<div dir="ltr"><div dir="ltr"><div class="gmail_default" style="font-family:courier new,monospace;font-size:small"><br></div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Sat, Feb 6, 2021 at 2:29 PM Johannes Doerfert via llvm-dev <<a href="mailto:llvm-dev@lists.llvm.org">llvm-dev@lists.llvm.org</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">Hi Konstantin,<br>
<br>
I didn't find the time to write new GSoC projects for 2021 yet but<br>
if you are interested we could probably set one up in this area. I<br>
also CC'ed Mircea who might be interested in this too, maybe as<br>
(co-)mentor.<br>
<br>
We could look at loop transformations such as unrolling and fusion,<br>
similar to the inliner work. Best case, we can distill a heuristic<br>
out of a model we learned. We could also look at pass selection and<br>
ordering. We started last year and I was hoping to continue. You<br>
might want to watch <a href="https://youtu.be/TSMputNvHlk?t=617" rel="noreferrer" target="_blank">https://youtu.be/TSMputNvHlk?t=617</a> <br>
<<a href="https://youtu.be/TSMputNvHlk?t=617" rel="noreferrer" target="_blank">https://youtu.be/TSMputNvHlk?t=617</a>> and<br>
<a href="https://youtu.be/nxfew3hsMFM?t=1435" rel="noreferrer" target="_blank">https://youtu.be/nxfew3hsMFM?t=1435</a> <<a href="https://youtu.be/nxfew3hsMFM?t=1435" rel="noreferrer" target="_blank">https://youtu.be/nxfew3hsMFM?t=1435</a>> .<br>
<br>
In case your interested in a runtime topic, I really would love to<br>
have a predictor for grid/block/thread block size for (OpenMP) GPU<br>
kernels. We are having real trouble on that end.<br>
<br></blockquote><div><div class="gmail_default" style="font-family:"courier new",monospace;font-size:small">Hi,</div><div class="gmail_default" style="font-family:"courier new",monospace;font-size:small"><br></div><div class="gmail_default" style="font-family:"courier new",monospace;font-size:small">We are working on an ML model that can predict the profitability of offloading a kernel to GPU. I feel that this problem is very much related. This problem will have the same challenges in terms of feature engineering and data preparation the one we are handling for our work.</div><div class="gmail_default" style="font-family:"courier new",monospace;font-size:small"><br></div><div class="gmail_default" style="font-family:"courier new",monospace;font-size:small">Abid</div><br></div><div><br></div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
I also would like to look at ML use in testing and CI.<br>
<br>
Let me know what area sounds most interesting to you and we can<br>
take it from there.<br>
<br>
~ Johannes<br>
<br>
<br>
On 2/6/21 4:35 AM, Сидоров , Константин Сергеевич via llvm-dev wrote:<br>
> Dear all,<br>
><br>
> I would like to continue the discussion of the GSoC project I mentioned in<br>
> the previous email. Now, when I know my way around the LLVM codebase, I<br>
> would like to propose the first draft of the plan:<br>
><br>
> * Improving heuristics for existing passes – to start the discussion, I<br>
> propose to start the project by working on `MLInlineAdvisor` (as far as I<br>
> understand, in this class the ML infrastructure is already developed, and<br>
> thus it seems to be a good idea to start there) and after that switching to<br>
> the other passes (e.g., `LoopVectorizationPlanner` seems to be a good<br>
> candidate for such an approach, and `LoopRotate` class contains a<br>
> profitability heuristic which could also be studied deeper).<br>
> * Machine learning models to select the optimizations – to the best of my<br>
> understanding, the key concept here is the pass manager, but here I don't<br>
> quite understand the technical details of deciding which optimization to<br>
> select. For this reason, I would like to discuss this part more thoroughly.<br>
><br>
> If the project mentors are reading this mailing list and are interested in<br>
> the discussion, can we start the discussion here?<br>
><br>
> By the way – I would like to thank Stefanos for the comprehensive response<br>
> to my previous questions that helped me to get started :)<br>
><br>
> Looking forward to a further discussion,<br>
> Konstantin Sidorov<br>
><br>
> вт, 19 янв. 2021 г. в 07:04, Сидоров , Константин Сергеевич <<br>
> <a href="mailto:sidorov.ks@phystech.edu" target="_blank">sidorov.ks@phystech.edu</a>>:<br>
><br>
>> Dear all,<br>
>><br>
>> My name is Konstantin Sidorov, and I am a graduate student in Mathematics<br>
>> at Moscow Institute of Physics and Technology.<br>
>><br>
>> I would like to work on a project "Machine learning and compiler<br>
>> optimizations: using inter-procedural analysis to select optimizations"<br>
>> during the Google Summer of Code 2021.<br>
>><br>
>> I have an extensive background relevant to this project - in particular:<br>
>><br>
>> * I have already participated in GSoC before in 2017 with mlpack<br>
>> organization on the project "Augmented RNNs":<br>
>> <a href="https://summerofcode.withgoogle.com/archive/2017/projects/4583913502539776/" rel="noreferrer" target="_blank">https://summerofcode.withgoogle.com/archive/2017/projects/4583913502539776/</a><br>
>> * In 2019 I have graduated from the Yandex School of Data Analysis — a<br>
>> two-year program in Data Analysis by Yandex (the leading Russian search<br>
>> engine); more info on the curriculum could be also found at<br>
>> <a href="https://yandexdataschool.com/" rel="noreferrer" target="_blank">https://yandexdataschool.com/</a>.<br>
>> * I have also been working as a software engineer at Adeptik from July<br>
>> 2018 to date, where I have predominantly worked on projects on applied<br>
>> combinatorial optimization problems, such as vehicle-routing problems or<br>
>> supply chain modeling. In particular, I have had experience with both<br>
>> metaheuristic algorithms (e.g., local search or genetic algorithms) and<br>
>> more "traditional" mathematical modeling (e.g., linear programming or<br>
>> constraint programming).<br>
>><br>
>> I would like to discuss this project in more detail. While it is hard to<br>
>> discuss any kind of exact plan at this stage, I already have two questions<br>
>> concerning this project:<br>
>><br>
>> (1) I have set up an LLVM dev environment, but I am unsure what to do<br>
>> next. Could you advise me on any simple (and, preferably, relevant) tasks<br>
>> to work on?<br>
>> (2) Could you suggest any learning materials to improve the understanding<br>
>> of "low-level" concepts? (E.g., CPU concepts such as caching and SIMD)<br>
>><br>
>> Best regards,<br>
>> Konstantin Sidorov<br>
>><br>
><br>
> _______________________________________________<br>
> LLVM Developers mailing list<br>
> <a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a><br>
> <a href="https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev" rel="noreferrer" target="_blank">https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev</a><br>
_______________________________________________<br>
LLVM Developers mailing list<br>
<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a><br>
<a href="https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev" rel="noreferrer" target="_blank">https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev</a><br>
</blockquote></div><br clear="all"><div><br></div>-- <br><div dir="ltr" class="gmail_signature"><div dir="ltr"><div>Abid M. Malik<br>******************************************************<br>"I have learned silence from the talkative, toleration from the intolerant, and kindness from the unkind"---Gibran<br>"Success is not for the chosen few, but for the few who choose" --- John Maxwell<br>"Being a good person does not depend on your religion or status in life, your race or skin color, political views or culture. IT DEPENDS ON HOW GOOD YOU TREAT OTHERS"--- Abid<br>"The Universe is talking to us, and the language of the Universe is mathematics."----Abid<br> </div></div></div></div>