Rosy

[cloud rivet]

as in whether that's good for games or not, but there's a lot you can reload though already is my thought, but idk

[astral hinge]

all my scripting is in c++ so it's pretty useful still

but you are right otherwise

[cloud rivet]

so you want to separate out your code into dlls

and reload individual bits?

like make your game a bunch of dynamic libraries and a main process?

[astral hinge]

livepp requires minimal changes to the source (just one line at the start of main iirc). it works by hot patching the code

you change a function, recompile the TU, then livepp will patch all references to functions in that TU

[cloud rivet]

oh do you have to turn off memory address safety

[astral hinge]

or something like that

wdym

[cloud rivet]

if you try to read the code in memory where your asm is pointing to in the disassembler

it will not be readable

while debugging

[astral hinge]

pretty sure the debugger still works

[cloud rivet]

it shows you where in visual studio where each line of the assembly code is in memory

no I am saying the operating system prevents it

you have to like turn that off or something idk

[broken fog]

address randomization?

[cloud rivet]

no that's different

[broken fog]

oh

[astral hinge]

you do have to change some compiler flags to use livepp

[cloud rivet]

yeah

[broken fog]

idk how that all works tbh

[cloud rivet]

on the left

you see the disassembly

[cloud rivet]

there's an address next to each line

that's where the instruction actually exists in memory

if it is writable

you could write and change it directly in the debugger

because you can see your process's virtual memory

but you can't unless you turn the safety off

[astral hinge]

changing asm in the debugger sounds super cursed

[cloud rivet]

well that's what you're talking about sounds like to me

[broken fog]

still less cursed than self modifying code

[astral hinge]

I don't think this is what live++ is doing anyway

[cloud rivet]

it's writing to those places in memory

[astral hinge]

it's not writing to those places

[cloud rivet]

ok idk

what is it patching then?

it has to be memory though right

[astral hinge]

nvm it is writing

yeah it has to update references to the functions

[cloud rivet]

right

[cloud rivet]

well I got a windows window via window apis, time for vulkan init I guess

looks like I just pass in my app and window instance into this:

typedef struct VkWin32SurfaceCreateInfoKHR {
    VkStructureType                 sType;
    const void*                     pNext;
    VkWin32SurfaceCreateFlagsKHR    flags;
    HINSTANCE                       hinstance;
    HWND                            hwnd;
} VkWin32SurfaceCreateInfoKHR;

and call vkCreateWin32SurfaceKHR

this is easier than SDL tbh :/

not that SDL was hard, but it is a painful library to have build and keep around imo, just extra unecessary steps, and I never built rosy on anything but windows so it was a complete waste of my time to use it

[broken fog]

you... interacted with win32? willingly?

[cloud rivet]

yup yup

that's how it's going to go with this thing

[broken fog]

[cloud rivet]

ah I used std vector all over the place when initializing vulkan

        std::vector<VkLayerProperties> availableLayers(layerCount);
        vkEnumerateInstanceLayerProperties(&layerCount, availableLayers.data());

I guess I need to build that allocator

was looking at volk also

vulkan is kind of a shit API lol

[broken fog]

oh no

[cloud rivet]

jk jk

[broken fog]

you do not get to say that after using win32

[cloud rivet]

maybe I sacrifice BDA and just go with directx12

[broken fog]

weren't you writing c

[cloud rivet]

I am

it's all C

I was talking about Rosy

that's C++

[broken fog]

so what is cpeepee stl doing in your code

oh right

[cloud rivet]

I think I can maybe make the vkGetInstanceProcAddr song and dance a macro though

[broken fog]

i should learn vk

[cloud rivet]

well I need a way to allocate memory before I can even query anything

so I'm gonna do that I guess

[broken fog]

malloc?

what kinda allocator do you need

[cloud rivet]

yeah I'm going to write my own arena allocator though instead of manually managing memory all over the place

[true moon]

Honestly if you're comfortable using alloca a lot of Vulkan can be done on the stack

at least anything involving the pattern where you query a count of things and then query the things themselves

or if you don't want to do that you could even just allocate a worst-case stack array of them

[cloud rivet]

I thought about that

idk feels ugly

[true moon]

I've never really used alloca but if I wanted to avoid malloc that's probably what I'd do in this case

and I'd probably get used to it and stop finding it ugly

[cloud rivet]

you know very first ever C++ application I wrote was a vulkan implementation, which is kind of crazy

and now my first ever C application is also a vulkan implementation

I don't recommend a vulkan implementation as a way to learn programming languages tbh

[true moon]

lmao

[astral hinge]

only vulkan init has this

which I guess is a lot of vulkan if you're making hello triangle

[cloud rivet]

well once I have the allocator and the vulkan loader, the rest of this is going to be trivial

[true moon]

yeah between querying for stuff and passing stack-allocated createInfo structs you can get a long way

the allocation matters more for the actual rendering

[cloud rivet]

yeah, I am going to do my vulkan memory all from scratch, no vma too

this is all the fun part

[obsidian spade]

I wrote a stack allocator just so I wouldn't have to worry about this in my RHI lol

[true moon]

what sort of stack allocator

Like just a regular allocator with heap memory, but operating in a stack data structure?

or an allocator that actually runs on stack memory

[obsidian spade]

heap memory, stack data structure

[true moon]

gotcha

[obsidian spade]

It's fast enough. It's not why my code is slow, so I don't care to optimize it 🙂

(doesn't even show up on the flame graph)

And I'm hitting that region enough frequently enough to where it's not really causing (L2+) cache misses to hit that, so I'm fine with it as is. No need to fix what aint broken 🙂

[cloud rivet]

had some challenges with clangd linting + msvc building and warnings setting up a unity build but it works now, I learned about IWYU pragmas

the lsp has some trouble with understanding, obviously, that a c file that is included by another c file would know about certain names so I'm seeing if I can solve with a #pragma once in a header file that includes deps, I don't see other repos doing this so not sure how they handle it

they just seem happy including things at a top level c file and that does the job for them, probably something I haven't figured out yet

[cloud rivet]

the pragma once seems to work

cl /showIncludes /c main.c
main.c
Note: including file: C:\Users\Bjorn\projects\code\palinode\palinode.h
Note: including file:  C:\Program Files (x86)\Windows Kits\10\\include\10.0.26100.0\\um\windows.h
... more windows and stdio.h includes
Note: including file: C:\Users\Bjorn\projects\code\palinode\palinode.c

it did not then include all that windows stuff all over again

I also looked at the /E output you can see the full output of the translation unit's contents

idk why those other projects don't do that, but seems to work ok

the ai codereview bot hates my unity build, I am going to have to like figure out a directive to tell it not comment on it

I only have ai review my code, I don't use it to write anything

it has saved me from a lot of painful things I overlooked, ai codereview is very valuable

[cloud rivet]

clangd sucks

it's a pile of garbage

[cloud rivet]

I have to do a lot of extra work just to make clangd lsp work

I have to keep unecessary header files around with forward declarations

I have define a .clangd file that's basically a duplicate of my batch file

only clangd doesn't support environment variables https://github.com/clangd/clangd/issues/1827

so it's worse than a batch file

and I have to hard code the vulkan header location

which the rest of the world references via the environment variable the vulkan sdk itself adds

and their feedback is, have your build system generate the clangd

you know maybe I will have my batch file write the json

cool cool

[cloud rivet]

honestly I should be grateful clangd exists at all, because it does work

[cloud rivet]

that's a nice error message 🤗

[cloud rivet]

progress lol

[broken fog]

yea you usually have cmake generate it

why not just do that

[cloud rivet]

I'm not using cmake

I'm so happy

I have a little shoestring arena

#include "common.h"
#include "palinode.h" // IWYU pragma: keep

internal VkResult query_instance_layers(Arena *arena) {
    u32 property_count;
    VK_CHECK(vkEnumerateInstanceLayerProperties(&property_count, NULL));
    DEBUG_PRINT("num instance layers: %d\n", property_count);

    VkLayerProperties *layers;
    layers = arena_alloc(arena, sizeof(VkLayerProperties)*property_count);
    if (layers == NULL) return VK_ERROR_INITIALIZATION_FAILED;

    VK_CHECK(vkEnumerateInstanceLayerProperties(&property_count, layers));

    for (int i = 0; i < (int)property_count; i++) {
        VkLayerProperties layer = layers[i];
        DEBUG_PRINT("Instance layer name: %s instance layer description: %s \n", layer.layerName, layer.description);
    }

    return VK_SUCCESS;
}

VkResult init_gfx(void) {

    Arena *arena = init_arena(1024 * 1024 * 256);
    if (arena == NULL) return VK_ERROR_INITIALIZATION_FAILED;

    VK_CHECK(init_vulkan());
    VK_CHECK(query_instance_layers(arena));

    deinit_arena(arena);
    return VK_SUCCESS;
}

clangd keeps adding includes I don't need

I don't need a build system

definitely not making my project even worse than I have just for clangd

anyway

I can do vulkan init now

the hardest part has just been fighting this linter, not anything else

but I have it under control now

[brisk chasm]

oh noes, the pointer star went to the right : |

[cloud rivet]

it gotta go somewhere I guess

[astral hinge]

1024* 1024* 256

[cloud rivet]

with C, if you want some memory, you gotta go hat in hand. you don't get it for free, putting the astrisk on the right is just the beginning if you want yourself some memory to put things in

[brisk chasm]

as long as its not sometype * somevariable 😄

this is the wurst

[astral hinge]

texas style pointers

[cloud rivet]

it's just a sign of gratitude, that, that you're thankful, for all the memory

[astral hinge]

(because the star is lone)

[brisk chasm]

https://tenor.com/view/wink-winking-lone-star-lone-star-winking-horse-winking-gif-23190293

ah, all these good 80s and 90s series 🙂

[cloud rivet]

https://tenor.com/bcY4I.gif

[astral hinge]

https://tenor.com/view/he-man-fabulous-wink-blink-gif-21848333

[cloud rivet]

you gotta brand that memory, to show it's the right kind

right

it's that nice aligned memory, kept hot on the l1 and l2 cache, none of this page on disk memory that came down from the bus

it's super fresh

so the asterisks gotta go on the right i'm afraid

https://tenor.com/bUnGT.gif

[broken fog]

i hope you're right

yeah clangd without a build system is a pain

but also, batch files are a build system, just a shit one

(when are you installing lunix btw)

[brisk chasm]

i think we need a 2nd version of this, one with a little chimney, where smoke comes out, to visualize the jailee is cooking 😄

[slim oak]

and also a variant of this sticker with demongod in it, with explosions in the background

[brisk chasm]

and the bike is a foldable mortar of sorts

but the bike is crossed out on some panel right behind it

[cloud rivet]

When windows adds required code signing for executables and paid dev accounts like Apple

So probably never

Also then Id have to use clang also to compile and I would be even worse off

[brisk chasm]

i cant parse the first half of the windows signing stuff

[cloud rivet]

Try and ship a program that others can use on a mac

[brisk chasm]

ah mac

i dont get it, just compile your program on a mac and sign it there then? what has windows to do with that?

