DaphneDSL: Integer matrix multiplication

[RetoKrummenacher]

Matrix multiplication with dense matrices filled with integers (int64_t) is not working.

i1 = reshape(seq(1,5,1),1,5);
print(i1);
i2 = reshape(seq(1,5,1),5,1);
print(i2);
print(i1 @ i2);

Output:

[pdamme]

Thanks for bringing this to our attention, @RetoKrummenacher. Currently, we only have matrix multiplication kernels for f32 and f64 value types (based on BLAS routines). We need either a general implementation of matrix multiplication for a generic value type, or specializations for typical integer value types.

[KostasBitsakos]

Hello, as far as Ι checked, blas does not natively support integer types.
Perhaps we could look for another third party library that supports integers matrix multiplication.

For now, I have implemented a solution, that uses the CastObj.h to transform integer inputs to double or float and then multiply the input matrixes.
I have also implemented the tests. It works fine, but of course it is not the optimal solution.

I could upload it and make a pull request.