[PATCH] D92722: [TableGen] [CodeGenTarget] Cache the target's instruction namespace

[Paul C. Anagnostopoulos via Phabricator via llvm-commits]

Paul-C-Anagnostopoulos added a comment.

I can't separate the performance increases of the two caching revisions, but I suspect the other one produces most of the improvement (https://reviews.llvm.org/D92674).

At least one backend requests the namespace 22 times, and the search is linear and must skip all the TargetOpcode instructions. There are about 200 of them, I think.

But if you think this is pointless, I'm happy to discard it. On the other hand, it's a total of 3 new lines of code.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D92722/new/

https://reviews.llvm.org/D92722