[cloud rivet]

https://developer.apple.com/documentation/security/notarizing-macos-software-before-distribution

[cloud rivet]

I was asked when I would install linux, and my reply was when Microsoft adds the same requirements for sharing software that Apple has

[brisk chasm]

emphasis on requirements i suppose

since you can sign programs ever since

[cloud rivet]

Well you have to pay like a hundred dollars per year to apple

You can’t just self sign

[brisk chasm]

actual code signing certs also cost quite a bit

lol i get it NOW

[cloud rivet]

I mean that would be one thing where I would definitely install linux that very day

I guess my UI will just be ASCII

[silver slate]

Dwarf Fortress time.

[cloud rivet]

that's my favorite game

[cloud rivet]

Microsoft's C documentation is amazing https://learn.microsoft.com/en-us/cpp/c-language/?view=msvc-170

it's easy to find things, it's easy to read, it has examples, it's so good

it goes into tons of detail, covers alignment, macros, the C runtime

how the compiler works

I think that and the open standard pdf are enough

no cppref required

[silver slate]

My eyes on that page. :')

[cloud rivet]

ah

I have cleared a visual path for me

so I know where to go

[silver slate]

but i wanna search and it's not doing anything

[cloud rivet]

yes

searching for anything C related is pointless on the internet

[astral hinge]

but cppref has it all

[cloud rivet]

cppref is unreadable

although

[silver slate]

Yes, but if this site is good, then why does it not work. 😭

[cloud rivet]

I'm sorry, I guess it's good for me, who clicks into things

:(

[silver slate]

I rarely ever use MSDN directly, usually I get to it via another search engine. :)

Because their ehm, navigation is problematic.

[cloud rivet]

also cppref is poluted with cpp details

[astral hinge]

there's a C section of cppref

[silver slate]

That's fair, I'm usually looking for specific things when I check the references, so I don't click around much.

[astral hinge]

if you look at c stdlib functions on the c++ part, then you'll get c++ info

[silver slate]

I'm sure there are other people who enjoy finding things via the MSDN site.

[cloud rivet]

ah you know, it's the search leads me to C++ stuff first, it does have a C only section

nice

[astral hinge]

ah

[cloud rivet]

ok another good resource

like the search for memset

it took me to C++ but then I looked further and saw there is a C result, but if I just browse the C section I can find what I want

see

evidence

searching for C is futile!

[slim oak]

I've been using it for the past ~13 years and it's been a very nice experience
they even added links to source code of the .NET a couple of years ago
For example:
https://learn.microsoft.com/en-us/dotnet/api/system.span-1?view=net-8.0
under Definition there is Source: Span.cs

[cloud rivet]

are you using DX12 for your project?

[slim oak]

DX11/12/Vulkan, DX12 and Vulkan are still WIP

[astral hinge]

man memset

[cloud rivet]

what is man

jk

[astral hinge]

a guy who tells you stuff

[cloud rivet]

C:\Users\Bjorn\projects\code\palinode>man memset
'man' is not recognized as an internal or external command,
operable program or batch file.

C:\Users\Bjorn\projects\code\palinode>

:<

[astral hinge]

if you install mingw it'll probably give you man

and a bunch of other stuff

[cloud rivet]

I was thinking about using DX12, since I am 100% windows only, but I cannot give up BDA :/

or slang

[cloud rivet]

I have WSL2 so I can use that

hey

that's great

I forgot I had that

more resources, thanks! keep em coming lol

I'll just keep this wsl2 ubuntu tab open

[silver slate]

grabs mirror and starts talking

[cloud rivet]

a short video of my code flow, showing errors, auto completion, and then build and test, everything is so fast and light now

just no problems not using Visual Studio

only thing is, if the app crashes I have to use the Visual Studio debugger

to capture the crash

it doesn't seem like other debuggers can do that

I tried it with both rad debugger and remedybg

but I can just type devenv palinode.exe and it will startup Visual Studio which is just feels like it takes forever, but there's no otherway I think, but it will capture the crash and tell me where it happened, which is like a must

ok ok, I want to start getting through vk init so I can render stuff

[brisk chasm]

you could start vs at the start already and just leave it in the background until you need it

[cloud rivet]

start the debugger you mean?

[brisk chasm]

just vs, so that you dont have to wait for it to start

[cloud rivet]

oh

[brisk chasm]

and then just attach the debugger when you need to later

[cloud rivet]

I'm not using vs I am using rad debugger

I would only use vs if I had a crash

[brisk chasm]

you were complaining about long vs startup times

[cloud rivet]

I mean

rad debugger is so fast it would take less time to launch rad debugger and hit f11 than it would to alt tab in to vs and do the same, and then rad debugger is just faster all around

idk

what I have determined is that rad debugger launches so fast, that it is actually faster to always quit rad debugger and just relaunch with my shortcut than it is to alt tab to rad debugger

[brisk chasm]

yes and when you need to capture a crash, you can use the already started vs which lingers in the background, is what i mean

[cloud rivet]

oh I see

yes

well hopefully I don't have any crashes

I am validating everything all over the place

[brisk chasm]

https://tenor.com/view/alexis-rose-schitts-creek-good-luck-fingers-crossed-cross-finger-gif-11400378180637421768

[cloud rivet]

aggressive validation:

internal VkResult query_instance_layers(AppContext *actx, Arena *arena) {
    if (!arena) fatal("query_instance_layers: null arena");
    if (!actx) fatal("query_instance_layers: null actx");
    if (!actx->gctx) fatal("query_instance_layers: null gctx");

    u32 property_count;
    VK_CHECK(vkEnumerateInstanceLayerProperties(&property_count, NULL));
    DEBUG_PRINT("num extension properties: %d\n", property_count);

    VkLayerProperties *layers;
    layers = arena_alloc(arena, sizeof(VkLayerProperties)*property_count);
    if (layers == NULL) return VK_ERROR_INITIALIZATION_FAILED;

    VK_CHECK(vkEnumerateInstanceLayerProperties(&property_count, layers));

    actx->gctx->layers_count = 0;
    for (int i = 0; i < (int)property_count; i++) {
        VkLayerProperties layer = layers[i];
        DEBUG_PRINT("Instance layer name: %s instance layer description: %s \n", layer.layerName, layer.description);
        for (int j = 0; j < num_default_layers; j++) {
            VK_STRCMP(default_layers[j], layer.layerName, {
                actx->gctx->layers_count += 1;
                break;
            });
        }
    }

    actx->gctx->layers = arena_alloc(actx->arena, sizeof(VkLayerProperties)*actx->gctx->layers_count);
    if (!actx->gctx->layers) fatal("Failed to allocate gfx layers");
    size_t current_layer = 0;
    for (int i = 0; i < (int)property_count; i++) {
        VkLayerProperties layer = layers[i];
        for (int j = 0; j < num_default_layers; j++) {
            VK_STRCMP(default_layers[j], layer.layerName, {
                if (current_layer >= (size_t)actx->gctx->layers_count) fatal("layer overflow");

                actx->gctx->layers[current_layer] = layer;
                current_layer += 1;
                break;
            });
        }
    }

    for (int i = 0; i < (int)actx->gctx->layers_count; i++) {
        VkLayerProperties layer = actx->gctx->layers[i];
        DEBUG_PRINT("Added instance layer: %s\n", layer.layerName);
    }

    return VK_SUCCESS;
}

no ub 🤞

this is my arena code so far:

Arena *init_arena(size_t min_size) {
    Arena *arena = NULL;
    arena = malloc(sizeof(Arena));
    if (!arena) fatal("Failed to init arena");

    size_t size = ALIGN_UP(CLAMP_MIN(min_size, ARENA_BLOCK_SIZE), ARENA_ALIGNMENT);

    arena->start = malloc(size);
    if (!arena->start) fatal("Failed to init arena data");
    void* expected_ptr = ALIGN_DOWN_PTR(arena->start, ARENA_ALIGNMENT);
    if (arena->start != expected_ptr) fatal("Allocated unaligned memory");

    arena->ptr = arena->start;
    arena->end = arena->ptr + size;
    return arena;
}

void deinit_arena(Arena *arena) {
    if (!arena) fatal("attempted to deinit a null arena");

    if (arena->start != NULL) {
        free(arena->start);
        arena->start = NULL;
        arena->ptr = NULL;
        arena->end = NULL;
    }
    free(arena);
}

#pragma warning(push)
#pragma warning(disable : 5045)
void *arena_alloc(Arena *arena, size_t size) {
    if (!arena) fatal("attempted to alloc from a null arena");

    void *ptr = NULL;
    ptr = arena->ptr;
    size_t available = arena->end - arena->ptr;

    // TODO: grow arena
    if (available < size) fatal("Arena overflow, available: %zu", available);

    // Align pointer
    arena->ptr = ALIGN_UP_PTR(arena->ptr + size, ARENA_ALIGNMENT); 

    // Validate pointer alignment
    if (arena->ptr > arena->end) fatal("Arena overflow after alignment, with available space: %zu", available);
    void* expected_ptr = ALIGN_DOWN_PTR(arena->ptr, ARENA_ALIGNMENT);
    if (arena->ptr != expected_ptr) fatal("Unexpected new pointer aignment");

    return ptr;
}
#pragma warning(pop)

5045 are spectre warnings, I get a ton of them whenever I compare anything related to pointers

[echo crystal]

nice

https://media.discordapp.net/attachments/851043005531815947/854980975326920724/image0-2-1.gif

what's the plan for this new project ?

