Sandbox for matrix multiplication, just having fun with fundamental algorithms.
-
For the CPU version (.cpp), you need to have CBLAS installed (if you want to compare against
cblas_dgemm(...)). IJK, IJK (with transposed B), and IKJ approaches are implemented -
The GPU version (.cu), uses shared memory to ensure data reuse, and the tiling approach. Of course the best option is to call
cublasDgemm(...)instead of reinventing the wheel
NOTE: In both implementations I use one-dimensional dynamically allocated arrays.