Smol optimizations #166
Smol optimizations #166
Conversation
i.e. push and pop at the end of the list instead of at the beginning
| BitVecSort256 = BitVecSorts[256] | ||
| BitVecSort264 = BitVecSorts[264] | ||
| BitVecSort512 = BitVecSorts[512] | ||
|
|
There was a problem hiding this comment.
Q: could you help me understand the behavior of this global mutable object when multiple SEVM objects are created?
for example, say we have two sevm objects:
from halmos.sevm import SEVM
sevm1 = SEVM(options)
sevm2 = SEVM(options)
will they have the same cache object, or a different one?
if the former, would it be safe when we run sevm1 and sevm2 in parallel in a thread or process pool?
in any case, can we avoid the use of global mutable objects?
There was a problem hiding this comment.
you're right I didn't think of that, both instances would access the same object because it's at the module level. I can make it read only to avoid this issue
this means that not every BitVecSort will be cached, only the ones deemed most common during initialization
|
I think the CI failures are now due to farcaster being in a bad state (farcasterxyz/contracts#275 (comment)), unrelated to the changes in this PR |
|
yes, i know that. cicada test was canceled because of that. i set fail-fast false, so that we can make sure that all but farcaster test succeed. let's merge after we confirm that. |
2 things:
Saves about 10% on some path-intensive benchmark (172M -> 156M function calls), coming mostly from caching