SGX driver outdated

[bismuthdrummer]

As a software dev, I am totally with notaz on the gravity of having disparity in software versions. It's all fine and dandy until you get questions, confusion, and irritation from the customers who have to live without the newer, shinier stuff. Or the ones who want the newer stuff but grab the wrong version of a pnd because they were confused (and worst of all you might even waste time troubleshooting what is ultimately just a version mismatch)... this is all avoided when you aim for universal support.

But to the point, I would like to see the progress made as well, so long as the world is stable it doesn't have to be absolutely uniform. If I start coding for Pandora I'd like to have recent drivers...

 

[lunixbochs]

The stuff I'm working on should give us the best of both worlds. I don't think it'll need multiple PND versions.

 

[bismuthdrummer]

The stuff I'm working on should give us the best of both worlds. I don't think it'll need multiple PND versions.

Then, sweet!

 

[gunrock]

I'm with Notaz and the drummer, though. The Pandora is too small a community to split software support on.

 

[elwing]

I'm with Notaz and the drummer, though. The Pandora is too small a community to split software support on.

I can only agree, but aren't the old driver bugged and glitched on the 1GHz pandora right now?

 

[Askarus]

The stuff I'm working on should give us the best of both worlds. I don't think it'll need multiple PND versions.

So, where's the pproblem then.

Let's see what comes out in the end.

I'm nweither against Notaz nor on his side. First I want to see what's the result in the end.

Not to try it is the worst thing. That's my opinion.

 

[ShleeDragon]

As if I'm deliberately preventing you to use the new drivers..


The drivers are publically available from TI, just download and get them working:

http://software-dl.ti.com/dsps/dsps_public_sw/sdo_sb/targetcontent/gfxsdk/latest/index_FDS.html


Good luck.

We don't have the experience. We listen to you. if you say no, we listen.

I'm sure the newer SOCs might work and the ESes will fail.. that's expected BUT the fact is.. people are running ESes.. so we need to respect them

 

[skeezix]

notaz does not have an opinion here; don't take it like he's for or against anything. He's purely talking technical feasibility..if someone is for or against notaz here, they don't understand the issue at all if it can be made to work, notaz is for it

step 1: get driver working on any device

step 2: given 1, determine its scope.. old devices, or just new? is there any ABI issue so pnds need to be built for either version?

step 3: depends on step 2 .. if ABI issues, then decide what to do, etc..

 

[Loonie]

I can only agree, but aren't the old driver bugged and glitched on the 1GHz pandora right now?

Precisely. We already have a situation where some software won't work on some Pandoras. Admittedly it's a fairly limited number of cases, but it's most certainly there.

 

[Askarus]

notaz does not have an opinion here; don't take it like he's for or against anything. He's purely talking technical feasibility..if someone is for or against notaz here, they don't understand the issue at all if it can be made to work, notaz is for it

step 1: get driver working on any device

step 2: given 1, determine its scope.. old devices, or just new? is there any ABI issue so pnds need to be built for either version?

step 3: depends on step 2 .. if ABI issues, then decide what to do, etc..

I support this.

 

[lunixbochs]

I'm not aware of any breaking EGL/GLES spec changes recently.

The newest version of the driver won't work on the original SGX. I plan to see if there's anything major technically preventing the chip from running on newer drivers or if they just stopped supporting it because it's old

 

[bismuthdrummer]

notaz does not have an opinion here; don't take it like he's for or against anything. He's purely talking technical feasibility..if someone is for or against notaz here, they don't understand the issue at all if it can be made to work, notaz is for it

step 1: get driver working on any device

step 2: given 1, determine its scope.. old devices, or just new? is there any ABI issue so pnds need to be built for either version?

step 3: depends on step 2 .. if ABI issues, then decide what to do, etc..

Nay, I understand, I wasn't meaning to polarize it in any fashion. Notaz placed emphasis on a potential rift in software support, and I concurred that should be a concern. That's it.

I certainly wouldn't say "don't research/develop it"... that would just be silly.

 

[lunixbochs]

I just compared the SGX 1.0.3 driver (original CC) to the 1.2.1 driver and the code is near identical. We have the 1.2.1 driver in the latest v1.9 graphics SDK. This means a new driver for the CC Pandora might be feasible.


I still haven't fixed the 1.9 kernel module, so the new driver doesn't run on my ghz pandora yet.

 

[ShleeDragon]

Let the SGX Hack-Fest begin!

 

[notaz]

I just compared the SGX 1.0.3 driver (original CC) to the 1.2.1 driver and the code is near identical. We have the 1.2.1 driver in the latest v1.9 graphics SDK. This means a new driver for the CC Pandora might be feasible.

While most files match there are large differences in a few of them, one containing code that runs on SGX itself and a few other ones.

 

[Wally]

Nothing a few ifdefs can't solve right?

 

[notaz]

ifdefs in binary driver blobs?

 

[lunixbochs]

I just compared the SGX 1.0.3 driver (original CC) to the 1.2.1 driver and the code is near identical. We have the 1.2.1 driver in the latest v1.9 graphics SDK. This means a new driver for the CC Pandora might be feasible.

While most files match there are large differences in a few of them, one containing code that runs on SGX itself and a few other ones.

That's definitely the case for the 1.2.5 driver blobs, but my analysis of the 1.2.1 vs 1.0.3 binaries didn't have the same conclusion. The files were close enough I could feasibly perform a manual analysis of each changed byte.

If I can get v1.9 working on the 1.2.5 core I'll look more deeply into supporting 1.0.3.