SERVICE

Gabriel's Journals

July 04, 2010

  • This last week I was able to successfully merge some simple CUDA kernels by hand. I also read generated the PTX code for them and was also able to merge those by hand, producing code very similar to what the generated PTX code from my merged CUDA kernels was.
  • Unfortunately, I am still waiting to hear back from a couple of the other memebers. I hope to do so soon so that I can begin finalizing the lessons.
  • I met with Dr. Feng and Dr. Gardner and we discussed my progress. There is a new architecture for NVIDIA cards that allows for multiple kernels to be run at once, so I may start looking into another direction for my master's thesis. Automatically merging kernels in software would be a backwards compatible approach, but should not be necessary for new GPUs.
  • Therefore, this next week I will look into what other aspects of GPU virtualization I can focus on. I will also begin looking through the lessons I have and seeing what needs to be done.