00:40 youngfaceatold: Yeah I still think that vertex shader can be used the way I said... have GL_POINTS, and do not write the output, and bind new vbo and get new attributes, that would make infinite iterations of vertex shaders.
00:42 youngfaceatold: You can write to vbo from egl.
01:13 Mangix: my nvmem-layout commits have mostly been merged. let's see if anything fails
10:07 youngfaceatold: Graphics is with compiler in the mix, the hardest it can get in programming, but it's not so hard when you take it to granules, it's just in fragment shader you would want to write to your chosen location in framebuffer object on older cards too, but that is just a dma packet it maps to so that when modifying the drivers fw would allow you to, without image on buffer load store extensions write to address array indices the data that you want in
10:07 youngfaceatold: fbo.
10:14 youngfaceatold: But just don't forget that dma controllers are really easy to be programmed.
10:18 youngfaceatold: There are even drivers which at most want the mode to be altered by driver, and that's the only modification
10:18 youngfaceatold: That's called fifo mode
10:18 youngfaceatold: Async mode of DMA
10:20 youngfaceatold: There are also incremental fixed modes and decremental, where you load the size and how many entries you want to copy/transfer
10:24 youngfaceatold: But there's a better way to wiggle around that, and those are conditionals that I talked in vertex case about
10:24 youngfaceatold: So you never write colors out until your shader has everything ordered out
10:25 youngfaceatold: I use that mode, cause there are problems otherwise with hw vendor ip portfolios.
10:26 youngfaceatold: So for me it's easier to wiggle around in userland shader engine
10:36 youngfaceatold: Cause look at it this way I compute supervision hashes, I only need their compute unit and output spec.
10:37 youngfaceatold: Those hashes have eagle vision of every pixel and incoming attribute
10:37 youngfaceatold: So the hardest I mean most unnatural
10:38 youngfaceatold: Is to get graphics iterations right, compute mode would work in all ways anyway as to how user requests hence wants
10:38 youngfaceatold: Graphics part is 3D api
10:39 youngfaceatold: That's the API and spec about the hw 3D register reference of these sets
10:40 youngfaceatold: There are some fixed function units meant to do fixed modes, which at modern set is bad design even rather.
10:41 youngfaceatold: Fixed function units that none needs
10:42 youngfaceatold: Those blocks still live in hw where they actually should be forced to migrate into SW.
10:43 youngfaceatold: But in both cases you can still control the needed from SW
10:43 youngfaceatold: it's just that those blocks consume silicon on the card
10:44 youngfaceatold: The biggest advantage of defeating the situation from userland is that it's ip portfolio safe, it never touched that land
10:45 youngfaceatold: Regardless if ff blocks be lifted to SW or consume still silicon in hw, for some so the others other
10:46 youngfaceatold: You still avoid getting into trouble by not going to deal with IP guarded units.
10:47 youngfaceatold: I guarantee that everything is good with my approaches.
10:47 youngfaceatold: You do not get hurt when coming together with my ideas
10:48 youngfaceatold: Cause I am so smart I do not ask for those troubles to become current or actual
10:49 youngfaceatold: I deal so that I am genius who can live with their restrictions
10:49 youngfaceatold: Cause those are not real restrictions in the things
10:49 youngfaceatold: You just can evade them different
10:51 youngfaceatold: In graphics we can push to the same base our own graphics api
10:51 youngfaceatold: Which is the most modern or any other chosen
10:52 youngfaceatold: Cause we are way better than directx or opengl or vulkan
10:53 youngfaceatold: The job is very simple but it's just very sane in terms of logics, it's very thoughtful
10:55 youngfaceatold: It's made possible to evade by conditionals that let you evade the output writes and delay those until superexecution was ready to output everything
10:56 youngfaceatold: It's all in the compatibility set in hw, called shader model
10:57 youngfaceatold: It's Microsoft that rules the world of graphics cause they started it
10:57 youngfaceatold: Microsoft is very big the biggest player there when we consider that they started it all too
10:58 youngfaceatold: That's why those hw units are controlled by Microsoft spec
10:58 youngfaceatold: Called shader model, or SM
10:59 youngfaceatold: But hw graphics card manufacturing companies can lift those units to programmable fabric
11:00 youngfaceatold: Microsoft mandates everything in the PC gaming market.
11:00 youngfaceatold: Even the low level hw blocks
11:01 youngfaceatold: We could brutally kick them out, but I chose to pay respect instead
11:01 youngfaceatold: Cause they did so good work, I do never kick them
11:06 youngfaceatold: In the same way I chose to pay respect for the IP of the hw blocks that manage Microsoft mandates.
11:07 youngfaceatold: By not touching or interfering
11:07 youngfaceatold: With their interests
11:07 youngfaceatold: But more like do not kick the out cause they deserve so good
11:08 youngfaceatold: They did not hurt me, why should I hurt them?
11:09 youngfaceatold: They allow for me to do everything I want, that's why I say this I pay respect for this
11:09 youngfaceatold: And give a kiss, not punching in the nuts that take a hike
11:12 youngfaceatold: I do not even do this against my will, it's just that they deserve their kiss, by ratifying that any expert is not abused by limiting their results
11:12 youngfaceatold: They just absolutely fully deserved their kiss
11:18 youngfaceatold: I can delegate mandate too, cause they themselves allow this
11:18 youngfaceatold: Cause they liked to find smart and capable people
11:19 youngfaceatold: They hence give me all rights
11:20 youngfaceatold: But in a mutation way, you have to be intelligent to understand those concepts, and they chose not to kick those in the nuts by default