Podcast: Porting Math Libraries Across Heterogeneous Architectures
28 May 2020
Codeplay Principal Software Engineer Mehdi Goli joined the Code Together podcast for a show about his work on porting the oneMKL math kernel library to run on Nvidia GPUs using SYCL and DPC++.
Mehdi gave a bit of background on his experience in implementing math libraries. He talks about how important it is to have a wide set of libraries available to developers and how this fosters a healthy developer community.
This quote from the podcast emphasizes why SYCL is so important for the future of portable heterogeneous programming:
"I think the missing key feature for existing math kernel libraries is portability across heterogeneous platforms. And I think this is because there's no common language to abstract out the memory model and execution model from various heterogeneous devices. So, in otherwords, how to glue together different parts of code written in different hardware. And I believe this problem can be solved by using SYCL as the unifying programming model."
Listen to or download the podcast.