[cloud rivet]

just do cool graphics stuff

:/

It's not a game™

[echo crystal]

i know I mean which stuff, if u know already

[cloud rivet]

my current todo list is (using just hard coded geometric shapes, no gltfs for now):

finish vk init
frame graph
instancing
animation
zoom/magnifier
MSAA
sky dome
env map
IBL!
Directional light
CSM Shadow
Point lights
Point shadows
AO

then I'll add textures and gltfs support I guess

[brisk chasm]

is that a C89 thing there?

void *ptr = NULL;
ptr = arena-ptr;
``` rather than
```c
void *ptr = arena->ptr;
```?

[cloud rivet]

I think it's the same after the compiler gets its hands on it

[brisk chasm]

yes, but

thats like 10 steps back

[cloud rivet]

how am I going to get those locs up without this

:|

[brisk chasm]

😄

[cloud rivet]

I will start with obj tbh I love obj

I will do gltf later I am so over asset stuff

[broken fog]

gltf sounds like a pain to nih ngl

[cloud rivet]

this is fully nih

everything from scratch

[broken fog]

yeah, fair enough

[cloud rivet]

oh you said pain to

[broken fog]

but the gltf spec is not very shrimple

[cloud rivet]

yeah

[broken fog]

it's not very convoluted either, tbf, but there's many levels of indirection

[cloud rivet]

yeah

it's cool though to render nice 3D stuff with PBR materials

so I will eventually do it

I'll actually start with a PBR roughness metallic workflow from the get go even for my geometric shapes

[broken fog]

make your own nih format

[cloud rivet]

I did that lol

that was my rsy asset format

is* I guess

I can alway go back to pick up where I left off if I ever wanted to

@broken fog do you have a project thread?

I thought I was following yours but I don't see it

[broken fog]

ye, https://discord.com/channels/318590007881236480/1332402779276316695

i haven't posted in forever though, so it's dead

[cloud rivet]

what are you working on?

[broken fog]

haven't had the time/motivation to work on it lately

[broken fog]

uni and work lol

[cloud rivet]

ah makes sense

do you do anything related for work or uni?

[broken fog]

would love to get back to platinum soon, hopefully

work is webdev which i've grown to hate lol but it pays the bills for now, in uni i get to do fun stuff at times tho

i actually added a little software rasterizer to my assignment for computer arch class this semester

assignment was to build a kernel so totally unrelated lol, actually super fun on its own but i though why not add 3d graphics to the vga driver

anyway yea trying to speedrun uni and get some money on top leaves little time for projects and i kinda burned out on it earlier this year as well

[cloud rivet]

oh that's cool

[broken fog]

but will get back to it soon™ cause honestly it was super fun

well except when i spent like a month debugging my bsdf

[cloud rivet]

not the burnout part, the project is cool I mean

[broken fog]

thanks heh

[cloud rivet]

my job is pure misery

[broken fog]

[cloud rivet]

one of the reasons I actually stopped rosy was to be able to be able to be more present in real life and less rushing to always work on my "game"

and also to have something I could do on my laptop

all this expensive software can only be on my pc

[broken fog]

sad, rosy was/is a very cool project as well

but hey, this new thing seems like it's gonna be a fun journey too

and by fun i mean probably a lot of type 3 fun

[cloud rivet]

I hope so!

[true moon]

You're using C, you don't need to do anything lol

You'll have more loc then you'll know what to do with

[cloud rivet]

Nice

[true moon]

As soon as you start getting into significant data structures, since it barely supports generic data structures so you'll have a lot of custom infrastructure per application

Even if you have a generic system that uses void* and a custom allocator (to avoid mallocing every node/whatever individually) you'll still have some extra glue to put in most likely

[cloud rivet]

I am going to pattern my storage after Sean Barrel stb stuff

Stretchy buffer and the other stuff

Since I have an allocator I always know exactly how much heap I have allocated without any need for introspection which is cool

Nothing will ever be allocated without my code

[broken fog]

or "templates at home" macro hell

[cloud rivet]

I'm not really into meta programming

I always take the 0.1x engineer more code approach

unless it becomes just too awful and a waste of time

I hate abstractions, a lot of engineers can't seem to wait to use latest thing, I'm not one of them, I like my imperative plain to read code. I went that way with Rosy, initially no helper functions, and then did a big, but actually easy, refactor that consolidated everything into really nice helper functions because by then I understood where they would be useful

that was really good I think starting with no helper functions, my rhi went from like nearly 10k lines of code to like 5k-6k

[echo crystal]

maybe u find this interesting https://caseymuratori.com/blog_0015

[broken fog]

inshallah we shall purge the oop brainrot 🙏

[cloud rivet]

I subscribe to Casey's substack

these 5 minutes I link to directly from his day 2 intro to C bascially explains why I am so excited about my current project https://youtu.be/KF29ePTqWa4?si=HaVRi33OxRCd42fI&t=4503

[cloud rivet]

the autosnapping is cool I want that in my UI

Like, literally, pretend you were a really great version of PKZip, running continuously on your code, looking for ways to make it (semantically) smaller.

And just to be clear, I mean semantically smaller, as in less duplicated or similar code, not physically smaller, as in less text, although the two often go hand-in-hand.

Like a good compressor, I don’t reuse anything until I have at least two instances of it occurring. Many programmers don’t understand how important this is, and try to write “reusable” code right off the bat, but that is probably one of the biggest mistakes you can make. My mantra is, “make your code usable before you try to make it reusable”.

that's a really good article

so one thing that I found valuable in my C++ though was I found private to be really useful for a particular exercise where I broke up my Editor.cpp into Editor.cpp and Level.cpp

and somewhere in a inbetween state Editor kept directly modifying level stuff because that's how it was originally when it was all the editor code

and eventually it just became such a huge problem that I just pulled the bits of data in Level.cpp that Editor was using and made them actually private

and that forced me to rewrite that code to be better

I could do that in C by using a forward declaration and defining something only in the C file I guess, but it was just a case where something being private forced me stop doing something bad, because enough was enough

I really had no idea what I was doing when I started on that editor code, so it was like an introduction to managing a bunch of assets different referencing to a bunch of things indirectly with colliding indices and I had to figure out how to make that work, and I had something that worked, but I still don't really feel great about the solution

throwing animation curves and key frames into the mix along with instances was like a bomb went off. I was so happy that I had instanced drawing done in like basically a day or two, and then I had 2 weeks of bugs to fix in the level editor as I realized more and more stuff had been broken

those were hard bugs too

[broken fog]

(although as the cancerous C++ specification continues to metastasize, it’s starting to add more options for this, but that is a separate issue)

[cloud rivet]

Spicy take: for a person who wants to make an indie game it would be really smart to:

1. Use an existing engine
2. Find some friends with a diverse set of talents to do it with, hopefully with more artists than engineers
3. Make it a small and short game, only make it bigger once you see people are actually willing to pay money to play it
4. Engineering is maybe the least important role on the team
5. Players do not care if you made your own engine custom, if you used C++ or Rust, Vulkan or OpenGL. They just want to play a fun game.

all those people who told me on the Game Dev League server that I was silly to make my own engine were in fact correct

[astral hinge]

assuming your goal is to make a financially successful game

[cloud rivet]

yes

[astral hinge]

I'm making a game I want to play, not necessarily one the market demands

[cloud rivet]

I think the people who are far along with their game on this server are the rare talented enough people to pull it off

a lot of the successful indie games that we all know about all built a growing community of fans early in the development process, fans that loved the games and those games spread through word of mouth,

[astral hinge]

I followed Minecraft development very closely in 2011. I'd check notch's tumblr for updates every day in my robotics class lol

[cloud rivet]

my first experience with minecraft was so much fun

https://github.com/romgrk/barbar.nvim looks cool

that's really nice

[gaunt drum]

I don't necessarily agree, depends on the game genre

I had many unsuccessful attempts at making games using engines, frameworks and spartan style

one might argue that that's the reason I don't have a game and I'm wasting time and people that used those engines and released something have some insight on it

though for comparison purposes, how valid is sampling from a pool of people that only experienced one side of the coin?

if they didn't make a, even a simple game, without any engine but they released a game using an engine, do that make that sample a valid one? not necessarily

engine or not, making games are hard. engine stashes that complexity under the rug. but that will leak, always and it is nasty to deal with

and never in any of my projects, I had the luxury of "oh, I'll just do the game and ignore rest of the stuff", it never worked that way

you always have to fight against the tool, framework, engine, lack of source code or lack of easily amendable source code, bad documentation, decades of tech debt and cruft

that's a solid trade-off, you should know what you are buying and what you are getting as a result

for single person projects, most projects one person can actually pull off and finish off (no early access lingering for years to come) don't require that much technical build up if you already have experience with making games to some degree

I think this is an important narrative to come against because there are many people gaslighting theirselves or (worse) other people into engines as if they are the absolute way to do stuff

so I'd argue what a game requires is a fun, executable game idea with a good loop. if you have that, you can do it however you want as long as you have clear idea about the process and have chance at financial success

[brisk chasm]

one could argue that this is bloat 😛

