x3042/ExactODEReduction.jl

Rewrite matrices class to utilize better cache access

sumiya11 opened this issue · 2 comments

sumiya11 commented
Rewrite matrices class to utilize better cache access
sumiya11 commented
  1. Use continuous (or even static) memory for vectors
  2. Specify the algorithm for finite fields exclusively. Whenever a QQ computation is requested, reduce to modular computations
pogudingleb commented

Comment to the second item: in CLUE, we found out that quite often it is actually faster to work over QQ, see Table 1 on page 9 in https://arxiv.org/pdf/2004.11961.pdf