As usual, a short hint: For the actual matrix data, you should use float[] and double[] arrays. Due to the differences between Java and C, a cuComplex[] array won’t work well here. Apart from that, the basic setup and workflow should be similar.
Speaking of things I’d love to see:
The SparseHoseCSR API would be very nice to include the ability to get and set values.
I say that because, once a sparse matrix has been multiplied, it is sometimes valuable to set diagonals to 0.
Thanks!
*** Edit ***
Speaking of things I’d love to see:
The SparseHostCSR API would be very nice to include the ability to get and set values.
I say that because, once a sparse matrix has been multiplied, it is sometimes valuable to set diagonals to 0.
Thanks!
Sure, the current classes are VERY basic, just wrappers and conversion functions. But maybe I should mention that at least the differentiation between a “Matrix” and a “Mutable Matrix” was on purpose: A CSR matrix is usually considered to be not mutable, because setting values in a CSR matrix usually means that the internal arrays have to be reallocated. That’s not a problem, of course, but won’t be very efficient - e.g. for each element of the diagonal that is set to 0, all arrays have be be resized by -1 and copied around. It’s certainly a trade-off, but maybe I’ll have a look at how real matrix libraries solve this.