tr_sessionGetNextQueuedTorrent() can be faster

[jordan]

tr_sessionGetNextQueuedTorrent() walks through all the torrents to get their queue positions. This is done in a loop, once per free queue slot.

This can be expensive if tr_sessionCountQueueFreeSlots() or tr_session.torrentCount are high. It would be better if the get-next func returned N torrents to avoid redundant walking and position calculations.

[jordan]

Added in r13708, r13709

[jordan]

r13710: fix r13709 oops

[JJTagy]

May also want to change header file (attached)

[jordan]

Looks like this knocked ~7% off the runtime in my tests. My test session had a large number of torrents, so in the average case the improvement would probably be smaller.

Before the change:

     8.71%  transmission-da  transmission-daemon    [.] tr_torrentNext
            |
            --- tr_torrentNext
               |
               |--91.32%-- tr_sessionGetNextQueuedTorrent
               |          queuePulse
               |          bandwidthPulse
               |          event_base_loop
               |

And after:

     1.42%  transmission-da  transmission-daemon    [.] tr_torrentNext
            |
            --- tr_torrentNext
               |          
               |--88.30%-- tr_sessionGetNextQueuedTorrents
               |          queuePulse
               |          bandwidthPulse
               |          event_base_loop

[x190]

Replying to JJTagy:

May also want to change header file (attached)

Fixed in r13711.