- This past week I updated the SERVICE site with pictures while continuing my research into virtualizing GPUs. I have found an interesting method that might allow more control over what is scheduled on a single GPU. In particular, it could allow multiple GPU kernels to be merged into one and run together. Combined with ideas from GViM (and Ocelot), this could be done at runtime by retrieving PTX code for the kernels and automatically merging them.
- In addition, I met with Dr, Feng and Dr. Gardner and decided that I should focus on finishing the lessons (as well as determining the research topic for my master's thesis).
- For this next week, I will try my hand at manually merging kernels and will read the PTX specification to try to merge the kernels at that level. Also, I will get in touch with the other members of the SERVICE project so as to get the most up-to-date lessons.