1 files changed, 21 insertions, 2 deletions

diff --git a/TODO.md b/TODO.md
index 7ab3287..f6c464b 100644
--- a/TODO.md
+++ b/TODO.md
@@ -42,11 +42,31 @@ It's more of a personal reminder than anything else.
   in non-posix systems
 * better man page
 * find a better way to distribute the large tables, especially khuge
+(or just generate them quickly, see below)
 * webapp (cgi)
 
 ## Technical stuff
 
-### Better pruning tables
+## Performance (optimal solver)
+* Khuge optimal solver: change direction of search when doing so leads to
+less branching (like nxopt). Need to add some info to EstimateData or to
+DfsData (like last moves on inverse/other scramble) and to change some of
+the logic of niss (allow for switching multiple times).
+* Light optimal solver: use drud table instead of khuge, with tricks as above
+and one more trick: if the last move is 180° avoid computing inverse cube
+and just use previous values for all 3 axes.
+
+## Coordinates, symmetries, pruning tables
+* use multiple threads to search for solutions in parallel
+* Faster pruning table generation: keep track of which positions are "nasty"
+(i.e. self-symmetric with respect to the base symmetry coordinate but not
+self-symmetric overall) by adding a function to struct coord and some datafield
+to struct symdata.
+* Faster pruning table generation: multithreading (divide table into large
+sections and use one mutex for each section to avoid too much locking)
+* Cleanup symcoord.c: some coordinates and symdata are never actually used;
+remove also sd_eofbepos and just use sd_coud for khuge (this changes the
+coordinate so the whole table must be generated again!)
 * Use pruning values mod 4 instead of mod 16 (or maybe not, I like the
 current system)
 
@@ -60,4 +80,3 @@ current system)
 * client/server architecture: run a server process in the background so that
   multiple client processess can send it queries and get results; this would
   open up the door for a web-based version or graphical clients
-* use multiple threads to search for solutions in parallel