-
Notifications
You must be signed in to change notification settings - Fork 32
RepresentationInference
Felix S. Klock II edited this page Jul 28, 2013
·
1 revision
Twobit's representation inference is a prototype that has been operating for several years now. In combination with the GreatPrimOpCleanUp, it is time to:
- take advantage of branch direction
- enlarge the tables used by representation inference
- widen inferred representation types as necessary to avoid the worst-case cubic time of this flow analysis
Note: Although the worst-case time of the algorithm should be cubic, some experiments that Will conducted while investigating ticket:419 led him to believe the worst-case time as implemented was something like O(n^8^) or O(n^9^).
Widening to object was implemented in v0.96, effectively limiting representation inference to linear time, but less drastic forms of widening should be implemented for future releases.