[clang] [Serialization] Don't read all declaration id eagerly when merge the tables (PR #95506)
Chuanqi Xu via cfe-commits
cfe-commits at lists.llvm.org
Thu Jun 20 18:24:31 PDT 2024
ChuanqiXu9 wrote:
> > Yeah, I think this patch may be conceptually good except the extra memory using. But benchmarking, I tried it locally but didn't find observable effects. So I'd like to land this after 19's branching to give it more baking time.
>
> Do you mean that no large increases in the memory use or that things are not getting faster either? If it's the latter, I would vouch for keeping the code as-is to avoid adding extra complexity. If this does lead to significant performance wins, though, it's a different story.
Yeah, it is the later. But it might not be too complex since all its uses are limited to the file only. My point is on the contrast, if it doesn't lead to significant regression, we should do it. Since it is rare that a single patch did herotical optimizations. Some times the performance gains from the long tails.
https://github.com/llvm/llvm-project/pull/95506
More information about the cfe-commits
mailing list