labath wrote:
> I would like to say I would love for this improvement to land :).
>
> I have observed that on Windows with a fair number (4K+) of small compile
> units the progress reporting completely dominates indexing time due to
> contention (presumably not just in the lock but also the IO layers), to the
> point that disabling it resulted in **very** large speedups (an operation
> which previously took 5s+ now was almost instant).
Thanks. I think that in your case we're limited by the speed of console updates
rather than the progress mutex. We've gotten similar reports for high ("high")
latency ssh connections and unusual console setups, and that's the angle I've
started working this from before I realized the it was bottlenecking the
parallelism. If you have some time, I'd be grateful if you could test this
patch out. The thing want the see is whether this completely resolves your
issue or if you still see a difference between disabled&enabled progress
reports. The reason I'm asking that is that I have a prototype which should
specifically help with the slow console issue, and I'm wondering if that is
necessary, or if this patch is sufficient.
https://github.com/llvm/llvm-project/pull/118953
_______________________________________________
lldb-commits mailing list
lldb-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits
