In-Fighting Continues Over OpenACC In GCC

[Daktyl198]

I don't get the issue...?

NVidia are doing exactly what is ENCOURAGED in the Open Source community: doing what benefits you, that might benefit others.
They even said that the PTX thing is just a back-end, meaning that with the framework laid out for the OpenACC front-end, Intel and AMD could come along and write their own back-end to it and help with the front-end.

So what if PTX isn't Open? If you don't like it, write an Intel/AMD back-end to it and use their cards. In the meantime, stop getting the the way of *nix users who want this feature on their nvidia cards.

[MaxToTheMax]

Michael, this is better journalism than usual for you. Good job.

[droidhacker]

Well there is a certain irony in criticizing NVIDIA's policy of keeping their code private and not choosing to reveal it to the world, and doing so under the cover of privacy and anonymity. If we don't have a right to know this guy's name, what right do we have to NVIDIA's IP?

You've got it all backwards. Hiding his name would be the equivalent of nvidia hiding THEIR name. His message is plain text, that like open source. If he GAVE HIS NAME, but encrypted the message, that would be like being nvidia.

[stativ]

Here's a thing ? the language support has frontend and backend. Anyone can write backend targeting whatever they want now that the work on frontend is done. Possibly blocking language support only because of its backend means that if someone else wanted to write OpenACC support, they would have to write the frontend code, too.

Now let's become more subjective. First, an anonymous source doesn't look very trustworthy, it looks more like a one person's rant which is not open to a discussion. Sure, it would certainly be better if all relevant technologies were free, but currently that can never be true, because of the microcode running on the CPU. The other thing is since when the output of the compiler has to be low-level code? Many compilers outputs bytecode like javac (and Java wasn't always very open), heck, there are even compilers that outputs JavaScript! Targeting PTX is not that different, not to mention that even nVidia itself doesn't target the GPU ISA directly in their software.

[deanjo]

You've got it all backwards. Hiding his name would be the equivalent of nvidia hiding THEIR name. His message is plain text, that like open source. If he GAVE HIS NAME, but encrypted the message, that would be like being nvidia.

There is still the sentiment that open development and discussion of that development is done in an open manner such as mailing lists. If the OP were to carry out his concerns in a public manner, not only can he have those concerns addressed point by point with out hearsay. I agree with others that choosing to voice those concerns behind a third party comes across as cowardly as he tries to shift his the responsibility and burden to others. While the GCC devs actually have the power to block it, the discussion and debate is open for everyone to participate in (which has been the FOSS way forever). Sometimes those "hey I have a concern" mails to the dev-lists do have persuasion to the ultimate decision that the devs make.

[XorEaxEax]

While I have absolutely zero interest in another NVidia proprietary solution for GPU acceleration (which is also why I have no interest in CUDA), the FUD spread by the 'anonymous source' is overbearing.

The 'give me 1 good reason why an open source compiler should allow itself to be sold to the most closed source backends anyway' comment is complete nonsense.

While I think GCC spending resources on implementing code generation against a proprietary solution would be really poor judgement and cause for some alarm, that is not the case here.

NVidia is footing the bill and are paying CodeSourcery to implement and maintain this code generation, it is also not in any way a dependancy as it's an optional component which you can omit from your build, and even if it is part of the GCC build you need to explicitly target OpenACC for it to be used.

Again I have no interest in PTX unless other manufacturers start to support OpenACC through it for their drivers and thus 'standarize' it across GPU's, however from what I've read the idea is to merge OpenACC with OpenMP, so the OpenACC implementation added to GCC which will currently only emit PTX byte code can be exteded to also emit whatever code any other GPU vendor wants.

So it's not as if OpenACC in itself is a PTX-only solution even though NVidia will only pay for PTX support to be implemented, it IS open source after all.

As for the commenter being anonymous, I have no problem with that. We are pretty much all anonymous on this very forum and we should be judged by what we say and not who we are.

And my problems with the anonymous source is what he/she says, which is at it's base an interesting observation of facts (NVidia is paying CodeSourcery to implement OpenACC support but only for their own PTX) but then he/she sprinkles it with so much FUD that it's clear it's an agenda at play rather than any objectivity.

[johnc]

You've got it all backwards. Hiding his name would be the equivalent of nvidia hiding THEIR name. His message is plain text, that like open source. If he GAVE HIS NAME, but encrypted the message, that would be like being nvidia.

I don't have anything backwards. I put forth an analogy to prove a point. Clearly you put forth a different analogy to prove a different point, which, while true, has nothing to do with the point I was making. This guy chooses to obfuscate his identity because he prefers privacy. That NVIDIA chooses to keep their IP secret for their benefit is no greater an evil.

I agree that he should be judged on his message alone which, as far as I can tell, is basically "NVIDIA uses closed source boo hoo". Yeah, tell us something we haven't known already.

[johnc]

And my problems with the anonymous source is what he/she says, which is at it's base an interesting observation of facts (NVidia is paying CodeSourcery to implement OpenACC support but only for their own PTX) but then he/she sprinkles it with so much FUD that it's clear it's an agenda at play rather than any objectivity.

I have to admit that I was amused about how he slipped in there that AMD is just so awesome.


AMD doesn't even know Linux exists.

[Marc Driftmeyer]

FWIW: You don't see this crap by Nvidia being pulled with LLVM/Clang community because they don't have jack squat influence over the direction of the project(s).

[johnc]

FWIW: You don't see this crap by Nvidia being pulled with LLVM/Clang community because they don't have jack squat influence over the direction of the project(s).

https://developer.nvidia.com/cuda-llvm-compiler