Yes, we need and welcome any user input. The purpose of starting with a basic version release of MAGMA was to test the level of interest in the library and get as much user feedback as possible. The interest so far has been very high and we plan to quickly extend MAGMA's functionality.
Although the one-sided factorizations can be accelerated using GPUs, they run fine on (homogeneous x86-based) multicores as well. This is not the case for the 2-sided though as they are slow on multicores because of memory-bound operations. Hybrid GPU-based computing can overcome these bottlenecks to get huge performance improvements. Therefore we think the highest impact will be to release the two-sided factorizations. They are the 1st step in developing eigensolvers and we have had many requests for them. We plan to release them by November 14.
magma_roadmap.pdf
- Slides summarizing MAGMA's current work and future plans
- (518.03 KiB) Downloaded 960 times