[echo crystal]

if u enjoy doing it, its not silly

[broken fog]

who needs tablines, open buffer search works great

[brisk chasm]

i dont think we are silly to make our own engines

its this kind of shit which interests us

everyone can make games, not everyone can make engines 😄 (big jk)

[gaunt drum]

too late

[brisk chasm]

as bjorn said, he might come back to this project 🙂

[gaunt drum]

there is no escape from NIH

[echo crystal]

why bother using a game engine?™

[brisk chasm]

exactly

[wraith urchin]

Guys we need to have a really inflammatory reddit post this year

[brisk chasm]

or we dont feed the trolls there

[wraith urchin]

It was good for the views

And the trolls didn't comment on the video for the most part

[slim oak]

just ask cybereality for some inflammatory reddit ideas

[brisk chasm]

noooooooooo

[cloud rivet]

I like it, I like being able to see current open buffers without having to do something without my fingers

why I gotta make my poor brain keep track of all that

or make these old hands have to type to know

without it, I'm constantly typing :ls and then I switch buffers

now I just see the list as tabs

[cloud rivet]

yes, also most of my favorite games are like made solo devs

[cloud rivet]

this is also very true

that games are hard

[wraith urchin]

Whatever you do, the bikeshed will come and find you

[cloud rivet]

me: man, making games is hard
also me: let me build renderer in C

although

void * is very nice, yes

everything is now void *

jk

I realized yesterday that I never did vk debug messages correctly in my previous projects, pMessages are UTF 8 strings

I never handled that

[broken fog]

list of open buffers in telescope tho

[cloud rivet]

that also requires typing

I just <leader>N buffer

[gaunt drum]

do the hard thing, declare feature freeze for two weeks

[broken fog]

fair enough, that's pretty nice

might give it a shot

how does it handle multiple panels

do you get one tabline per panel?

cause most tabline plugins are kinda ass with multiple views and that's why i don't use em much

[silver slate]

void* my beloved

[cloud rivet]

no, just one tab bar, it's just the list of open buffers you get with :ls so it's the same for all the panes

[broken fog]

oh

[cloud rivet]

man, the stuff people are sharing in #showcase is incredible

I need some pixels

[cloud rivet]

I love rad debugger tbh, so clean

^^ my VK_CHECK macro was wrong, it ran a failed thing multiple times and I figured it out from that

[astral hinge]

why don't you use a function for that

[cloud rivet]

-#define VK_CHECK(res) \
+#define VK_CHECK(x) \
 do { \
+    VkResult res = (x); \
     if (res != VK_SUCCESS) { \

[astral hinge]

macro moment 😩

[cloud rivet]

yeah I could

I use a function for fatal

I like keeping it inconsistent I guess

[astral hinge]

I just don't see the advantage of using a macro for this

but everyone uses one for some reason

[cloud rivet]

yeah idk

[brisk chasm]

macros is the reason why i dont want to write c

[cloud rivet]

idk you do you

it's all good

[cloud rivet]

@alpine basin talked me into using vcc for my shader in this project, so if I don't have pixels for a bit it's all their fault

[alpine basin]

uwu

[cloud rivet]

also starting with mesh shaders too :P

[echo crystal]

:0

[brisk chasm]

if you face any problems, then we have gobi right at our finger tips to fix shit 🙂

[cloud rivet]

yeah I mean it's all fun and learning, it's not a game™

[echo crystal]

https://tenor.com/view/puppet-hand-controlling-gif-17554100

[echo crystal]

it's basically a renderer right?

[brisk chasm]

of course, its all fun here

[cloud rivet]

yes to both

[cloud rivet]

hrm I will try push descriptors also

idk

this init will take a while

[astral hinge]

use bindless

[cloud rivet]

I do use bindless

but I still need a descriptor set

[astral hinge]

why push descriptorz

if you just have one set

[cloud rivet]

idk honestly, I'm just going to look to see if it's less work

if it's more work then I won't do it

rosy is still on vk 1.3 and was looking at v 1.4 features

[cloud rivet]

Current progress

[x] vkGetInstanceProcAddr;
[x] query_instance_layers(); 
[x] query_instance_extensions();
[x] init_instance();
[x] create_debug_callback();
[x] init_surface();
[x] init_physical_device();
[x] query_device_layers();
[x] query_device_extensions();
[x] init_device();
[ ] init_allocator();
[ ] init_presentation_queue();
[ ] init_swapchain();
[ ] init_draw_image();
[ ] init_descriptors();
[ ] init_command_pool();
[ ] init_command_buffers();
[ ] init_sync_objects();
[ ] init_commands();
[ ] init_shaders();
[ ] init_data();
[ ] init_default_sampler();

[astral hinge]

missing one checkbox
[ ] the_game();

[cloud rivet]

If I did make a game it would be like snake or tic tac toe or something

[astral hinge]

I think you said you just wanted to make a renderer which is a fine goal imo

[cloud rivet]

Yeah

Not making a game

Just if I was

My ambitions have plummeted to the floor on that

[cloud rivet]

I think I'm just going to render to the swapchain image at first

and then follow up with figuring out how to allocate memory for images after I have some pixels

I'm going to use a host allocator too

I want to allocate all of the memory my process uses

on the host and gpu

oh I wonder if vcc supports mesh shaders

Entry points #
Shader entry points need to be annotated with one of:

fragment_shader
vertex_shader
compute_shader
Also requires setting local_size(x, y, z)

nope

that's fine

[cloud rivet]

oh

the video says can be used for any vulkan shder stage

https://www.youtube.com/watch?v=SPWDLMc-9h4&t=21305s

that's a good talk

I'll try it

oh maybe I use vcc instead of msvc for the application 🤔

really cool stuff

[cloud rivet]

My first pass at allocator just so I can have a draw, depth and msaa image is just keep it simple and just over allocate and blow up if it ooms, but already have ideas for a more complex allocator that handles fragmentation and has good reallocation

I know I can just look it up, just want to try my ideas

[cloud rivet]

[x] vkGetInstanceProcAddr;
[x] query_instance_layers(); 
[x] query_instance_extensions();
[x] init_instance();
[x] create_debug_callback();
[x] init_surface();
[x] init_physical_device();
[x] query_device_layers();
[x] query_device_extensions();
[x] init_device();
[ ] init_allocator();
[x] init_presentation_queue();
[x] init_swapchain();
[ ] init_draw_image();
[ ] init_descriptors();
[ ] init_command_pool();
[ ] init_command_buffers();
[ ] init_sync_objects();
[ ] init_commands();
[ ] init_shaders();
[ ] init_data();
[ ] init_default_sampler();

tons of work left still

[cloud rivet]

Almost got to a clear color before I got too tired

[cloud rivet]

I always learn something new when I go through vk init

I will be making my own math library in C which I will also be able to use in my shader

I am on board with the vcc thing

A new level

I am thinking about using C to write C

That would help with the math and vk loading

[cloud rivet]

Had to work super late today

[true moon]

Me every day because I go to work late and then have to work late to get my hours in

It sucks I need to stop doing it

[cloud rivet]

I now have one pixel

gonna work on resize tomorrow, maybe a small animation, and then start on a vcc shader and see if I can get a triangle or something

if work permits

[cloud rivet]

it did not

I'm probably working late the rest of the week

[cloud rivet]

my next vulkan renderer after this one I will write in object oriented PHP

[gaunt drum]

where there is one there are many

[cloud rivet]

all I had time for today :/

start on handling resize tomorrow I guess

maybe idk

[cloud rivet]

This is rosy waiting for me when i had to work late last night

[cloud rivet]

my window is super dirty, need to clean it lol

[brisk chasm]

please knuddle rosy for me 🙂

[cloud rivet]

win32 is ezpz

I guess vcc triangle is next

I wasted so much time on SDL in the past when I never did a single thing that ever ran on non-windows, could have just been using the win32 all this time and no need for a gitsubmodule, dll nonsense

just following the how to instructions like a sheep

I am quarantining my win32 code though in case I ever did want cross platform, which I think is a better solution than just jumping right into SDL or glfw.

glfw is a lot less worse than SDL though

in terms of overall dependency burden

idk wasn't a huge deal, but I dont' have to worry about it

if I want Palinode to run on mac/linux I'm just going to write that code myself also

[cloud rivet]

look at that bright green strip since I started on the C renderer, it dimmed the rest of the year

[broken fog]

damn i wish my git history looked anything like that lol

this year it's been pretty fuckin barren

[echo crystal]

damn the consistency is impressive

this is mine

[cloud rivet]

Nice

This is all I do pretty much

I just work

[broken fog]

ah if that includes work then it's more understandable

if it's just your projects

[cloud rivet]

Yeah I work on open source projects hosted on GH for my job

lol

In fact I made all my personal projects private when I had my extreme anti social moment earlier this year and have kept them private

But you can still see my private contributions there

[hushed creek]

anyone know of an offline tool that generates that graph? I've been looking for one for ages

the closest I got to having one was when I had a gitea instance but I retired it ages ago

[wraith urchin]

You are using git right?

[hushed creek]

yes but not on github, and bitbucket doesn't show such graphs out of the box

[wraith urchin]

Yeah I'm also not using GH.

Would be interested

but tbh, the graph is not neccesarily a good reflection of work done

especially on personal projects

Its more of a reflection of how diligent you are at breaking tasks up into tiny commitable chunks

[hushed creek]

think it's more about streaks than how many commits you have per day

[wraith urchin]

Yeah but I often go 2-3 days without making a commit

But I'm still working

I just make one giant one after that time

My public GH is a disaster

[hushed creek]

that's kinda worrying to me I averaged something like 2 commits per day these past 3 years tho this includes a half year hiatus

I need my small commits to feel like nothing's going to go wrong XD

I was curious about the graph because I legit don't remember ever taking a single break this past 1 year to 1.5 yrs and I wondered what the streak would look like XD

[cloud rivet]

I make many small commits

I cannot deal with large diffs

My brain is too small for it

If it builds I commit

[wraith urchin]

For work stuff where someone is reviewing PRs, yeah large diffs suck. But for my personal stuff, I just get bogged down with the overhead of git stuff

[cloud rivet]

I use PRs on my own project

[wraith urchin]

You review them yourself?

