llvm dev - Apr 2013 - [LLVMdev] [cfe-dev] [Announcement] 3.3 Release Planning!

On Apr 3, 2013, at 2:07 AM, Renato Golin <renato.golin at linaro.org>
wrote:
> On 1 April 2013 22:05, Bill Wendling <wendling at apple.com> wrote:
> We would like to support ARM again.
> 
> Hi Bill,
> 
> Glad you asked! ;)
> 
> I'm getting the test-suite bot green (a few minor tweaks and we're
good) and that should get us well ahead of what we've ever been on ARM.
Though, bootstrapping seem to fails a few check-all tests. I'll look into
that as soon as the test-suite bot is fully green.
> 
> Just to make sure we're talking about the same things, the requirements
for release are:
> 
> * Green direct check-all (clang-native-arm-cortex-a9 bot, green)
> * Green direct test-suite (clang-native-arm-lnt bot, almost green)
> * Green self-host check-all (no bot yet, some failures, will look into it
next)
All of this, yes. :)
> * Green self-host test-suite?
This would be nice.
> * Anything else?
The above are pretty much the requirements for other binaries. Of course, I
expect (hope?) that the community will take each release candidate and test
their own code with it.
> I don't want to have this for release-only, but as continuous
integration. Though, I hope it'll be good for all future releases.
> 
I do too. The progress you're making is great!
> It's probably best to produce the binaries on a Cortex-A9 (Panda ES),
since they're the most common target and the binaries work quite well on
A15s.
> 
I agree. Though I'll also let Jim and Evan chime in on what they think.

Thank you!
-bw
> Sylvestre,
> 
> If we do end up creating ARM binaries for the general public, your input
and expertise will be greatly appreciated! ;)
> 
+1 :)

On Apr 3, 2013, at 1:15 PM, Bill Wendling <wendling at apple.com> wrote:
> On Apr 3, 2013, at 2:07 AM, Renato Golin <renato.golin at linaro.org>
wrote:
> 
>> On 1 April 2013 22:05, Bill Wendling <wendling at apple.com>
wrote:
>> We would like to support ARM again.
>> 
>> Hi Bill,
>> 
>> Glad you asked! ;)
>> 
>> I'm getting the test-suite bot green (a few minor tweaks and
we're good) and that should get us well ahead of what we've ever been on
ARM. Though, bootstrapping seem to fails a few check-all tests. I'll look
into that as soon as the test-suite bot is fully green.
>> 
>> Just to make sure we're talking about the same things, the
requirements for release are:
>> 
>> * Green direct check-all (clang-native-arm-cortex-a9 bot, green)
>> * Green direct test-suite (clang-native-arm-lnt bot, almost green)
>> * Green self-host check-all (no bot yet, some failures, will look into
it next)
> 
> All of this, yes. :)
> 
>> * Green self-host test-suite?
> 
> This would be nice.
> 
>> * Anything else?
> 
> The above are pretty much the requirements for other binaries. Of course, I
expect (hope?) that the community will take each release candidate and test
their own code with it.
> 
>> I don't want to have this for release-only, but as continuous
integration. Though, I hope it'll be good for all future releases.
>> 
> I do too. The progress you're making is great!
> 
>> It's probably best to produce the binaries on a Cortex-A9 (Panda
ES), since they're the most common target and the binaries work quite well
on A15s.
>> 
> I agree. Though I'll also let Jim and Evan chime in on what they think.
cortex-a9 sounds fine. It'd be cool if we verified those binaries run on a
cortex-a8 and a cortex-a15, too. It'd be very, very strange if they
didn't, but hey, catching very, very strange problems is what release
testing is for, right? :)

At least for now, I do think we should constrain "ARM is a supported
target" to mean armv7. For purposes of testing, armv6 is effectively a
different target.

-Jim
> 
> Thank you!
> -bw
> 
>> Sylvestre,
>> 
>> If we do end up creating ARM binaries for the general public, your
input and expertise will be greatly appreciated! ;)
>> 
> +1 :)
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130403/e0d0db13/attachment.html>

On 3 April 2013 21:15, Bill Wendling <wendling at apple.com>
wrote:
>
> > * Green direct check-all (clang-native-arm-cortex-a9 bot, green)
> > * Green direct test-suite (clang-native-arm-lnt bot, almost green)
> > * Green self-host check-all (no bot yet, some failures, will look into
it next)
>
> All of this, yes. :)
Great! This is feasible for the 3.3 time frame.

> > * Green self-host test-suite?
>
> This would be nice.
I'm expecting that, once the self-host check-all is fixed, we won't have
problems here. The few check-all problems (about 3) seem to be regarding
configuration, not real bugs in the compiler, so I'm not expecting any
problem in the self-hosting test-suite.

But it'll take a while to create the bots. There were some, er, issues with
the Arndale (http://www.spinics.net/lists/kvm-arm/msg03723.html) and the
Chromebooks are not the best rack machines... ;)

Anyway, my aim is to get those 4 bots up and green by the 3.3 release.

> cortex-a9 sounds fine. It'd be cool if we verified those binaries
> run on a cortex-a8 and a cortex-a15, too. It'd be very, very strange
> if they didn't, but hey, catching very, very strange problems is what
> release testing is for, right? :)
Actually, this is a good idea, I'll set up a Beagle as a buildbot and see
what happens to it.

> At least for now, I do think we should constrain "ARM is a supported
> target" to mean armv7. For purposes of testing, armv6 is effectively
> a different target.
I agree. LLVM does a pretty good job at compiling for other ARM
architectures, but corner cases can be quite bad, and it's usually on them
that people will compare LLVM to GCC. A warning that we're primarily
focusing on v7 for the time being would be in order.

cheers,
--renato
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130403/edf1bcbf/attachment.html>

Renato,
>> cortex-a9 sounds fine. It'd be cool if we verified those binaries
>> run on a cortex-a8 and a cortex-a15, too. It'd be very, very
strange
>> if they didn't, but hey, catching very, very strange problems is
what
>> release testing is for, right? :)
> Actually, this is a good idea, I'll set up a Beagle as a buildbot and
see
> what happens to it.
AFAIR, there were several beagles in LLVM Lab. You might want to check
with Galina about how live they are.

--
With best regards, Anton Korobeynikov
Faculty of Mathematics and Mechanics, Saint Petersburg State University