Closed mulimoen closed 4 years ago
Thanks, that's a very clean implementation.
This might not be the most efficient, but the performance should be predictable in this case.
Indeed, it might be best to dynamically choose which matrix should be converted, but this would need benchmarks first. I think it's best to first have this implementation, then later we'll see if the performance needs to be improved. And since the performance is predictable, this means the calling code is free to perform its own manipulations to enforce best performance.
This is a slightly modified version of the one in #209, which uses some symmetries for the
CSC^2
->CSR^2
case.The incompatible storage case is now handled by converting the second matrix to the opposite storage. This might not be the most efficient, but the performance should be predictable in this case.
Closes #209