[PATCH] D116821: [DebugInfo][InstrRef] Move instr-ref controlling flag out of TargetOptions
David Blaikie via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Mon Jan 31 17:01:00 PST 2022
dblaikie added a comment.
In D116821#3285685 <https://reviews.llvm.org/D116821#3285685>, @probinson wrote:
> In D116821#3280137 <https://reviews.llvm.org/D116821#3280137>, @rnk wrote:
>
>> Regarding a value cutoff, I think that makes sense. To date, LLVM has never intentionally thrown away correct debug info, because historically we've had so preciously little of it. However, as we get better at tracking debug info, I think we're going to need more heuristics like this to cutoff excessive amounts of debug information.
>>
>> Anecdotally, debug info for mostly trivial inlined call frames (unique_ptr::get, operator*, etc) takes up a huge chunk of PDBs for Chrome. I think it might be time to start developing heuristics, modes, config files, or flags that discard trivial information like this to save file size and compile time.
>
> FTR, Sony downstream has implemented an option that will automatically treat certain methods and functions as `nodebug`. The criteria, IIRC, are: method with in-class definition; explicitly marked `inline`; or attribute `always_inline`. It gets rid of a lot of the fluff. If there's interest I can put it on my to-do list to post a patch.
Hmm - I'd at least be curious to know more about the criteria used. Those seem a bit more aggressive than I'd have figured - non-trivial inline functions in headers seem like they could be quite confusing to treat as nodebug.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D116821/new/
https://reviews.llvm.org/D116821
More information about the llvm-commits
mailing list