On 03/21/06 03:25, Laurent GUERBY wrote: > A casual read of tree-vrp.c showed that symbolic_range_p is mostly > used to do nothing, did I miss something? May be it's in another file. > That's correct. We mostly punt on symbolic ranges because they are fairly expensive to track. We do try to use symbolic range information in some places like 'if (a_3 > b_10)'. If b_10's range is [a_3, +INF] then we know the conditional is false.
Do you have anything specific in mind that you would like to track? Any kind of heavy symbolic processing will likely be very expensive. And VRP already is a hog.
