Feature #869
SmallFpImpl: different repr for computation and for storage?
Start date:
22 Apr 2016
Due date:
% Done:
0%
Estimated time:
Description
I would like to use unsigned short
to store SmallFp
values, but still have a decent range i.e. without the restriction that the square of the modulus fit in the "store representation".
I want to do a big computation (needs more than 8Gbytes, and coeffs are modulo some prime with at most 15 bits); currently the coeff values are stored as unsigned int
(presumably 4bytes) when presumably they could be fitted into unsigned short
... though I'm not actually sure that DistrMPolyInlFpPP
can actually make use of this (depends on alignment inside a structure).
Might it make sense to use two different reprs for (external) storage of SmallFp
values, and for (internal) computation with those values?