[PATCH] D93040: [InlineFunction] Use llvm.experimental.noalias.scope.decl for noalias arguments.
Jeroen Dobbelaere via Phabricator via cfe-commits
cfe-commits at lists.llvm.org
Sat Jan 23 14:25:20 PST 2021
jeroen.dobbelaere added a comment.
In D93040#2517769 <https://reviews.llvm.org/D93040#2517769>, @jdoerfert wrote:
> In D93040#2517758 <https://reviews.llvm.org/D93040#2517758>, @xbolva00 wrote:
>
>> http://llvm-compile-time-tracker.com/compare.php?from=344afa853fcfcc085cb5c957b4a07c7ea013bb1b&to=2b9a834c43cb1f93d33958c14b695896bb4e9c1e&stat=size-text
>>
>> Codesize regression 1% for tramp3d. Can you check it?
>
> I assume this to be a secondary effect of having the instructions in the first place. Maybe some unroll or inline size threshold needs to be thought about them. At the end of the day, we might not be able to avoid something like this as we make `!noalias` correct, though, I imagine the threshold theory which can be resolved.
The loop unrolling/rotating and the cleanup patches are not yet committed. You can see the effect of those here:
http://llvm-compile-time-tracker.com/compare.php?from=344afa853fcfcc085cb5c957b4a07c7ea013bb1b&to=eaf871f4e7fde26cd755cc4c2d67f2c244c66f18&stat=size-text
Also see D95141 <https://reviews.llvm.org/D95141> for links to more results.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D93040/new/
https://reviews.llvm.org/D93040
More information about the cfe-commits
mailing list