usx95 wrote:

> I also wonder if it is better to do something else for functions with large 
> CFGs. E.g., instead of entirely skipping them, we could just visit all nodes 
> ones and not iterate to a fixed-point.

I have tried doing that earlier to reduce improve performance on LLVM but had 
no success. The primary bottleneck in practice seems to be the number of join 
operations we need to do even in a single iteration and this is proportional to 
the number of blocks. 
I would recommend to not do anything more clever here. Set this limit 
relatively high would make almost all human-written code in the scope of the 
analysis.

https://github.com/llvm/llvm-project/pull/170444
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to