Closed psathyrella closed 8 years ago
All of our time is spent calculating the DP table. Main gain has been avoiding calculating it.
This issue concerns speeding calculation of DP table.
E says: loop unrolling?
Looks like the gcc option -Ofast
is, among many other things, unrolling loops that it deems useful.
Yes, though it can only fully unroll loops if the number of iterations is fixed at compile time.
huh, this is still open... well I spent a bunch of time optimizing/caching for the loops a month or so ago. There's always more to do, but I don't need this open any more.
i.e. in trellis::Viterbi/Forward see if we can reduce the number of times we go through the loops.
This may or may not be equivalent to the pre-alignment they talk about in the repgenhmm paper (i.e. assuming each position in the sequence can only be a certain state in the V or J)