[PATCH v2] linux-gen: sched: use restrict keyword to help optimizer #2123
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
odpbuild
changed the title
MatiasElo
reviewed
Sep 23, 2024
| @@ -1196,7 +1196,7 @@ static inline int balance_spread(int grp, int prio, int cur_spr) | |||
| return new_spr; | |||
| } | |||
|
|
|||
| static inline int copy_from_stash(odp_event_t out_ev[], uint32_t max) | |||
| static inline int copy_from_stash(odp_event_t *restrict out_ev, uint32_t max) | |||
There was a problem hiding this comment.
Could also use static inline int copy_from_stash(odp_event_t out_ev[restrict], uint32_t max) to maintain the same style but either method is OK for me.
There was a problem hiding this comment.
Yes, but I felt that it is a bit weird syntax and harder to understand. But maybe it is just me. Did not change it in v2 though.
MatiasElo
approved these changes
Sep 23, 2024
Now that -fno-strict-aliasing was added in the default compiler flags, the compiler cannot use type based alias analysis when optimizing and has to assume potential aliasing if it cannot prove otherwise. This slowed down scheduling because the compiler could not deduce that the out_ev parameter of copy_from_stash() does not alias sched_local. Most of the slowdown seems to come from the use of the ev_index and num_ev fields of sched_local.stash in the loop body. Using local variables instead would solve the issue for the most part but it appears that qualifying the out_ev pointer with restrict has a similar effect with gcc and clang even though the problematic accesses have a different type from *out_ev. Fix the problem by using the restrict keyword to indicate that *out_ev does not alias with anything else within the function. Signed-off-by: Janne Peltonen <janne.peltonen@nokia.com> Reviewed-by: Matias Elo <matias.elo@nokia.com>
JannePeltonen
force-pushed
the
sched-perf-res
branch
from
13a356e
to
07930f6
Compare
odpbuild
changed the title
|
v2: review-tag |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Now that -fno-strict-aliasing was added in the default compiler flags, the compiler cannot use type based alias analysis when optimizing and has to assume potential aliasing if it cannot prove otherwise. This slowed down scheduling because the compiler could not deduce that the out_ev parameter of copy_from_stash() does not alias sched_local.
Most of the slowdown seems to come from the use of the ev_index and num_ev fields of sched_local.stash in the loop body. Using local variables instead would solve the issue for the most part but it appears that qualifying the out_ev pointer with restrict has a similar effect with gcc and clang even though the problematic accesses have a different type from *out_ev.
Fix the problem by using the restrict keyword to indicate that *out_ev does not alias with anything else within the function.