Silk.NET•3y ago

But that seems incredibly unnecessary

SSQDev and why you didnt implement an audio decoder as well?

Lyris the Kitori•6/14/23, 7:27 AM

didnt need it

KingOP•6/14/23, 7:28 AM

You take from the gpu only to give it back to the gpu

KKing But that seems incredibly unnecessary

Lyris the Kitori•6/14/23, 7:28 AM

a) "hardware accellerated" does not always mean GPU
b) its not integrated into the many graphics APIs my engine supported so its nessesary

KingOP•6/14/23, 7:29 AM

Surprised no one thought out a fancy way to just... pipe the GPU video decode directly into the gpu output whatever, the moment that feature was introduced

KingOP•6/14/23, 7:29 AM

Without having to bounce it to the cpu first

Lyris the Kitori•6/14/23, 7:29 AM

you can with things like DXVA or Vulkan Video

Lyris the Kitori•6/14/23, 7:29 AM

but

Lyris the Kitori•6/14/23, 7:29 AM

thats not this

KingOP•6/14/23, 7:29 AM

aah

Lyris the Kitori•6/14/23, 7:30 AM

bigger problem is video isnt always accellerated with the GPU, there are other video accellerator devices that arent a GPU

Lyris the Kitori•6/14/23, 7:31 AM

and even if it is a GPU, you cant guarentee that the formats will be the same, or anything like that, or even that the driver will let you share that memory

Lyris the Kitori•6/14/23, 7:31 AM

the APIs and contexts would generally have to be the same

Lyris the Kitori•6/14/23, 7:31 AM

and FFmpeg will manage all that itself

KingOP•6/14/23, 7:32 AM

True... I guess I'm ignoring the whole enterprise segments that love their rackmounts with like 5 dedicated video decoder cards

KingOP•6/14/23, 7:33 AM

And maybe apple? I have no idea if their media encoder "engine" counts as being part of the gpu

Lyris the Kitori•6/14/23, 7:33 AM

however that works, it probably doesnt help me anyway

Lyris the Kitori•6/14/23, 7:34 AM

when id have to somehow make it interop with GL

SQDev•6/14/23, 7:34 AM

(not from your code, just found online) Doing this just to get the duration of the video...

Lyris the Kitori•6/14/23, 7:35 AM

there is no good way to make CPU-less GPU to GPU frame copies here
a) API/usage agnostic on the FFmpeg side
b) API/usage agnostic on my side
c) API/usage agnostic on the GPU side

SSQDev (not from your code, just found online) Doing this just to get the duration of t...

Lyris the Kitori•6/14/23, 7:35 AM

me when i

SQDev•6/14/23, 7:37 AM

well you have ->duration already calculated i believe

SQDev•6/14/23, 7:37 AM

ffmpeg gives weird values num den to calculate it

SSQDev well you have ->duration already calculated i believe

Lyris the Kitori•6/14/23, 7:37 AM

calculated by FFmpeg :^)

SQDev•6/14/23, 7:37 AM

lol

KingOP•6/14/23, 7:38 AM

I assume the negative check is there cause ffmpeg reports weird stuff if you feed it a corrupted video file?

Lyris the Kitori•6/14/23, 7:38 AM

i have no idea why that is there i wrote this code a year ago

Lyris the Kitori•6/14/23, 7:39 AM

this is all the documentation i have for myself

KingOP•6/14/23, 7:39 AM

lol

SQDev•6/14/23, 7:39 AM

KingOP•6/14/23, 7:39 AM

"This time I"m DEFINITELY documenting it properly so future me will have no problem understanding why I do what I do"
You 2 weeks later: What in the fuq

KKing I assume the negative check is there cause ffmpeg reports weird stuff if you fee...

SQDev•6/14/23, 7:40 AM

since ffmpeg supports 23468248624 formats, not all return correctly for all videos i believe

KKing "This time I"m DEFINITELY documenting it properly so future me will have no prob...

SQDev•6/14/23, 7:40 AM

lmao

KKing "This time I"m DEFINITELY documenting it properly so future me will have no prob...

Lyris the Kitori•6/14/23, 7:40 AM

well random negative checks i probably didnt bother documenting, nor did i probably even understand at the time, most of this code is reworked from examples/other codebases

SQDev•6/14/23, 7:40 AM

me in the mean time

Lyris the Kitori•6/14/23, 7:41 AM

theres very little documentation for FFmpeg, so a lot of what i wrote is things sussed out from other projects

LLyris the Kitori theres very little documentation for FFmpeg, so a lot of what i wrote is things ...

SQDev•6/14/23, 7:41 AM

i agree

SQDev•6/14/23, 7:41 AM

such a big project having almost no useful documentation feels weird

SQDev•6/14/23, 7:41 AM

all those big companies just internally care for their projects i guess

LLyris the Kitori well random negative checks i probably didnt bother documenting, nor did i proba...

KingOP•6/14/23, 7:43 AM

Ah, pretty much my exact experience trying to do imgui things lol

KingOP•6/14/23, 7:44 AM

"I don't know why this is important but apparently everything explodes if you don't do it... ??? oh well"

Lyris the Kitori•6/14/23, 7:45 AM

i've found imgui to be well documented and easy to use, theres nothing in my imgui code id describe like that

KingOP•6/14/23, 7:46 AM

All I remember from my past attempt was I tried doing it on my own instead of copypasting an example, and everything just went maximum bork, and I couldn't figure out why until I realized I wasn't doing a boolean check on something or the other, which seemed completely pointless and redundant, but not doing it just broke everything horribly

KingOP•6/14/23, 7:47 AM

I think I spent a whole day trying to figure out why it was necessary... but neither stack overflow, reddit or the official examples/documentation made so much as a mention of it

SQDev•6/14/23, 7:47 AM

that happened to me multiple times in the past

SQDev•6/14/23, 7:49 AM

especially if i work on something very big, multiple source files thousands of lines of codes, if i just stand up to somewhere for a sec without finishing, i always end up writing incorrect code when i am back to finish it

SQDev•6/14/23, 7:49 AM

missing something thanks to

KingOP•6/14/23, 7:49 AM

Oh yeah... the whole "train of code" thing

KingOP•6/14/23, 7:50 AM

If you take a break and come back, and everything just feels like mystery heiroglyphs and you have no idea why the heck you did it that way

KingOP•6/14/23, 7:50 AM

Until you starts from the beginning and sort of get the train of thought rolling on the same track again