Airlie: "Why Sharing Code With Windows Isn't Always A Win"

[phoronix]

Phoronix: Airlie: "Why Sharing Code With Windows Isn't Always A Win"

Following the news today of Intel sharing ~60% of their GPU driver code-base between Windows and Linux and working to bring the Intel Graphics Compiler (IGC) to Mesa in 2021, not everyone is enthusiastic about those prospects...

http://www.phoronix.com/scan.php?page=news_item&px=Airlie-Thoughts-On-Code-Sharing

[airlied]

"For end-users at least and consumers not concerning themselves with the development model, all indications are the IGC'ified Mesa performance should be faster than the current compiler back-end."

For a benchmark led website like this i find that a very strange statement, the vendor said it so it must be true?

[Michael]

"For end-users at least and consumers not concerning themselves with the development model, all indications are the IGC'ified Mesa performance should be faster than the current compiler back-end."

For a benchmark led website like this i find that a very strange statement, the vendor said it so it must be true?

Obviously when there is support available I'll be benchmarking it. At this point there isn't any reason to doubt their claim outright and imagine they wouldn't be wasting their time in the effort if the performance wasn't looking promising.

[duby229]

Obviously when there is support available I'll be benchmarking it. At this point there isn't any reason to doubt their claim outright and imagine they wouldn't be wasting their time in the effort if the performance wasn't looking promising.

Yeah, well, you remember the claims Intel made about its OpenCL Beignet project.... I expect this to be another project with no community at all, that for all the good it does, its just as good as closed source any way.

[ms178]

By the way, is the "community/open-source-first workflow" a hard requirement for such a big code contribution from a company which is already working on other parts of the Mesa stack? Or how would such an integration look line? Mesa and IGC sound like two separate projects which somewhere magically find together (or not).

[microcode]

Hey, it could go the other way entirely. We'll know when we see it.

[bug77]

Yeah, well, you remember the claims Intel made about its OpenCL Beignet project.... I expect this to be another project with no community at all, that for all the good it does, its just as good as closed source any way.

Keep in mind, even if you can't actually engage in development, open source code can still be audited or used for academic research/teaching.

[duby229]

Keep in mind, even if you can't actually engage in development, open source code can still be audited or used for academic research/teaching.

Audited, sure... Research and teaching, Nope, make that a big fat no.

[cqcallaw]

While I'm extremely sympathetic to Arlie's argument for sustainable, community-driven development processes, I wish him best of luck convincing decision-makers that this development process is competitive. Performance on AAA gaming titles is still an important metric for many review sites, and performance on Intel's Linux driver stack lags Windows in this domain, sometimes by a substantial amount.

To catch up, Intel's Linux development will have to move faster than Windows development, with fewer heads and less funding. That's a tall order indeed. Attempting to simultaneously build consensus with a famously opinionated and litigious community sounds like a recipe for burnout.

[duby229]

While I'm extremely sympathetic to Arlie's argument for sustainable, community-driven development processes, I wish him best of luck convincing decision-makers that this development process is competitive. Performance on AAA gaming titles is still an important metric for many review sites, and performance on Intel's Linux driver stack lags Windows in this domain, sometimes by a substantial amount.

To catch up, Intel's Linux development will have to move faster than Windows development, with fewer heads and less funding. That's a tall order indeed. Attempting to simultaneously build consensus with a famously opinionated and litigious community sounds like a recipe for burnout.

Except that Intel has tried to exclude the Open Source community several other times in the past, the most famous being Beignet, and in every single case it was a total failure that cost Intel millions of dollars and years of lost time. Precedence for this already exists, precedence that Intel themselves already set.