM

#

this is painful

#

I need a mesh small enough so that I can debug it easily

#

but it should also be big enough for me to be able to actually generate meshlets out of it

#

ok sphere it is

#

sphere is good

wispy spear Jan 7, 2024, 12:19 AM

#

rip cubes

wicked notch Jan 7, 2024, 12:20 AM

#

cubes have 12 faces

wispy spear Jan 7, 2024, 12:20 AM

#

can you use deth's doom.wad?

wicked notch Jan 7, 2024, 12:20 AM

#

not gud

wispy spear Jan 7, 2024, 12:20 AM

#

i know cubes are not good

wicked notch Jan 7, 2024, 12:20 AM

#

wispy spear can you use deth's doom.wad?

that has thousands of triangles bleakekw

wispy spear Jan 7, 2024, 12:20 AM

#

ah too much

wicked notch Jan 7, 2024, 12:20 AM

#

I do 128 triangles max

#

which is already painful to debug but manageable

wispy spear Jan 7, 2024, 12:21 AM

#

then some icosahedron-esque shape ye

wicked notch Jan 7, 2024, 12:21 AM

#

low poly sphere KEKW

wispy spear Jan 7, 2024, 12:21 AM

#

add subdivs

wicked notch Jan 7, 2024, 1:13 PM

#

assuming my code is correct

#

(huge assumption)

#

these are the meshlets for suzanne

#

how the hell do I validate this

#

@frank sail if I give you vertex position in a txt file (ready to be copy pasted in C++) can you try loading it in frogfood and checking if the meshlets are correct? KEKW

wispy spear Jan 7, 2024, 1:22 PM

#

hmm could you feed it as vertex positions into fastgltf and have it write a gltf?

wicked notch Jan 7, 2024, 1:23 PM

#

oh yeah I could write a GLTF

#

let me try

#

after this I really gotta go study otherwise my exam tomorrow is gonna go baad

wispy spear Jan 7, 2024, 1:24 PM

#

lol

wicked notch Jan 7, 2024, 1:24 PM

#

it's not my fault

#

ok

#

graph partitioning is cool once you understand it

wispy spear Jan 7, 2024, 1:24 PM

#

its UEs fault for taking so long

wicked notch Jan 7, 2024, 1:24 PM

#

(I still don't understand it but I can fumble my way through at least)

#

the meshlets being all the same size is sus..

#

ffs writing a gltf file is too involved

#

for (auto i = 0; i < meshlets.size(); ++i) {
    const auto& meshlet = meshlets[i];
    auto accessor = new cgltf_accessor();
    accessor->type = cgltf_type_vec3;
    accessor->component_type = cgltf_component_type_r_32f;
    accessor->offset = 0;
    accessor->stride = sizeof(SVertex);
    accessor->count = meshlet.size();
    accessor->buffer_view = &meshletBufferViews[i];
    auto attribute = new cgltf_attribute();
    attribute->type = cgltf_attribute_type_position;
    attribute->data = accessor;
    auto primitive = new cgltf_primitive();
    primitive->type = cgltf_primitive_type_triangles;
    primitive->attributes_count = 1;
    primitive->attributes = attribute;
    auto mesh = cgltf_mesh();
    mesh.primitives_count = 1;
    mesh.primitives = primitive;
    meshes.emplace_back(mesh);
}``` who said memory leaks weren't useful

wicked notch Jan 7, 2024, 2:06 PM

#

📎 sphere_meshlet.glb 📎 suzanne_meshlet.glb

#

I just want to see the draw calls

#

they are both vertex only

#

so no index buffer

wispy spear Jan 7, 2024, 2:09 PM

#

i cant load them with my loader : > it only accepts triangulated primitives 😢

wicked notch Jan 7, 2024, 2:12 PM

#

alright I have a hack

wispy spear Jan 7, 2024, 2:12 PM

#

please ignore me, my renderer wouldnt know if a meshwas indexless or not either

wicked notch Jan 7, 2024, 2:18 PM

#

alright time to engage overdrive mode

#

Imma write an OpenGL gltf viewer in 5 nanoseconds

wispy spear Jan 7, 2024, 2:30 PM

#

ah

#

nuh uh

#

fastgltf has a gltf opengl viewer

wicked notch Jan 7, 2024, 2:32 PM

#

true

#

let me try it

#

  Target "fastgltf_gl_viewer" links to:

