LLVM lazy load #1694
LLVM lazy load #1694
Conversation
robdockins
force-pushed
the
rwd/llvm-lazy-load
branch
3 times, most recently
from
93cfc3c to
55c34fc
Compare
| -- | Look up the 'L.Define' for a symbol defined in an 'LLVMModule' | ||
| lookupFunctionDef :: LLVMModule arch -> String -> Maybe L.Define | ||
| lookupFunctionDef lm nm = | ||
| find ((fromString nm ==) . L.defName) $ L.modDefines $ modAST lm |
There was a problem hiding this comment.
We've gone from a Map lookup to a list lookup, which seems somewhat unfortunate. Could this be a performance bottleneck in practice?
There was a problem hiding this comment.
I'm not sure; I don't have a good sense of how hot this lookup function is. I'd be surprised if it is a major bottleneck, but it's a shame to be going backward in terms of algorithmic complexity.
Ideally, the llvm-pretty library, or some intermediate layer would provide better lookup functionality for this stuff so it doesn't have to be redone everywhere downstream.
There was a problem hiding this comment.
@eddywestbrook, would you be willing to take a quick look at these heapster-releated changes and give a sense of whether this is a concern for you?
robdockins
force-pushed
the
rwd/llvm-lazy-load
branch
from
55c34fc to
4f4e1d7
Compare
|
I'm not sure what happened with the MacOS / saw-remote-api test, but given it passed on other platforms I'm willing to write it off as a transient failure. |
Pull in Crucible changes that enable lazy loading of Crucible CFGs from LLVM.