Re: Calc research & funcalc ...

Hi Michael,

Just a little note of warning. The forum you're posting to is one of
developers who typically intersperse their replies into edited-down
E-mail contexts. They also read through the whole E-mail to ensure they
got it all.

Thanks for the warning. So you saying we should not include the previous part of the conversation 
in our replies?

Sure, we do this each year - always fun; this year in Albania -
starting early next week: do join us. Next year's location to be
announced but will happen.

Unfortunately, I have a big deadline which is due this Friday, so I am not going anywhere :( I’d 
very much like to join next year though!

Correct; so far. Always more to do there. We'd love help as well - if
you have students interested in the more vexed engineering of
re-assembling our aircraft in-flight =) the interesting (and partially
solved) issue of more deeply unwinding the implicit intersections for
more precise dependencies and a decent view of the dependency graph is
something we'd love help with eg.

Haha! Probably not how we should sell it to them though. It is surprisingly difficult to “sell” the 
idea to the students in projects when you
mention the word “spreadsheets”. Most people know what a spreadsheet is but do not always 
understand or appreciate the complex
wiring behind the scenes. Last time, we threw around a bit parallelism and compiler stuff in the 
hopes that it would attract some students.

I believe some students would be interested in helping out, especially on a real-world application. 
I’ll pass it on to my supervisor and
see if he has been approached by any students. As previously mentioned, my own time is 
unfortunately exceedingly scarce.

While Calc seems daunting, it's really only ~400k lines of code and
attacking a reasonably well understood problem (vs. say writer layout =).

That should be fine :) Especially if a student hones in on a specific feature and with the major 
refactoring that Calc has undergone.

Are the benchmarks from the OpenCL kernel code publicly available? We would love to have some idea 
of the performance boosts
on all the six spreadsheets.

Mvh/Best regards,

Alexander Asp Bock, PhD student
Computer Science Department<https://computerscience.wikit.itu.dk/>
IT University of Copenhagen<http://en.itu.dk/>

On 22 Sep 2018, at 19.19, Michael Meeks 
<michael.meeks@collabora.com<mailto:michael.meeks@collabora.com>> wrote:

Hi Alexander,

On 21/09/18 12:57, Alexander Bock wrote:
Yes, that’s very close :) Thanks! I can see there is also a tentative
LibreOffice Conference

Sure, we do this each year - always fun; this year in Albania -
starting early next week: do join us. Next year's location to be
announced but will happen.

The threaded variant you refer to is the one from the slides right? As I
understand, the interpreter is not fully parallel but runs cell
arrays/formula groups in parallel.

Correct; so far. Always more to do there. We'd love help as well - if
you have students interested in the more vexed engineering of
re-assembling our aircraft in-flight =) the interesting (and partially
solved) issue of more deeply unwinding the implicit intersections for
more precise dependencies and a decent view of the dependency graph is
something we'd love help with eg.

While Calc seems daunting, it's really only ~400k lines of code and
attacking a reasonably well understood problem (vs. say writer layout =).

ATB,

Michael.

--
michael.meeks@collabora.com<mailto:michael.meeks@collabora.com> <><, GM Collabora Productivity
Hangout: mejmeeks@gmail.com<mailto:mejmeeks@gmail.com>, Skype: mmeeks
(M) +44 7795 666 147 - timezone usually UK / Europe