A letter to PAL developers

(email sent to PAL developer list)

———–

Hi all,

I hope everyone is having a great summer. This email is going out to the 32 developers who have put significant time into PAL so far. Pretty incredible how much the library has progressed in the last month!

Ola has put together some scripts and bench-mark code to show where the library stands right now. (not completely up to date) (James’ contribution this morning provides massive improvement to some of the image processing functions)

https://parallella.github.io/pal-stats/summary.html

If you want your free Parallella board, send me your shipping address and we will get you a board asap. Most of the functions don’t meet the performance criteria, but I don’t have time to be stingy.

We will be shipping this US postal so there may be some import charges depending on where you live? (this unfortunately has to be your responsibility)

Initially, we will send out one board per person. Additional boards (for those who contributed more than one function will be sent out later). If you want them shipped now, demonstrate how you have achieved at least  60% 40% [UPDATE: forgot that we lowered target] of peak theoretical processor performance on each function.

Let me know if you want a different communication channel set up. So far most things have been managed through GitHub and it seems to work reasonably well. If you think we should have a mailing list approach instead/addition let me know.

Thanks,
Andreas

One Comment

  • Adam says:

    Your example of optimized pal function p_popcount.c does not seem to look like optimized when I looked into the summary report located in https://parallella.github.io/pal-stats/summary.html. Twice the size of x86 and 100x longer to process. I believe comparison test is not done properly. Please see further details in https://github.com/parallella/pal/issues/206 . A little bit clarification on how to compare performance. If I am correct, the best comparison is to compare size and processing speed of a function between math.h library and pal.h library.
    About communication channel, git pal looks okay to me.

Leave a Reply to Adam Cancel Reply