>70% of the project is in pure assembly>vbroadcasti32x4>punpcklqdq>no slop detected What's your reply AI-tards?
>>108955212you just can't spot the ai code because competent developers are using a good model.
>>108955269More or less, that being said people don't really know what is or what isn't AI. And they blame everything that breaks being coded using Codex/Claude. It could be completely AI Free but since it's shit, must be an AI.By that same logic the whole Linux Desktop is written by AI but that doesn't make sense as it's pretty much as shit as it was 10/15 years ago, the difference is Wine improved a lot.
>>108955212>portable>full of assembly
>>108955269this, they are coding like huwite men not like jeets
>>108955212AI still isn't good enough to come up with something by scratch from a reference design. It can't even make a compliant web browser from all of the specs, never mind trying to make a competitive video codec design from scratch.AI slop is only useful for regurgitating existing ideas. It literally couldn't make something like AV2 if it wanted.
>>108955548It's called optimizing for specific platforms while having a generic fallback implementation. Crazy concept, right?
>42.4 = 70The absolute state of assemblytards.
>>108955212I don't get paid enough to care. You will have an AV2 decoder written in nodejs and you will like it.
>>108955548Why not? He just has to target a couple architectures, then recompile. The era of asm not being portable ended in the mid to late 90s
>>108955591see the other half of that bar that's as brown as your skin? That's also assembly. https://github.com/ressiwage/DAV2DAlso it's 80%
>>108955212rav1d might be slow, but at least its safe.
When is 4chan gonna support AV2?
>>108955619Rav1d would not be possible without Dav1d. They literally took its C code, put in a blender and whatever Rust paste came out the other end that's Rav1d.
>>108955269>>108955537I've asked AI about it and xhe said it is fully written by human>>108955591lol baited, cope and seethe AI-tards... cope and seethe>>108955624Next sprint, trust
>>108955561Why do indians/malaysians come to /g/ and use 2016 internet lingo like 'huwite' in an attempt to fit in?
>>108955619>might be slowThen it's fucking worthless you brownie holy shit what memory safety does a fucking decoder need? Speed is far more relevant to its operation.
>>108955689what do you call it when 4chan does agile? fagile
>>108955703>decoder memory safetyEXRReadPlugin::decodeBlockAppleEXR
>>108955703>holy shit what memory safety does a fucking decoder needpeak /g/, you do realize these decoders are used and can be exploited in software that like 99.9% of people on earth are actively using?
>>108955212i was afraid they were going to use rust for dav2d glad they didn't
>>108955703>I want to get RCE'd by opening a videoyou know you're decoding hundreds of images and videos every time you use your PC
>>108955609OP is not github, looks more like gitlab, show the real one snailboy
>>108955609>42.4%>half>2 halves together makes 70%Your brain on assembly.
>>108955793Rustards in shambles, they are spamming the AI-assisted suicide watch emergency channel right now>>108955583>AI-tards>optimizationBy God! Blasphemy! Are you going to say we need to read the code? Impossible!
>>108955689how would an ai know how the code was written?
>>108955858Why should I know? I'm just vibing. If my AI says it is then it is what it is, simple as>>108955807>>108955820Have your AI-tards brain devolved and forget how to do a web search? I mean it is a guarantee but I don't think it'll be this fast. Progress!
https://jbkempf.com/blog/2026/dav2d/Here's the announcement as well.>The current dav2d tree already contains a feature-complete AVM v15 decoder supporting both 8-bit and 10-bit decoding.>One reason the project has progressed so quickly is that dav2d does not start from scratch. AV2 shares many concepts with AV1, and dav1d already solved a number of architectural questions around threading, SIMD organization, testing, portability and API design.>While AV2 requires substantial new decoder code, a lot of the experience accumulated over years of dav1d development transfers directly to dav2d.
>>108955609A lot of the assembly is copypaste, but that's fine.
>>108955858It looks for the use of em dashes.
>>108956072>writing a reusable codeTruly an ingenious invention>>108956080Correct—
>>108955269>AI generated assembly(You) don't write assembly. AI is still ass at assembly beyond all reason.
>>108956137>Truly an ingenious inventionI was more pointing out that the percentage is somewhat inflated because of the amount of copypaste.It's not like every line of assembly is artisanally hand written. And then there's having a routine in SSE2, and then in AVX2, and then AVX512, and then the same routine in ARM, and in RISC and in LOONGARCH.It's not that you can just copypaste the routine in these cases, but once you have worked out the routine in one architecture, porting it to another is relatively easy.
>>108956175it did it correctly :3
>>108956227>doesn't even compileBinary or gtfo """martin"""
>>108956265womm
>>108956220It is true only to a certain degree, each platform has its own quirks and the engineers had to adapt to them when chasing performance. I think at this stage, using only test samples, the performance is quite good. Looking at dav1d and how it is comparing to libaom in the early days is staggering, I still can't comprehend why AOM initially refuses to build a performant software decoders and instead banking on hardware manufacturer catching up, yeah they had a good backing and money but still without dav1d we would still be stuck with h265/x265Them learning on their strategic mistake and basically releasing dav2d ahead of the standard is not an easy feat, the engineers holding back on blindly riding the AI train when they literally can just ask and will get unlimited SOTA models token should be appreciated>>108956400Aww nice, just another broken, mentally ill trying to look like cute girl aesthetics. Thighs, """Martin"""? >,,<
>>108956175and it will not be able to, at least on par with expert human engineers. Writing assembly need a very specific skillset and deep understanding of the underlying hardware, not to mention countless iterationsIf AI managed to, by itself, reimplement AV2 decoders with similar performance only from spec I'm afraid that's AGI
Can anyone explain what's the thinking behind making video files smaller by 20% at the price of making deciding them 70% more expensive? Are HDD prices that bad or something? Or is everyone streaming 8k?
>>108958425* decoding
>>108958425It's a multifaceted answer. The easiest one is that streaming services want to keep their storage and network footprint as small as possible, while shifting the onus onto the consumer. The second is that, generally, this puts pressure onto hardware manufacturers to implement hardware decoding for those new codecs, so consumers don't experience as much of, or any change in their service. In general it is worth it to use the new codecs, but the hard part is pushing adoption.
libsvtav1 was like 10x better than dav1d
>>108955212Are you under the delusion that AI can't code assembly?
>>108957814Yes >>108958889
>>108958841>X is better than YWithout dav1d we'll stuck with libaom>>108958824Yep trading storage (perpetual) for compute (one time)
>>108955212>dav2dIsn't that the rapper who killed his girlfriend?
>>108961832No that's 0J
>>108957814>>108961088the demand for assembly programmers is so tiny when compared to every other type of programmer that it may as well be a rounding error thoughbeit
>>108958841And when was that?SVT-AV1 is an encoder, dav1d is a decoder.
>>108955212Call me when svt-av2 comeOtherwise I'm staying with SVT-AV1-Essential
>>108955212I get paid for delivering a bunch of json, I don’t actually care about solving le problems