    glfw::glfw```

#

bruh

#

ah gotta execute the python script myself

#

smh

wispy spear Jan 7, 2024, 2:41 PM

#

; (

wicked notch Jan 7, 2024, 2:42 PM

#

after fixing the build issue it crashes

#

probably can't handle non indexed meshes

#

well it doesn't matter

#

Iris is still there ready for us

#

alright moment of truth

#

welp

#

at least I see something

wispy spear Jan 7, 2024, 2:47 PM

#

ah you are trying to make to write a meshlet algorithm?

#

and want to see if it spits any meshlets out

wicked notch Jan 7, 2024, 2:47 PM

#

yes, using graph partitioning

wispy spear Jan 7, 2024, 2:47 PM

#

aaah

wicked notch Jan 7, 2024, 2:47 PM

#

hmm lots of empty draws

#

ah I'm very stupid

#

????

#

what in the frick is this

wicked notch Jan 7, 2024, 3:12 PM

#

holy

#

fucking shit

#

I've done it

#

#

it's done

wispy spear Jan 7, 2024, 3:12 PM

#

: O

#

meshlet looking meshlet scene

#

all these wolf faces everywhere

wicked notch Jan 7, 2024, 3:13 PM

#

it's over

#

I did it

wispy spear Jan 7, 2024, 3:13 PM

#

time to study

summer gyro Jan 7, 2024, 3:13 PM

#

wicked notch

coool

wispy spear Jan 7, 2024, 3:13 PM

#

we celery later

wicked notch Jan 7, 2024, 3:14 PM

#

hold up

#

I have to calm myself down first bleakekw

wispy spear Jan 7, 2024, 3:14 PM

#

: )

wicked notch Jan 7, 2024, 3:14 PM

#

Also

#

📎 the_thing.cpp

wispy spear Jan 7, 2024, 3:15 PM

#

i think you are the smartest frog among us

wicked notch Jan 7, 2024, 3:15 PM

#

so that I don't lose this

wispy spear Jan 7, 2024, 3:15 PM

#

wicked notch

wicked notch Jan 7, 2024, 3:16 PM

#

jesus..

#

we did it

wispy spear Jan 7, 2024, 3:16 PM

#

for for for if 😄

#

did potrick do something else?

wicked notch Jan 7, 2024, 3:23 PM

#

as far as I'm aware the only ones using METIS are:

#

me

#

unreal

#

that's it KEKW

#

alright NOW it is time to prepare for exams

wispy spear Jan 7, 2024, 3:25 PM

#

ah

#

= )

frank sail Jan 7, 2024, 7:54 PM

#

wicked notch <@288411937962721305> if I give you vertex position in a txt file (ready to be c...

Sure

#

Gltf is preferable

wicked notch Jan 7, 2024, 7:57 PM

#

we got the thing already

#

now it's time to make this actually work KEKW

#

this generates meshlets close to MAX_PRIMITIVES

#

but it doesn't actually respect the upper bound

#

and honestly, I have no idea how to make it work besides splitting the offending meshlet in 2

#

https://youtu.be/eviSykqSUUw?t=1172 sir brian says "managed to coerce it with small slack and fallbacks" though I have no idea what "slack" is or which fallbacks he used

wispy spear Jan 7, 2024, 7:59 PM

#

mayhaps run it by Suslik

#

see what he thinks

wicked notch Jan 7, 2024, 7:59 PM

#

suslik made his own clusterizer

#

like the god he is

wispy spear Jan 7, 2024, 7:59 PM

#

because he gave up on metis neh?

