[Paraview] Building ParaView v5.1.2 on Linux from source with support for GPU and SWR.

[Chuck Atkins]

Hi Frank,

I tracked down the problem.  The way the superbuild was calling ./configure
for mesa was causing the default -O2 flags set by mesa's ./configure to get
dropped, thus resulting in an unoptimized mesa build.  I just merged a fix
for both issues into the superbuild.

----------
Chuck Atkins
Staff R&D Engineer, Scientific Computing
Kitware, Inc.


On Fri, Oct 7, 2016 at 4:39 PM, Chuck Atkins <chuck.atkins at kitware.com>
wrote:

> which are not copied into the archive when packing the binaries using:
>>> “ctest -V -R cpack-paraviewsdk-TGZ”
>>>
>>
>> This should have been fixed a while ago.  I'll try to reproduce it and
>> make sure the packaging is correct.
>>
>
> I see the same problem.  I have a fix for the packaging and will get it up
> shortly.  It probably won't make it for the RC tonight but it will be fixed
> before the release.
>
>
> 4.     Compared to the ParaView v5.1.2 binary for Linux, the compiled
>>> version with mesa v12.0.3 is substantially slower than with the OpenSWR
>>> version shipped with the binary.
>>>
>> That certainly should not be the case.
>>
>
> The good news is that I was just able to reproduce the problem.  Even
> better I could isolate it to the mesa build from the superbuild.  Swapping
> out the mesa libs in the release binary on the web site, on a single
> 14-Core Haswell CPU I see a difference of 25MiPolys / sec with the
> superbuild libraries vs 118MiPolys / sec with the website binaries.
> That's roughly a 4.5x performance difference, i.e. not good.  I don't know
> yet if it's a mesa problem or how we build it.  I'll dig deeper and let you
> know.
>
> Thanks for the bug!
>
> - Chuck
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/paraview/attachments/20161011/a0bd3e0c/attachment.html>