Optimize none
base privatization, add eager Vojdani privatization
#1552
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
none
privatizationsThe
none
privatization in base analysis is a sliced copy of the old unsound Vojdani privatization (which was removed in #736). Thus, it is still based onsync
:write_global
doesn't side-effect, but adds to local state andsync
does all the side effects. This is an unnecessarily roundabout way to be (almost) as imprecise as possible.What's perhaps worse is that it
sync
s unconditionally (so not just at join points) and each time iterates over the entirecpa
. This could be a stupid hidden cost when wanting super-fast analysis without any privatization, e.g. ourlarge-program
example conf. This is quadratically bad in program size (nodes * variables) compared toprotection
privatization almost none of this would happen.In this PR, I've added
NonePriv2
, which doesn't rely onsync
(except for the inevitable branched thread creation) and surprisingly fails fewer tests when made the default privatization on our regression suite (it's more precise or more sound?!).It's constructed according to some old traces related work and still has global variables in local states (but reads them joined with global unknowns).
Finally, I've added
NonePriv3
, which never puts any globals into local states (in multi-threaded mode), and hasread_global
andwrite_global
directly usinggetg
andsideg
, respectively.This is what probably should just replace
NonePriv
, but I've added them all right now to make investigating differences and benchmarking slightly easier.Eager Vojdani privatization
#736 removed the old unsound Vojdani privatization which triggered the whole traces research direction. Its unsoundness was due to an attempt at lazy reading (which was only hinted at in some thesis/paper but I cannot find anymore where).
In this PR, I've added
VojdaniPriv
, which is the sound Vojdani privatization as described in his thesis and various papers. This does eager reading, which should avoid the unsoundness. It also doesn't usesync
because it handleslock
andunlock
itself, but that may also be a mistake when trying to represent the original analysis faithfully.I don't know if we'd want to have this back, but it could also make for interesting benchmarking.