PGI and CLANG compatibility

Dear PGI user forum,

I would like to ask you whether there is compatibility between clang and pgi compilers.

Mainly, I want to ask you if there is OpenACC Support for
Clang/LLVM. If there are already implementations, please provide me this information.

All the best,

Sotirios

Hi Sotirios,

I don’t think Clang has OpenACC support as of yet. For OpenACC, I’d recommend using PGI.

Our LLVM based compilers which use PGI’s front end and an LLVM back-end, are inter-operable with Clang compiled objects so you can build the OpenACC source with PGI and then link with clang objects.

Hope this helps,
Mat

Hi Mat,

in case of a compiler extension which relies on code analysis and transformation then it would not be possible by just linking objects together.

So my next question is whether someone can tap into the PGI toolchain, or it is not allowed since it is not open source.

All the best,

Sotirios

in case of a compiler extension which relies on code analysis and transformation then it would not be possible by just linking objects together.

Sorry but I’m not clear what you mean by this?

In general, mixing PGI OpenACC compiled objects can be linked with objects built with other compilers.

So my next question is whether someone can tap into the PGI toolchain, or it is not allowed since it is not open source.

Sorry, but again I’m not sure what you’re asking. Can you please give more detail on what you’re trying to do?

Thanks,
Mat

Hi Mat,

So I am providing you more info concerning my question.

In case of a compiler which can scale out the execution of GPU programs from one GPU to almost any number automatically. This compiler so called “Mekong” relies on code analysis and transformation. However, this will work only with cuda .

My question is then whether this can be either further generalized and instead of using cuda, you can use openACC. I can imagine if someone can connect those two things then the user can write code with pragmas (openACC) for many gpus (Mekong). My question is if this is allowed at some extend from pgi compiler since it is not open source.

All the best,

Sotirios

Hi Sotirios,

Now that I’m back in the office, I had a bit more time to look at Mekong. Are you part of the this project? If so, I’m thinking the best route would be to contact your NVIDIA collaborators (we’ve worked with them before) and have them start a dialog with PGI on how we could help support your efforts.

Best Regards,
Mat