Helloooo Stephan =) On 3/15/25 12:52 PM, Stephan Verbücheln wrote:
Am Donnerstag, dem 13.03.2025 um 02:37 +0100 schrieb Dirk Lehmann:GPU shaders should be clearly respected as software (and not hardware). Therefore, `intel-media-va-driver-non-free` is rightly in the Debian archive `non-free`.First of all, I completely agree that: 1. shaders are computer programs 2. the shaders in that package are unfree
okay, I have looked a bit into the sources of the upstream repository.
And I have good news!
All source files I checked in
* media_driver/agnostic/*/kernel{,isa}/*/*.c
which you listed at begin of this thread looks like as they are
* EITHER SHADERS NOR FIRMWARE !
These datasets are looking more like n-dimensional arrays.
Kernels in image- and video processing are mathematical matrices.
I.e. on images they will be scrolled (mathematical operation:
"symmetric convolved") through the images as (real-time)-filter for
anti-aliasing, sharpening, etc. In this `intel-media-va-driver`-case
they are seeming to be used to interpolate frames. I would assume,
for example if the video footage has a frame-rate of 60 fps and the
rendering-context (i.e. display) 144 fps then additional frames will
be generated/interpolated.
Here a Wiki link for the case of image processing:
* https://en.wikipedia.org/wiki/Kernel_(image_processing)#Convolution
Yes, such tasks are usually done by shaders from GPU, but I'm quite
sure that the data above are just the Kernel-Matrices itself. Would
be nice, if somebody can verify that the exported symbols in the
corresponding header files are really no executable code containing.
Question of licensing
*********************
I'm not a lawyer, but just want to start a discussion about a possible
sublicensing. Even in my last post I had data explicitly excluded
from the general case
On 3/13/25 2:37 AM, Dirk Lehmann wrote:
> [...]
>
> As you expect, in detail from technical point of view, firmware is a
> special kind of software. It is not data -- and it is executable in
> the sense that it changes the state of "physical hardware" in time.
>
> [...]
From technical point of view, I would
* consider these kernel-matrices as binary images
As like images are painted with tools like Gimp, the kernel-matrices
were generated by the `KernelBinToSource.exe` tool (see source
comments), whose sources are available in the upstream repository,
too.
But I'm not a lawyer and don't know in detail how to deal with it. I
would try do something like this:
The license of sources itself grants to the kernel-matrices-images
besides copy, modify, merge, publish, distribute also explicity
* sublicensing
For me as not-lawyer it is a license of the class
attribution-share-alike license. The question should be if it is
allowed to sublicense the kernel-matrices-images as
* CC BY-SA
(Creative Commons Attribution-ShareAlike Licenses) ?
as this license should be better for datasets -- and if so, if the CC
BY-SA is compliant to our DFSG. But I really have to less knowledge
about that topic, and which formalities are needed to be respected for
sublicensing.
Greets,
Dirk =)
OpenPGP_0xE2A3766F21F02BD5.asc
Description: OpenPGP public key
OpenPGP_signature.asc
Description: OpenPGP digital signature