wicked notch Jan 7, 2024, 8:00 PM

#

I don't think he ever thought of METIS KEKW

wispy spear Jan 7, 2024, 8:00 PM

#

ah, i thought there was some conversation about it happening

#

possible that im mixing shit

wicked notch Jan 7, 2024, 8:00 PM

#

regardless, I think I'll make my own clusterizer too

#

METIS is too garbage

#

I wonder if I can abstract away the graph too

#

in CSR format

wispy spear Jan 7, 2024, 8:02 PM

#

took sir brian just a man-year

#

if you start now, we might have something by new years eve 2024/2025

wicked notch Jan 7, 2024, 8:03 PM

#

I'm more of a child than a man if we compare me to brian bleakekw

wispy spear Jan 7, 2024, 8:03 PM

#

true, but, its worth a try heh

#

although you are italian, you probabyl look more manly than him already

wicked notch Jan 8, 2024, 1:29 AM

#

TODO fix self reference

wicked notch Jan 8, 2024, 3:03 PM

#

clusterPrimitiveBuffer.emplace_back((face * 3 + 0) - minPrimitiveIndex);
clusterPrimitiveBuffer.emplace_back((face * 3 + 1) - minPrimitiveIndex);
clusterPrimitiveBuffer.emplace_back((face * 3 + 2) - minPrimitiveIndex);``` I have no idea what I'm doing ![KEKW](https://cdn.discordapp.com/emojis/666849321462792234.webp?size=128 "KEKW")

wicked notch Jan 8, 2024, 3:33 PM

#

Hmm maybe I'll stray away from meshoptimizer's design completely

#

but EXT_mesh_shader kinda requires it

wicked notch Jan 8, 2024, 6:39 PM

#

man

#

every second that passes I realize how big zeux's brain is and how small mine is in comparison agonyfrog

distant lodge Jan 8, 2024, 7:49 PM

#

its ok, zeux is a legend

#

maybe in 20-30 years we will reach his level

wicked notch Jan 8, 2024, 8:01 PM

#

I gotta use some more heuristics to make sure my partitions are topologically and spatially local

#

like using distance as edge weights

#

but there's still the problem of generating the micro index buffer

#

which I have no idea what it represents

#

(I don't have an idea about the vertex index buffer either bleakekw )

glass sphinx Jan 8, 2024, 8:25 PM

#

hmm

wispy spear Jan 8, 2024, 8:25 PM

#

maybe that's the "slack" sir brain was talking about 😄

glass sphinx Jan 8, 2024, 8:26 PM

#

the micro indices point into a per meshlet vertex index buffer
the per meshlet vertex index buffer points to actual vertices

wicked notch Jan 8, 2024, 8:26 PM

#

Yeah but what does it represent

#

for example, we know the index buffer is basically an half-edge list for a mesh graph

#

also with edge weights I get real funky meshlets bleakekw

glass sphinx Jan 8, 2024, 8:27 PM

#

https://tenor.com/view/mord-mit-aussicht-murder-with-a-view-bjarne-mädel-dietmar-schäffer-scheffer-gif-26112731

Tenor

#

too philosophical for me

#

froge_bleak

wicked notch Jan 8, 2024, 8:27 PM

#

yep I can get that

#

unfortunately unless I deeply understand things I can't reason with them

#

it's a big flaw of mine

glass sphinx Jan 8, 2024, 8:28 PM

#

i see

#

hm

wicked notch Jan 8, 2024, 8:28 PM

#

I think I'll start with a cube mesh

#

And look at what meshoptimizer outputs

#

just single meshlet, no funky business

wispy spear Jan 8, 2024, 8:30 PM

#

glass sphinx https://tenor.com/view/mord-mit-aussicht-murder-with-a-view-bjarne-m%C3%A4del-di...

deccer face & body reveal : (

glass sphinx Jan 8, 2024, 8:30 PM

#

froge_love

wispy spear Jan 8, 2024, 8:30 PM

#

i have my laptop on the belly most of the day : )

#

whats the difference between yours and potrick's, lvstri?

#

besides meshoptimizer

#

or are you saying potrick's way is shit?

#

or cant do something you want to do?

wicked notch Jan 8, 2024, 8:32 PM

#

wispy spear or cant do something you want to do?

this

#

I'm dumb

#

I can't partition a graph in a good way

#

Also I don't understand microindex buffers

#

How to generate them at least

wispy spear Jan 8, 2024, 8:34 PM

#

then we need to find people who can explain it better to you

wicked notch Jan 8, 2024, 8:43 PM

#

alright this looks promising

#

I changed the heuristic to (1.0 / (distance * distance + 1.0) * 2^20)

wispy spear Jan 8, 2024, 8:45 PM

#

how come the arches have no meshlet looking meshlets?