[cloud rivet]

For the overall goal of what I am working on

[hushed creek]

bjorn if you keep self deprecating yourself or whatever the term is at each and every single step you'll eventually join deccer in the lame tombstone club

there are far more practical reasons for small commits than "my brain is too small for it"

[cloud rivet]

AI reviews it and I go over it too. I don’t use AI in my editor but I do value AI reviews

And I like seeing a PR to refer back to

To understand why I made a change

[hushed creek]

copilot has been doing real good at work, it's so amazing at generating and reviewing corporate code slop XD

[wraith urchin]

Isn't that what the commit message is for?

[hushed creek]

all the shit I'd have to write otherwise, copilot is like "no no allow me"

[cloud rivet]

Not for me

Also easy to git bisect to fix bugs

anti-AI sentiments will not age well imo. I don’t use AI for my personal code because I am having fun but my code at work goes BRR

[hushed creek]

anyway credit where credit is due, I shipped something that worked and had unusually many features, made a legacy class be testable and added a test suite to the project for it, all in one morning and wouldn't have happened without copilot (w/ claude sonnet) generating most of the boilerplate

I was of course gaslit into thinking there was a real deadline today, there wasn't but that's another story

[wraith urchin]

I didn't start out being anti-AI. The industry drove me to it

[hushed creek]

the point is before AI I would've said "yeah this will take at least two or three days"

[cloud rivet]

The reviews also frequently find real issues

[hushed creek]

the thing was even able to "understand" the underlying business logic when generating the tests, they weren't like, shallow or random

[cloud rivet]

That would have taken me a lot of misery and time to debug

There’s no way I am turning it off ever

[hushed creek]

but my favorite part was the fact that I was able to provide detailed logging because it kept generating good and accurate Console.WriteLine's for every single case

wish I could use it at home but I hate products with pay-per-usage that can balloon in costs

[cloud rivet]

I do not have these issues. I use Claude Code and my daily token spend is like sometimes $30 to $50 a day

[hushed creek]

fifty fucking dollars a day

[wraith urchin]

They made copilot usable for free to a certain extent

I actually used it for a few months

[hushed creek]

what a steal you cant even buy food with that money

I AM BEING SARCASTIC

[wraith urchin]

but then I turned all that shit off

[cloud rivet]

I am not using a restricted model with a $20 monthly plan

[hushed creek]

bjorn youre wasting your money hire someone to make your projects instead

[bronze socket]

yeah cut out the middleman and just hire the 3rd worlders that power the mechanical turk directly

[cloud rivet]

Idk my code works, the customer is happy, my employers and coworkers are happy, you can see my code it’s all open source

[wraith urchin]

Hire firefly

[hushed creek]

yeah hire me and I'll have claude generate the code

[wraith urchin]

I'm also against it for rule 8 reasons which I won't go into

[cloud rivet]

I didn’t write a single line of that

I vibe coded it with claude code

[bronze socket]

as a side note I hate the proliferation of the term vibe coding

[hushed creek]

me too

that said if you're getting paid for the code you're generating, then $50 is a steal

when I said I wouldn't pay for it I mean for dusk

[bronze socket]

yeah it's definitely a steal until the value gap closes

[hushed creek]

because if I had $50 to spend per day for dusk

I'd spend them on food and I wouldn't be having a job

I'd be working full time on it

when put into that perspective, $50 is an immense amount of money

[cloud rivet]

I am not paying it, it’s company money

[bronze socket]

is $18k/yr about median in the philippines or is it high

[hushed creek]

philippines?

[bronze socket]

I thought you were in the philippines, am I misremembering

[hushed creek]

nope I'm in romania, it's the country with the highest inflation in the EU and crazy prices relative to income

and $50 a day would still get me three VERY big meals

[bronze socket]

oh that's... close

[hushed creek]

it's just such an insane amount of money I just can't

[wraith urchin]

I though hungary was worse?

[hushed creek]

I don't think they have been for a while, but they're not far

[bronze socket]

eastern europe shithole competition

[wraith urchin]

Romanians like to dunk on Hungary right?

[hushed creek]

hmmm we don't dunk on our neighbors that much tbh but I guess there is some enmity there given our history

[cloud rivet]

Cost of living in the sf bay area is ridiculous

[hushed creek]

like how russia thinks about ukraine, some hungarians think about transylvania the same

[wraith urchin]

Oh we dunk on our neighbours all the time here. But I think they can handle it lmao

[cloud rivet]

Berkeley minimum wage is $19.58 per hour and it is not a livable amount of money

[hushed creek]

there's always been a trend here, for people near the border with hungary to go do groceries and get gas in hungary, because it's cheaper

[cloud rivet]

The cost of living is very high here

[hushed creek]

those near the border with bulgaria do the same

[cloud rivet]

You can’t pay for housing, much less food with minimum wage.

[hushed creek]

nobody comes to buy here because we're too expensive XD

[cloud rivet]

We deserve it though

Idk which neighbor you mean

[bronze socket]

CA definitely does

how many neighbors does canada even have

does greenland count?

[cloud rivet]

Us and denmark and Russia

I think

[bronze socket]

I guess russia across the north pole yeah

[cloud rivet]

Yeah

[true moon]

AI as a linter strikes me as potentially useful just because it's basically out of the development loop and simply tags things for closer inspection that you might not have looked at closely yourself, could catch security issues or something I guess

Then it can simply be ignored if you aren't interested in using AI tools

Unlike those horrendous "AI-first" policies where companies mandate that code is written with AI

[brisk chasm]

?

[wraith urchin]

Technically we have two now but yes I did mean you guys lmao

[wraith urchin]

We have a land border with Denmark & USA

[cloud rivet]

I might be in vancouver in 2026

software you can love is planning on a conference in 2026 but they haven't announced it yet

I don't write zig anymore but I still like the community

[wraith urchin]

It's a zig conference?

[cloud rivet]

not really

more like a systems programming conference

the last one hardly had any zig talks

but it is mostly zig people who go because it is strongly attached to the zig core dev team

it's similar to handmade

I would totally hire Firefly btw

I'm not a manager anymore so I don't make hiring decisions or know what the head count is but https://www.conductorone.com/careers/ and https://www.conductorone.com/docs/baton/faq/ <-- is what I build it is very boring stuff

[bronze socket]

boring is good

the software engineers make runs to tijuana?

[hushed creek]

https://tenor.com/view/we-dont-do-that-here-black-panther-tchalla-bruce-gif-16558003

[gaunt drum]

sometimes I abandon stuff before I start committing as I don't like starting to commit immediately when I start out and have a lot of momentum

after I replied to Bjorn several days ago, I had a self reflection on my failures and decided to go with godot and suck it up btw. I am still terminal NIH though, it is just a tactical retreat for the moment. will probably use godot as frontend after most of the gameplay becomes solid. in any case, thanks Bjorn

[cloud rivet]

godot seems really capable and awesome

if I want to build a game, an actual game, I would use an engine

I know that's heresy

it's still a lot of work to make a game even with an engine

players don't care how the game was made

I just realized I forgot to handle minimize

I was going to work on triangle, but I need to make sure I don't crash on minimize when I get home from work

I'm starting with instance drawing an animation from the very get go

shoehorning it in afterwards was painful with Rosy

like the assumption that there was only one thing and it was static, that was all over the code in surprising places

one thing of each type

[vagrant musk]

Yeah, player mostly just cares:

• game runs well
• game looks good (optional in some cases)
• game feels good (even if buggy, plus bugs can be beneficial to the game sometimes)

[astral hinge]

I think this is the best way to use LLMs for coding

Linus said they are basically an advanced linter and I agree

[cloud rivet]

I just asked Claude to check out an older release of a repo I'm working on and determine how a deprecated config flag works

it's going through all the code and doing that while I do something else

I can also set authorization tokens for services like AWS whatever for sandbox environments and ask it to generate client models and http request code for some API by giving it the api doc link

it'll make requests, test the apis, and write all the code for that

it's not just an advanced linter

it does all this while I am doing other things

I review the code and test it myself

to be AI adverse, to constrain it, to think it's just categorically bad because AI is being used bad in some places, it's just a mistake

and incorrect

I am not a braindead ai proponent, I actually think it's destructive and largely bad

but it makes my job easier, the tech works as advertised

I can run multiple instances of claude while it does this for different connectors I have in progress

it's ridiculous

there's no way i'm going to say no I'm not using this

and if I did, I'd get fired

Claude Output:

⏺ Key Differences Between v0.0.9 and Main

  v0.0.9 Implementation:
  - Had configurable sync-grant-sources flag (default: false)
  - Applied GrantExpandable conditionally only when flag was enabled
  - Used in pkg/connector/helper.go:146-155 for ACL processing
  - Added expansion for team/group member AND admin roles

  Current Main Implementation:
  - No configuration flag - grant sources are always synced
  - GrantExpandable is now always applied in two places:
    a. groups.go:103-110 - When group members are other groups
    b. projects.go:90-96 - For project access grants to groups

  What's the same:
  - Both use GrantExpandable annotation with EntitlementIds
  - Both target group membership expansion
  - Both use Shallow: true for some cases

  What's different:
  - Scope: v0.0.9 was broader (ACL-based), main is more targeted (group membership only)
  - Configuration: v0.0.9 was optional, main is always on
  - Implementation: v0.0.9 used ACL processing, main uses direct group membership APIs
  - Expansion types: v0.0.9 included admin roles, main focuses on member relationships

