-
Notifications
You must be signed in to change notification settings - Fork 435
Object Clustering
Mohamad Barbar edited this page Dec 24, 2021
·
5 revisions
For staged analyses (currently only for flow-sensitive analysis), objects can be better mapped to identifiers for use in bit-vectors according to the auxiliary analysis's results.
Merged; available in mainline.
-node-alloc-strat=dense
must be used.
-
-cluster-fs
: perform clustering for flow-sensitive analysis based on auxiliary Andersen's analysis (default: false). -
-cluster-regioned
: cluster one region at a time (default: true). -
-cluster-region-aligned
: align first assigned identifier in each region to the word boundary (default: true). -
-cluster-method
: clustering criterion to use;single
,complete
,average
, orbest
wherebest
performs all 3 and picks the best performing one (default: best). -
-pt-type
: bit-vector type to use for the auxiliary and main analysis;sbv
,bv
(unsupported with clustering), orcbv
(default:sbv
).
For more options to improve flow-sensitive analysis performance, see the page on hash consed points-to sets and the page on versioned staged flow-sensitive analysis.