After introduction of a itp correction based on number of running tests we have the odd situation that running more then 6 tests leads to a smaller total itp (and so total cores). Here the formula is changed that the total itp for more than 6 running tests is fixed to the same value as if tunning 6 tests.

[locutus2]

…ts we have the odd situation that running more then 6 tests leads to a smaller total itp (and so total cores). Here the formula is changed that the total itp for more than 6 running tests is fixed to the same value as if tunning 6 tests.

[locutus2]

It was always odd that a user which run more than 6 tests at once starts to loosing cores in the total. This fix keeps the total itp the same for 6 or more tests.

We had discussed this also here #1831. But i factored this out for an own PR.

Maintainers please check this. I have no local fishtest instance for really testing this but i have checked the changed code part in an online python interpreter and there it works as expected.

[locutus2]

I could use in this PR also my original formula from the other PR:

capped_count = min(count, 6)
itp *= 36.0 / (36.0 + capped_count * capped_count ) * capped_count  / count 

But i think this hiddes the fact that we can here two different functions combined (aka piece wise function) with the common point at 6 tests. So the if statement seems more explicit.

[vondele]

The original choice was actually on purpose... it is there as an incentive to those having many tests running, to prioritize their own tests by pausing overlapping ideas.

[locutus2]

@vondele
But that is mainly archived by the behavior of the formula up to 6 tests. if i would use 12 instead of 6 my formula means that each test uses only half the resources instead of even less as in current version. And i think that should be the decision of the dev. But i think both ways are ok. Most important its contrary to my aesthetic sensibilities because its seem unlogical and counterintuitive ;-). I know in Stockfish we have several such code but at least there we gained elo!

[locutus2]

As the point of view of the maintainers are now clear and no further discussion come up i close the PR.