  So yes, you're essentially doing the same thing for group members now, but it's always enabled and more focused on direct group
  membership relationships rather than the broader ACL-based approach in v0.0.9.

that saved me so much fucking time

it's also correct

I'm gonna go back to talk about Palinode instead of work though 😅

[bronze socket]

yeah that sounds like a blessing

[cloud rivet]

it is

[gaunt drum]

indeed, unless someone is teenager, most people end up with NIH due to control fetish

and tool you use affects the way you do stuff. you can obviously fix all tool related issues, people are doing it all the time

[vagrant musk]

eh, I end up with NIH because the only game engine I actually like working with doesn't have the features I need for the games I want to make

[astral hinge]

I have NIH due to a phobia of using unreal

[gaunt drum]

one example that annoys me with godot + gdscript for example that you lack custom value types

as a result I code around it and design stuff around it. and will probably offload heavy work to rust with gdextension at some point

[vagrant musk]

like it gives me all the control I really need for the game it's just
I don't want to lay out like a million trees by hand and setting up density regions using objects is a pain, lol

[cloud rivet]

Lyz you use godot yes?

I remember you were sharing something like that

I think this server should be friendly to graphics programming using engines

I'm not sure

it's still graphics programming?

[echo crystal]

yes

[astral hinge]

It is if you're programming graphics

[alpine basin]

There's nothing in the rules against using pre-existing engines

It's true we have a culture of NIH but we've had quite a few people come through here while using existing engines

[astral hinge]

Proposal to instaban people that use an engine

Including a custom engine

[broken fog]

yes but consider

number go up

[wraith urchin]

Frankly I had not considered this, you are in fact correct

[cloud rivet]

the graph for me represents more my daily commitment, and not the total amount of work done. I get one commit in every day, and sometimes I just have a small commit

it's like a habit tracker

the hue of the color does tell me something about what that day was like for me

the days on which I have no work were either very good days or very bad days

[vagrant musk]

If you’re doing graphics stuff, yes
… unfortunately tree placement is not really graphics

The ideal way for me to do that would probably be to make my own extension, which is something I don’t really want to learn
Terrain 3d extension might have what I need by this point but idk; last I checked on it I don’t think supported foliage colliders, which are kinda important for trees

[cloud rivet]

Ah

[cloud rivet]

holy shit, my app doesn't crash when I minimized it, and I didn't have to do anything extra

hrm I do see a VVL though

I remember having to write code to handle minimized to not render during that moment

with SDL

which I should do anyway even here, but the app didn't crash

[cloud rivet]

idk, interesting

time for vcc finally

[astral hinge]

vichichi as deccy calls it

[cloud rivet]

fancy

that should be it for win32 apis I need to deal with, just have to handle keyboard and mouse events

don't plan to add any audio or anything

Vcc supports advanced C/C++ features usually left out of shading languages such as HLSL or GLSL, in particular raising the bar when it comes to pointer support and control-flow:

Unrestricted pointers
Arithmetic is legal, they can be bitcasted to and from integers
Generic pointers
Generic pointers do not have an address space in their type, rather they carry the address space as a tag in the upper bits.
True function calls
Including recursion, a stack is implemented to handle this in the general case
Function pointers
Lets you write code in a functional style on the GPU without limitations
Arbitrary goto statements - code does not need to be strictly structured !

nice

Non header-only libraries, including parts of the C and C++ standard libraries, will not work on the GPU without additional work, because of the way Vcc works.

no malloc/free, that's fine

Function and data pointers are not portable between host and device. Taking addresses and passing them across in either direction is UB - since Vulkan lacks a unified addressing extension this is a limitation we cannot address at this time.

hrm

so I'm going to write my own math library and use it for both the shader and for application and see how that plays out

I think that is doable given the constraints here

[cloud rivet]

hrm

:\Users\Bjorn\projects\code\shady\build>dir bin\Release
 Volume in drive C is OS
 Volume Serial Number is 5C1A-ED43

 Directory of C:\Users\Bjorn\projects\code\shady\build\bin\Release

07/11/2025  08:04 PM    <DIR>          .
07/11/2025  08:04 PM    <DIR>          ..
07/11/2025  08:04 PM           721,920 aobench_host.exe
07/11/2025  08:04 PM           717,312 checkerboard.exe
07/11/2025  08:04 PM        52,219,456 LLVM-C.dll
07/11/2025  08:04 PM           507,392 opt_oracle.exe
07/11/2025  08:04 PM           720,896 runner_test.exe
07/11/2025  08:04 PM           736,768 slim.exe
07/11/2025  08:04 PM           239,104 test_builder.exe
07/11/2025  08:04 PM            15,360 test_dict.exe
07/11/2025  08:04 PM           227,328 test_math.exe
07/11/2025  08:04 PM            15,872 test_util.exe
07/11/2025  08:04 PM           716,288 vcc.exe
              11 File(s)     56,837,696 bytes
               2 Dir(s)  379,791,679,488 bytes free

C:\Users\Bjorn\projects\code\shady\build\bin\Release>slim.exe --help
  --log-level debug[v[v]], info, warn, error]
  --shd_print-internal                          Includes internal functions in the debug output
  --shd_print-generated                         Includes generated functions in the debug output
  --no-dynamic-scheduling                   Disable the built-in dynamic scheduler, restricts code to only leaf functions
  --lift-join-points                        Forcefully lambda-lifts all join points. Can help with reconvergence issues.
  --execution-model <em>                   Selects an entry point for the program to be specialized on.
Possible values: Compute, Fragment, Vertex, RayGeneration, Callable,
  --target <c, glsl, ispc, spirv>
  --output <filename>, -o <filename>
  --dump-cfg <filename>                     Dumps the control flow graph of the final IR
  --dump-loop-tree <filename>
  --dump-ir <filename>                      Dumps the final IR
No target specified, defaulting to a generic one.
  --entry-point <foo>                       Selects an entry point for the program to be specialized on.
  --word-size <8|16|32|64>                  Sets the word size for physical memory emulation (default=32)
  --pointer-size <8|16|32|64>               Sets the pointer size for physical pointers (default=64)
  --subgroup-size N                         Sets the subgroup size the program will be specialized for.
  --use-native-tailcalls                    Sets the subgroup size the program will be specialized for.
  --use-native-fncalls                      Sets the subgroup size the program will be specialized for.
unknown filename extension '--help', interpreting as Slim sourcecode by default.Failed to read file '--help'

hrm

C:\Users\Bjorn\projects\code\shady\build\bin\Release>vcc.exe --help
  --log-level debug[v[v]], info, warn, error]
  --shd_print-internal                          Includes internal functions in the debug output
  --shd_print-generated                         Includes generated functions in the debug output
  --no-dynamic-scheduling                   Disable the built-in dynamic scheduler, restricts code to only leaf functions
  --lift-join-points                        Forcefully lambda-lifts all join points. Can help with reconvergence issues.
  --execution-model <em>                   Selects an entry point for the program to be specialized on.
Possible values: Compute, Fragment, Vertex, RayGeneration, Callable,
  --target <c, glsl, ispc, spirv>
  --output <filename>, -o <filename>
  --dump-cfg <filename>                     Dumps the control flow graph of the final IR
  --dump-loop-tree <filename>
  --dump-ir <filename>                      Dumps the final IR
  --entry-point <foo>                       Selects an entry point for the program to be specialized on.
  --word-size <8|16|32|64>                  Sets the word size for physical memory emulation (default=32)
  --pointer-size <8|16|32|64>               Sets the pointer size for physical pointers (default=64)
  --subgroup-size N                         Sets the subgroup size the program will be specialized for.
  --use-native-tailcalls                    Sets the subgroup size the program will be specialized for.
  --use-native-fncalls                      Sets the subgroup size the program will be specialized for.
Missing input file. See --help for proper usage

these seem like the same program

[cloud rivet]

it works

as in vcc works, I'm not rendering anything yet

#include <shady.h>

descriptor_set(0) descriptor_binding(1) uniform_constant sampler2D texSampler;

location(0) input native_vec3 fragColor;
location(1) input native_vec2 fragTexCoord;

location(0) output native_vec4 outColor;

fragment_shader void main() {
    native_vec4 tint = (native_vec4) { fragColor.x * 0.8f, fragColor.y * 0.8f, fragColor.z * 0.8f, 1.0f };
    outColor = texture2D(texSampler, fragTexCoord) * tint;
}

C:\Users\Bjorn\projects\code\palinode>vcc -I"C:\Users\Bjorn\projects\code\shady\build\share\vcc\include"   vcc\shader.c
clang version 19.1.1
Target: x86_64-pc-windows-msvc
Thread model: posix
InstalledDir: C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\Llvm\x64\bin
file=vcc\shader.c tmpfile=vaGDatonjiMM9B7F2yys0c1A6ZuTPKNC
built command: clang -c -emit-llvm -S -g -O0 -ffreestanding -Wno-main-return-type -isystem"C:\Users\Bjorn\projects\code\shady\build\bin\Release/../share/vcc/include/" -D__SHADY__=1 --target=spir64-unknown-unknown -Xclang -fpreserve-vec3-type -o vaGDatonjiMM9B7F2yys0c1A6ZuTPKNC "vcc\shader.c" "-IC:\Users\Bjorn\projects\code\shady\build\share\vcc\include"
Clang returned 0 and replied:
LLVM IR parsed successfully
Done

C:\Users\Bjorn\projects\code\palinode>spirv-val.exe vcc\output\shader.spv
C:\Users\Bjorn\projects\code\palinode>spirv-dis.exe vcc\output\shader.spv -o vcc\output\shader.spvasm
C:\Users\Bjorn\projects\code\palinode>spirv-cross.exe vcc\output\shader.spv --output vcc\output\shader_clean.glsl

that produced a valid spirv and this glsl via spirv-cross

#version 450
#if defined(GL_ARB_gpu_shader_int64)
#extension GL_ARB_gpu_shader_int64 : require
#elif defined(GL_NV_gpu_shader5)
#extension GL_NV_gpu_shader5 : require
#else
#error No extension available for 64-bit integers.
#endif

#if defined(GL_KHR_shader_subgroup_basic)
