Library generation for linear transforms
Yevgen Voronenko · 2008
The roots of this thesis lie in the grand vision of my advisor Markus Püschel. Back in 2000, after listening to one of his talks, I thought it was a neat idea to generate a full library from some very high-level rules that define the algorithm. Yet it was not at all clear that this would be possible. I thank Markus for providing me with the never-ending source of inspiration and countless weekends proofreading drafts of various papers and this thesis. This work would not be possible without major contributions made by Franz Franchetti, who developed a number of key mathematical insights, on which I heavily built my work. With Franz we spent long hours arguing about how our rewriting systems should and should not work. I would like to extend my special thanks to Peter Tang, Dmitry Baksheev, and Victor Pasko whom I had a chance to work with at Intel. Peter provided me with great encouragement to work on new and exciting projects, just to “see what happens”. I enjoyed having worked with Dmitry and Victor, and really hope to be able to continue our fruitful collaboration. I bring my sincere apologies for letting you guys suffer through my e-mail lag. Many of the interesting experimental results of this thesis are possible due to work by Frédéric de Mesmay, who helped me just because it was interesting, and Srinivas Chellappa who maintained