Closed MBkkt closed 1 year ago
Reloading it should not be necessary.
@axboe
But behavior with single load and load on every iteration could be different? Because kernel can produce new cqe while it is iterating. Or I don't understand it right?
That is obviously true, but I would suspect most loops to process CQEs are pretty light (and hence fast) and therefore the likelihood of getting a new entry while looping is pretty small. So while it doesn't hurt to reload the cq tail on every loop, it'd probably be better to avoid it.
Thanks.
This commit fix this inefficiency https://github.com/axboe/liburing/commit/c16017c714b4400feba079c7681baf5186c9c8dc
Could you explain why do you do load on every iteration?
https://github.com/axboe/liburing/blob/master/src/include/liburing.h#L295