#extension GL_KHR_shader_subgroup_basic : require
#elif defined(GL_NV_shader_thread_group)
#extension GL_NV_shader_thread_group : require
#elif defined(GL_ARB_shader_ballot) && defined(GL_ARB_shader_int64)
#extension GL_ARB_shader_int64 : enable
#extension GL_ARB_shader_ballot : require
#else
#error No extensions available to emulate requested subgroup feature.
#endif

vec4 _57;

struct _100
{
    uint _m0;
    uint64_t _m1;
};

struct TreeNode
{
    uint64_t _m0;
    uint _m1;
};

struct _229
{
    uint _m0;
    uint _m1;
};

struct _266
{
    uint _m0;
    bool _m1;
};

struct _370
{
    uint _m0;
    TreeNode _m1;
};

struct JoinPoint
{
    TreeNode _m0;
    uint64_t _m1;
    uint _m2;
};

struct _478
{
    uint _m0;
    JoinPoint _m1;
};

layout(binding = 1) uniform sampler2D texSampler;

layout(location = 1) in vec2 fragTexCoord;
layout(location = 0) in vec3 fragColor;
layout(location = 0) out vec4 outColor;
shared TreeNode scheduler_vector[1][64];
shared TreeNode active_branch[1];
shared uint actual_subgroup_size[1];
shared uint64_t resume_at[1][64];
shared uint scheduler_cursor[1];
shared uint64_t next_fn[1];

#if defined(GL_KHR_shader_subgroup_basic)
#elif defined(GL_NV_shader_thread_group)
#define gl_SubgroupInvocationID gl_ThreadInWarpNV
#elif defined(GL_ARB_shader_ballot)
#define gl_SubgroupInvocationID gl_SubGroupInvocationARB
#endif

uint generated_init(vec4 outColor_physical, out uint _RESERVED_IDENTIFIER_FIXUP_68_physical, out vec3 fragColor_physical, out vec2 fragTexCoord_physical, vec4 _RESERVED_IDENTIFIER_FIXUP_837_physical, uint stack_ptr)
{
    _RESERVED_IDENTIFIER_FIXUP_68_physical = gl_SubgroupInvocationID;
    fragTexCoord_physical = fragTexCoord;
    fragColor_physical = fragColor;
    return stack_ptr;
}

uint main(out vec4 outColor_physical, uint _RESERVED_IDENTIFIER_FIXUP_68_physical, vec3 fragColor_physical, vec2 fragTexCoord_physical, vec4 _RESERVED_IDENTIFIER_FIXUP_837_physical, uint stack_ptr)
{
    vec4 _61;
    _61.x = fragColor_physical.x * 0.800000011920928955078125;
    _61.y = fragColor_physical.y * 0.800000011920928955078125;
    _61.z = fragColor_physical.z * 0.800000011920928955078125;
    _61.w = 1.0;
    outColor_physical = texture(texSampler, fragTexCoord_physical) * _61;
    return stack_ptr;
}

uint generated_fini(vec4 outColor_physical, uint _RESERVED_IDENTIFIER_FIXUP_68_physical, vec3 fragColor_physical, vec2 fragTexCoord_physical, vec4 _RESERVED_IDENTIFIER_FIXUP_837_physical, uint stack_ptr)
{
    outColor = outColor_physical;
    gl_Position = _RESERVED_IDENTIFIER_FIXUP_837_physical;
    return stack_ptr;
}

void main()
{
    vec4 outColor_physical;
    uint _RESERVED_IDENTIFIER_FIXUP_68_physical;
    vec3 fragColor_physical;
    vec2 fragTexCoord_physical;
    vec4 _RESERVED_IDENTIFIER_FIXUP_837_physical;
    uint stack_ptr = generated_init(outColor_physical, _RESERVED_IDENTIFIER_FIXUP_68_physical, fragColor_physical, fragTexCoord_physical, _RESERVED_IDENTIFIER_FIXUP_837_physical, 0u);
    uint stack_ptr_1 = main(outColor_physical, _RESERVED_IDENTIFIER_FIXUP_68_physical, fragColor_physical, fragTexCoord_physical, _RESERVED_IDENTIFIER_FIXUP_837_physical, stack_ptr);
    uint stack_ptr_2 = generated_fini(outColor_physical, _RESERVED_IDENTIFIER_FIXUP_68_physical, fragColor_physical, fragTexCoord_physical, _RESERVED_IDENTIFIER_FIXUP_837_physical, stack_ptr_1);
}

neat

cool af

time for that triangle

I can see the spirv too via vcc\output\shader.spvasm

disassembly I mean

When cross-compiling, certain identifiers are considered to be reserved by the implementation. Code generated by SPIRV-Cross cannot emit these identifiers as they are reserved and used for various internal purposes, and such variables will typically show up as RESERVED_IDENTIFIER_FIXUP or some similar name to make it more obvious that an identifier has been renamed.

idk what that was, so looked it up

i'm gonna go exercise then work on a triangle, I don't have a dynamic render pipeline yet or anything, nor anything other than a clear color and a present

[brisk chasm]

https://www.youtube.com/watch?v=skW3clVG5Fo

discord sold their data to youtube again

[silver slate]

I wouldn't be surprised, probably one of the few apps that can actually secretly listen to all the things it could "advertise" to me.

Of course it would still not know what the hell I want because I mispronounce brand names for laughs.

[brisk chasm]

hehe

[cloud rivet]

I just assume there’s no privacy on the internet. Everything I type, do or look at is happening in a public square anyone can see.

[true moon]

It's just disappointing

[cloud rivet]

I don’t care personally. I internalized it all long ago.

I am working through Casey Muratori’s Computer Enhance course and writing my own math C library that runs both on the cpu and the gpu will maybe be a fun way to practice what I am learning in the course.

I am going to try and get good at understanding SPIR-V disassembly in the same way I am learning about disassembly on the cpu

[cloud rivet]

I spent a bit of time, and I'm not using textures right now anyway, but I don't see any evidence that vcc supports bindless textures, and am not sure how to use the nonuniformEXT or declare a pointer of sampler2D or even an array of sampler2D

hrm

plumbed some of the depths of the emit spirv code

and how the annotate code worked

I'll just experiment a bit with just non-texture stuff and then probably go to slang, I don't want to bother anyone about it with questions

I'm a ways away from using textures anyway

maybe I risk asking a question when I get to that point that I want to use a texture

and after spending some more time looking through the compiler code

[astral hinge]

you can't make a pointer of samplers (how would you get the pointer?) but I don't see why vcc wouldn't support an array of them

hmm you'd have to ask gob about it

[cloud rivet]

yeah I don't want to ask until I am sufficiently worthy, which may be an unatainable goal

in shady.h I see

typedef __attribute__((address_space(0x1001))) struct __shady_builtin_sampler2D* sampler2D;

I think the builtins are generated from spirv grammar maybe

[astral hinge]

lame

[wraith urchin]

You are worthy

[astral hinge]

don't feel afraid to ask

[astral hinge]

I bet they are hardcoded tbh

[cloud rivet]

I have searched the full codebase

I looked through the memory addressing stuff and the emit code

the only reference to sampler2D is in the example and that works with the compiler

[astral hinge]

man your question would've taken 20 seconds for gob to answer

[cloud rivet]

I built the compiler from this source code

I bet

:P

anyway I need to actually write some code to be able to draw even a triangle

[cloud rivet]

having issues with creating the shader objects in vulkan with vcc

I'm going to file a bug

this here is a comparison of slang and vcc spirv output if anyone is interested https://gist.github.com/btipling/1e4ec7f8c55494eb8302a448e13215a6

for similar fragment shaders

that's the wrong output actually

I fixed it, that was the vert output

made an issue https://github.com/shady-gang/shady/issues/48

I had another copy paste fail, I had the wrong code in that frag.slang in that gist that's fixed too now

[cloud rivet]

so the Slang support for neovim is pretty good, for visual studio it felt very broken in comparison

[cloud rivet]

I'm gonna use an nvidia only extension :/

VkPhysicalDevicePerStageDescriptorSetFeaturesNV

so I can use dynamicPipelineLayout

hrm

I think that breaks renderdoc

hrmmmm

lemme see if it has that, since nintendo switch probably uses this

oh

renderdoc doesn't even support vulkan 1.4

RIP renderdoc

Nsight Graphics™ 2025.2 frame debugging supports all of Vulkan 1.4.310.0.

VK_NV_per_stage_descriptor_set supported

it's not supported

it's in the not supported list

https://github.com/baldurk/renderdoc/issues/3625

fine I can't use it then

I'll use 1.3 also

[cloud rivet]

k

gross

??

oh

A

where did I get that if statement from

now, unfortunately, I need to set up the bindless descriptor sets, so I can send push constants to this thing

and after that I'll have to write a bunch of allocators

so I can allocate memory on the GPU

then I can have buffers and geometry and images

and I'll be able to do msaa and have a draw image

this went way quicker than it did with either of my C++ VK renderers

noice

[vagrant musk]

T
a
l
l
T
r
i
a
n
g
l
e

[cloud rivet]

It can’t know how to scale until I do a bunch of work

I need to implement descriptors and start a math.c and then I could send an orthographic projection with push constraints

If I want to pull from gpu memory I need to start on the memory allocator

I learned a lot about debugging windows applications yesterday

I installed procmon, wingbd, and a bunch of other tools to debug what caused renderdoc to fail to launch my app

I figured it out and that works now

Those tools are great

Nsight is still failing to launch

[broken fog]

the vk project 1 week to tringle meme is real

[cloud rivet]

I also learned that there’s nvidia gou crash app helper

[cloud rivet]

Well, in my defense I am doing a bunch of new things this time and had to work late three days

[broken fog]

yea not saying you're slow or anything, just poking fun

work sucks :(

[cloud rivet]

But if this were C++ and I was using vk bootstrap and VMA and slang from the start it would take me less than a day

But it would not be as much fun

Everyone starting out should probably just take Sascha’s new tutorial and use that

You have to go through it and at the end you have a good starting point, it’s a good tutorial

The slowest thing about it would be how long it takes a Vulkan document page to load it takes like 10 seconds

They should just have no js html pages

[dark saffron]

vkspec.pdf + sumatrapdf if you're on windows or a any good reader elsewhere

[cloud rivet]

Oh I will check sumatrapdf out thank you

[dark saffron]

i can read the vk spec on even a potato pc with it

peak windows software, same aura as notepad++ or paint.net

[cloud rivet]

Yes I like notepad++ I am trying to get back into neovim for editing , but I find it easier to open up all my references projects in notepad++. That is a bunch of C projects to look how they to do things with C. I used it to search through shady code

@dark saffron I am fully onboard now with the using real languages for shaders. Thank you for building vcc, I think it is great and would like to get it to work, going back to slang was just to make sure I could make progress, it feels like a big step back through. I was planning on writing shared code. I am going to spend more time on what I'm running into with vcc

I will get nsight to launch my application, maybe whatever is causing that issue is something that is a sign of some larger problem

I really want to write shaders in C

[dark saffron]

Thanks! I saw your issue I'm addressing some other bugs but I'll get on it asap