#

coincidence? or just not enough verticles?

wicked notch Jan 8, 2024, 8:45 PM

#

this is what I want

#

regular, equal sized, spatially local meshlets

wispy spear Jan 8, 2024, 8:46 PM

#

ah

#

das ist what i thought

#

another weird question

#

do you perhaps need models with properly made "source" meshes

#

before even processing them into meshlets

wicked notch Jan 8, 2024, 8:47 PM

#

not really, I can optimize them myself

wispy spear Jan 8, 2024, 8:47 PM

#

ok

wicked notch Jan 8, 2024, 8:47 PM

#

so long as they aren't completely degenerate

wispy spear Jan 8, 2024, 8:48 PM

#

do those arches have weird vertices compared to the curtains?

wicked notch Jan 8, 2024, 8:48 PM

#

yeah

#

meshes that have a big axis difference are a problem

wispy spear Jan 8, 2024, 8:49 PM

#

ah, weird

#

can you somehow "unwrap" meshes onto a "plane" and go from there?

wicked notch Jan 8, 2024, 8:50 PM

#

I don't think they're that much of a problem, just an annoyance

wispy spear Jan 8, 2024, 8:50 PM

#

hmm no, that doesnt even make sense

wicked notch Jan 8, 2024, 8:50 PM

#

knot looks good

wispy spear Jan 8, 2024, 8:50 PM

#

uniform bends everywhere : )

#

how do my complex cubes look like?

wicked notch Jan 8, 2024, 8:51 PM

#

I have not tried

#

and maybe you'll break my thing

#

because I can't really handle 1 sized partitions KEKW

wispy spear Jan 8, 2024, 8:51 PM

#

heh, not my intention

#

but it has rotated bits in it too

wicked notch Jan 8, 2024, 8:51 PM

#

oh, yeah

#

I don't take transforms into account at all yet

wispy spear Jan 8, 2024, 8:52 PM

#

SM_Deccer_Cube_Complex.gltf i think its called

wicked notch Jan 8, 2024, 8:52 PM

#

there's no transforms yet sadly

#

remember that I'm testing this on IrisGL bleakekw

#

some meshletisms end up squished

#

probably a tringle count problem

wispy spear Jan 8, 2024, 8:57 PM

#

: )

#

but has that typical shape

#

what is that mesh now? suzanna?

wicked notch Jan 8, 2024, 8:58 PM

#

the froge

wispy spear Jan 8, 2024, 8:58 PM

#

oh : )

wicked notch Jan 8, 2024, 8:58 PM

#

honestly the frog is a great test mesh

#

lots of meshlets, but it's single primitive

wispy spear Jan 8, 2024, 8:59 PM

#

📎 SM_Deccer_Cubes_Textured_Complex.gltf

#

textures dont matter i think

wicked notch Jan 8, 2024, 9:05 PM

#

I guess my deccer cubes role will be revoked now

wispy spear Jan 8, 2024, 9:06 PM

#

: D

#

that was unexpected

wicked notch Jan 8, 2024, 9:07 PM

#

yeah like I said, I don't handle transforms yet

wispy spear Jan 8, 2024, 9:07 PM

#

oki

#

i didnt parse that, it seems

wicked notch Jan 8, 2024, 9:07 PM

#

they're a hassle when you don't understand whatever the hell you're doing bleakekw

#

clion crashed

#

nevermind

wispy spear Jan 8, 2024, 9:09 PM

#

are you on EAP?

wicked notch Jan 8, 2024, 9:09 PM

#

nope

wispy spear Jan 8, 2024, 9:09 PM

#

kk, eap is quite unstable

wicked notch Jan 8, 2024, 9:09 PM

#

CLion.exe was using 100% of my CPU for whatever reason

#

I guess it didn't like me renaming a variable

wispy spear Jan 8, 2024, 9:09 PM

#

ah indexingisms

#Iris - A Journey through OpenGL and beyond to learn Graphics

nobody fucking knows