Skip to main content
9 events
when toggle format what by license comment
Jan 8, 2023 at 14:41 vote accept bobinthebox
Jan 8, 2023 at 14:41 comment added bobinthebox The GitHub repo of the course if the following: github.com/Foundations-of-HPC/Foundations_of_HPC_2022
Jan 7, 2023 at 14:17 comment added Victor Eijkhout @bobinthebox For my own information, which course are you taking? Institution, number?
Jan 5, 2023 at 15:47 comment added Victor Eijkhout Your interpretation is largely correct. It is indeed a big pointwise reduction, followed by a scatter. However, the intermediate result of the reduction is not stored on process zero or anywhere. The only thing you know is the inputs and the final output. Indeed, for efficiency the intermediate result is probably not assembled anywhere. Btw, you have p vectors, not p-1.
Jan 5, 2023 at 15:44 comment added Victor Eijkhout I think 2011 predates both MPI 3 and (of course) 4. But for the basics you can indeed use that book.
Jan 5, 2023 at 15:17 comment added bobinthebox That book is also in the bibliography of my course (thanks for this). I was referring to Pacheco's book "An Introduction to Parallel Programming", the one of 2011, not the original one. Do you think this one is old too? @VictorEijkhout
Jan 5, 2023 at 15:15 comment added bobinthebox Thanks for your review. I'd like to be sure to understand how Reduce_scatter works in this case. I have p-1 vectors of size m whose sum is the solution, one per process. The reduction is applied then entry-wise on each vector, ending up with process 0 having the correct result, and then the result is split in p segments, where the i-th segment has size local_m and sent to rank i. Is that correct?
Jan 4, 2023 at 18:22 history edited Victor Eijkhout CC BY-SA 4.0
added 127 characters in body
Jan 4, 2023 at 16:26 history answered Victor Eijkhout CC BY-SA 4.0