[llvmlinux] More VLAIS code ...

Jan-Simon Möller jsmoeller at linuxfoundation.org
Wed Nov 20 11:43:52 UTC 2013

Hi all,

On Wednesday 20 November 2013 10:29:24 Renato Golin wrote:
> On 20 November 2013 10:18, Tinti <viniciustinti at gmail.com> wrote:
> > What kind of adaptation do you want? If you have a patch or something, I

I have no patch at hand just now. I'd need to sit down a bit more on it.

> >> can talk to him, or you can send him and copy me, if you want. I think
> >> the
> >> latter is better, since I'm not directly involved in the issue, myself.
> > 
> > This code inserts a VLAIS.
> Unless we have an alternative, I don't think "please, re-implement your
> code without VLAIS" would cut.

Yes full ack. This is just an example how the VLAIS code in crypto propagates 
- again. 

> The main issues are:
>  * VLA is much simpler to code and understand
>  * GCC produces very good code from it

We need to educate here the difference between what we call VLAIS and flexible 
members in structures / VLA .

> So, at the toolchain group, we're tackling this problem from two sides:
>  * Trying to find alternatives to VLAIS that are simple and perform well,
> and convincing kernel folks to use it
>  * Improving GCC to generate good code on those cases and hopefully emit a
> warning on VLAIS
> But for now, we'll have to suffer a bit... Which means, for every VLAIS
> patch, we should email back the submitter asking for a different approach.
> Submitting an alternative would make it a lot easier to be accepted. Also,
> feel free to copy me on any email, especially if it's a Linaro engineer.

Well - "suffer a bit" is "quite a bit" in this case. It is the WIFI mac80211 
part. So unless we fix it (or revert this patch :( ), we have no wifi.
Which does not cut for android either.

Ok, let me draft up an email to the authors ... 

> I'm sorry I can't do much more than that. They have some very hard
> deadlines for feature support and performance improvements, and I don't
> want to be in between them and their targets, as I know I'll be shot very
> quickly and thoroughly.

Yes, indeed. But if we don't speak up now, we have no chance lateron.
This is why we need to start tracking linux-next, too.


More information about the LLVMLinux mailing list