Source

nrnr / perfnotes.txt

Author Commit Message Labels Comments Date
Peter Shinners
Cleaned up the finished implementation and test code
Peter Shinners
got the problems worked out. settling on uint16 indices with 32k chunks as the current fastest.
Peter Shinners
rework algorithm to work in smaller chunks and use smaller data types than size_t for indices
Peter Shinners
Minimize code and branches in the search loop for 20% speedup.
Peter Shinners
Minimize code and branches in the search loop for 15% speedup.
Peter Shinners
Code is in working mode Fixed problems with the bisecting and finding subset