As part of our research in Phylanx, our team has worked to create GPU support in Blaze. In this seminar Jules Pénchot, an intern from Université Paris-Saclay, presents his library, Blaze CUDA. Continue reading
Tag: CRTP
Phylanx Seminar: Revealing the Magic of Blaze
Over the past several weeks, many team members have been asking how aspects of the Phylanx project are implemented. In this seminar, Hartmut explains the techniques used by Blaze to optimize matrix operations. He implements a matrix addition example which uses types, templates, and curiously recurring template patterns (CRTP) to reduce the number of temporaries made during the execution of the code. By avoiding these extra allocations in a compiler friendly way, Blaze can drastically reduce the amount of time it takes perform matrix operations. You can find links to the seminar materials below:
Seminar Video: https://www.youtube.com/watch?v=F5E8cOqHmRU
Naive Matrix Implementation: http://stellar.cct.lsu.edu/files/phylanx_seminars/03.21.18_seminar_blaze_magic/naive_matrix_03.21.18.cpp
Optimized Matrix Implementation: http://stellar.cct.lsu.edu/files/phylanx_seminars/03.21.18_seminar_blaze_magic/optimized_matrix_03.21.18.cpp