x86: Fix slight bug in `shared_per_thread` cache size calculation.
commit47f747217811db35854ea06741be3685e8bbd44d
authorNoah Goldstein <goldstein.w.n@gmail.com>
Tue, 18 Jul 2023 04:14:33 +0000 (17 23:14 -0500)
committerNoah Goldstein <goldstein.w.n@gmail.com>
Wed, 19 Jul 2023 01:56:25 +0000 (18 20:56 -0500)
treecc0be765c317ef17d6d73638ac221f4a24f00d03
parent2037f8ad016c05232bb170dbcb1a1d0ffbaeec20
x86: Fix slight bug in `shared_per_thread` cache size calculation.

After:
```
    commit af992e7abdc9049714da76cae1e5e18bc4838fb8
    Author: Noah Goldstein <goldstein.w.n@gmail.com>
    Date:   Wed Jun 7 13:18:01 2023 -0500

        x86: Increase `non_temporal_threshold` to roughly `sizeof_L3 / 4`
```

Split `shared` (cumulative cache size) from `shared_per_thread` (cache
size per socket), the `shared_per_thread` *can* be slightly off from
the previous calculation.

Previously we added `core` even if `threads_l2` was invalid, and only
used `threads_l2` to divide `core` if it was present. The changed
version only included `core` if `threads_l2` was valid.

This change restores the old behavior if `threads_l2` is invalid by
adding the entire value of `core`.
Reviewed-by: DJ Delorie <dj@redhat.com>
sysdeps/x86/dl-cacheinfo.h