- changed status to resolved
Improve performance of conversion between mzed_t and mzebd_t
Issue #4
resolved
Conversion between the bitsliced and the flat representation of matrices is expensive enough to cost about 15% of time for multiplication of reasonably sized. We should attempt to improve the performance of this conversion, for instance by constructing some conversion tables which deal with more than //e// bits at a time.
Comments (1)
-
reporter - Log in to comment
3e0c605eafab fixes this issue.