[cloud rivet]

thanks! if you need any information from me I will keep monitoring the issue, I have to read my github inbox for work so I am always looking at it

[cloud rivet]

ok I got both nsight and renderdoc working now I have some layer messages to go through from nsight though

ID,,Origin,Source,Message
20,,Target,NVIDIA Nsight Graphics,Detected VK_LAYER_KHRONOS_validation without VK_VALIDATION_FEATURE_DISABLE_UNIQUE_HANDLES_EXT; this will disabled all NVIDIA hardware specific features (e.g. profiling)

sumatrapdf is very nice

[bronze socket]

my preferred pdf viewer is mupdf, I should try sumatrapdf though

[cloud rivet]

it opens the vk spec with no issues

I have been using just my browser to read pdfs

and it's so much better than a browser

[bronze socket]

same but that has given me issues with super image heavy pdfs

[cloud rivet]

yeah this and the spirv spec are rough in my browser

[bronze socket]

sumatrapdf looks better than mupdf for proper reading tbh, mupdf is like the mpv of pdf readers

as in it'll just give you the pages and nothing else

[cloud rivet]

I can't believe I haven't heard of this reader before

[bronze socket]

same, I actually only found mupdf because I saw one of its maintainers comment on a harfbuzz issue

[cloud rivet]

never had an internal msvc compiler error before

C:\Users\Bjorn\projects\code\palinode\build>cl     /TC     /nologo     /std:clatest     /Wall     /Zi ..\src\main.c     /Fe:palinode.exe     /I ..\include     /I C:\VulkanSDK\1.4.313.2\Include     /link     /LIBPATH:C:\VulkanSDK\1.4.313.2\Lib     vulkan-1.lib     User32.lib
main.c
C:\Users\Bjorn\projects\code\palinode\src\gfx.c(775): warning C4034: sizeof returns 0
C:\Users\Bjorn\projects\code\palinode\src\gfx.c(657) : fatal error C1001: Internal compiler error.
(compiler file 'D:\a\_work\1\s\src\vctools\Compiler\Utc\src\p2\main.cpp', line 258)
 To work around this problem, try simplifying or changing the program near the locations listed above.
If possible please provide a repro here: https://developercommunity.visualstudio.com
Please choose the Technical Support command on the Visual C++
 Help menu, or open the Technical Support help file for more information
  cl!InvokeCompilerPassW()+0x5971b
  cl!InvokeCompilerPassW()+0x5971b
  cl!InvokeCompilerPassW()+0x596e8
  cl!InvokeCompilerPassW()+0x595f3
  cl!InvokeCompilerPassW()+0x591b5

I'm finding all the edge cases this weekend

[astral hinge]

Sumatra gives you little more than that as well tbh

[brisk chasm]

its my preferred pdf reader on binbows too

[cloud rivet]

I'm going to periodically do a YT video as part of a series where I give myself 1 hour to build a win32 software rasterizer from scratch in C and see how far I get. Not like a daily thing, just whenever I feel like it

idk how to build that so I will likely not get very far

at first

I don't care if anyone watches, I just want to do it and see how much progress I end up making

without using AI of course

[echo crystal]

cool

[astral hinge]

what if you became a vtuber at the same time

[wraith urchin]

I'll watch!

[cloud rivet]

I clicked a link to "tangled instructions" on the maximal reconversions proposal doc and it was quicker to alt tab to the pdf in sumatrapdf and type and search for "tangled inst" than it was to load the link to the spirv docs

Unfortunately, maximal reconvergence cannot guarantee a single behavior for switch statements. There are too many different implementations for a switch statement, restricting the divergence and reconvergence behavior would have serious negative performance impacts on some implementations. Instead, shader authors should avoid switch statements in favor of if/else statements if they require guarantees about divergence and reconvergence.

hrm

I was thinking about spirv instruction count differences between vcc and slang and wondering if it actually matters

in the case of assembly fewer instructions translates to less work done by the CPU

but SPIR-V is an IR and not the end result

and maybe it all washes out in the end, and maybe it might even be the case that more instructions can result in the implementation optimizing better? idk

[astral hinge]

check out radeon gpu analyzer for a tool that can show actual gpu assembly

you can input glsl or spirv

[cloud rivet]

Latest AMD drivers (AMD Software: Adrenalin Edition™ on Microsoft Windows® and amdgpu-pro on Linux®).

I think I need an AMD gpu to try it

let me see if nvidia has this

[astral hinge]

RGA works without an AMD GPU

[cloud rivet]

oh, it's mentioned in the requirement section for the software

oh

For all non-offline modes (DirectX® 12, DXR, Vulkan®):

I'm going to try the two spvs I got from slang and vcc

that have the same functionality

I realize vcc is early phase, I'm just curious

the end goal is for spirv and implementers to support real language features on the GPU just like one can get with CUDA

on the GPU graphics pipeline I mean

neat!

that's the slang

this is vcc

vcc actually has fewer instructions for the vertex shader

pretty much similar

despite like 10x the spirv instructions

nice

[brisk chasm]

more or less instructions doesnt always mean slower or faster

[cloud rivet]

yes, instruction level parallelism can play a part where you enable the CPU to understand if instructions can be run in parallel

but think it makes a difference

on the GPU the whole uniform vs divergence plays in too yeah

which is what I was just reading about

[astral hinge]

the compiler could also roll or unroll code or generate instructions with more or less latency

GPUs have ILP too btw

but they don't necessarily have fancy graph analysis hardware to detect hazards and reorder instructions like CPUs do

[cloud rivet]

I see

[astral hinge]

I edited the statement because some GPUs (e.g. pre-RDNA 3 AMD GPUs) do actually automatically track hazards

[cloud rivet]

the amount of spirv difference was striking, but you just have to measure I guess to know if it is signficant and can't just go and wildly gesture at it and say "this bad"

[brisk chasm]

you actually have to profile, to be able to say "this bad", unless you memorized each combination of instructions

[cloud rivet]

yes

[brisk chasm]

its like using fastinvsqrt on modern cpus these days, because its some assembly hackery it must be faster than it written out in plain c or something

[cloud rivet]

the computer enhance course's main problem that you learn is to improve the performance of calculating the haversine distance of 1m pairs of locations extracted from a json file, part of it involves a sqrt function, so I guess the course will go over that part of the problem

I'm going to write my own gltf parser so I'm interested in the json part of that course

he picked json specifically because it was a poorly performing format that you encounter in the real world

I'm going to have to write my own profiler code since palinode is fully NIH, I hope the course covers that too

the basics seem obvious I guess

[brisk chasm]

good luck

[cloud rivet]

thanks!

this may alll seem like a lot but, this is all way less work than making a game heh

and I'm not stressing about progress

and more in line with what I actually want to do

[brisk chasm]

ye no need to rush anything

[cloud rivet]

which is to learn and build skill

[cloud rivet]

I'm trying to convince this ai to let me merge my triangle code and it keeps finding things

Fix format specifier for long variable.

The format specifier %d is incorrect for a long variable. This could cause undefined behavior or incorrect output.

[cloud rivet]

I did do one of those videos today but even though I tested my mic in the video before, the full one hour my mic didn't record anything. I then tried a new video and it recorded audio too. I don't know what happened. I uploaded the video anyway, totally silent. If I redid the video I'd lose another hour and it wouldn't be really the first one, because I learned a lot in that hour. I also spent the last 20 minutes struggling because I didn't realize I had initialized my should_run main loop boolean as false and couldn't figure out why the window wasn't staying open. I thought I had gotten something wrong with the win32 API. I figured it out like 2 minutes before I ran out of time. You can't hear anything I said in my confusion or how much I laughed when I figured it out, but anyway. Each video I learn a lesson or two, and in this one it's to uh, read my own code.

it's fun to do tbh

I'm going to keep going with that

I kind of one to do another one today, but it's best for my learning if I don't do it twice in one day, for persistent memory

I'm going to maintain a public github repo with each attempts full code

[true moon]

Fast forward 2 years and we're on video day 685 and Bjorn is optimizing his string compare function for uOp cache usage

[cloud rivet]

that's the dream

what got me interested in graphics programming wasn't really ever video games, it was the demo scene

what got me hooked on video games for a brief moment there was how much progress I had made and how fun it was to work on Blockens

I didn't understand it would actually escalate to be a game, until I started building something that looked like a world and it was fun and addicting to keep working on it

then I ran into problems with lack of not having fundamentals and I did the foundations thing, then retried my hand at a game with Rosy and it wasn't at all like Blockens, it was a slog

[gaunt drum]

was blockens the zig one?

[cloud rivet]

yeah

I made it private, but there's a thread with videos

foundations was also zig

zig and opengl go well together

well

before cimport got removed

I don't know what it's like now

[gaunt drum]

there is translate-c and not so well documentated last time I checked translateC build stage

[cloud rivet]

I like the devs working on it, but I'm happy with C

it's interesting, I woudln't have gotten into any of this if it wasn't for zig

[gaunt drum]

what is precursor to zig experiment?

[cloud rivet]

like 10 years ago I built a little 3D renderer with Metal

and had done some 2D software side draw things with Java and Kotlin

[gaunt drum]

cool, I had some minecraft bukkit modding stuff and had some experiments with godot/unity but aside from that I always gatekept myself from gamedev itself "oh, you can't do it alone"

but it is my childhood dream, yeah

[cloud rivet]

nice

I thought what demo scene people could do in 3D with a tiny bit of code was just so impressive to me and something I've always wanted to do, I don't really want to do all the things to make the program small, I just thought if they can do that with so little code I should be able to do something

[gaunt drum]

I started to take it seriously during 2021/2022 and had many unsuccessful attempts with existing engines and my own NIH stuff. left my job during 2023 and somehow didn't starve yet XD