The summer is over, but my project is not! For the past 4 weeks, I’ve been working mostly on two things:
- translation to CUDA code of the most common matrix allocation function
- translation to CUDA code of the SpMV algorithm (Sparse Matrix to dense Vector multiplication)
The first being a function which translates a given COO-formatted matrix into an RSB-formatted one, I used heavily the CUDA Unified Memory in order to ease my job and it seemed to work. Then, I found working on the SpMV algorithm a lot harder than I thought since the implementation was contained (and replicated) inside a C file of 50’000+ lines. I translated a simpler version into a CUDA kernel and it still took quite a bit of time to do so, since I discovered a memory issue in it and there was not enough time to fix it. For this reason, I am not able to show you results, but I can assure you that it will be fixed in the future.
This bing said, I would still like to thank all the people that have helped me throughout this experience. First of all, dr. Ezhilmathi Krishnasamy of the University of Luxembourg, my mentor, really helpful, really patient and always with a clever suggestion to improve my work. Kostantinos and Alexander, my Summer-of-HPC-mates, who always gave me a nice joke to cheer me up while working. Also, all the guys at the PCOG department of the University of Luxembourg. A really big hug and a big thank you for the wonderful experience. I hope to see you again in the future.