Originally in #1430, but as that needs more work, creating a separate PR. Sadly, in most cases (except LAFOV), the functions that I parallelised were pretty fast already.
There's more work to do += etc, as it might be a bad idea to parallelise this for VectorWithOffset, as then the nD-Array functions could create a lot of threads. I guess it needs specialisation for Array<1,elemT>.
Will likely be rebased on top of #1442 once that is merged, although I could merge as well as it's independent.
Originally in #1430, but as that needs more work, creating a separate PR. Sadly, in most cases (except LAFOV), the functions that I parallelised were pretty fast already.
There's more work to do
+=
etc, as it might be a bad idea to parallelise this forVectorWithOffset
, as then the nD-Array functions could create a lot of threads. I guess it needs specialisation forArray<1,elemT>
.Will likely be rebased on top of #1442 once that is merged, although I could merge as well as it's independent.
@markus-jehl I haven't parallelised
is_contiguous
, as that's also a bit more work, see e.g. https://stackoverflow.com/questions/9793791/parallel-openmp-loop-with-break-statement. Alternatively, we introduce an extra member (if it turns out to be worth it, as that'd occur a lot in current nD Arrays).