SF's root

[lech]

Why SF repeats the main evaluation (root) for each iteration?

[Ralph Stoesser]

Why SF repeats the main evaluation (root) for each iteration?

It makes no sense. id_loop() is the appropriate place for root evaluation.

[mcostalba]

It is there to let root_search() to follow the same template of normal search(), see also the surrounding comments.

I agree could be moved in idle_loop, but because the advantage is zero (about 20 saved evaluation out of many millions) perhaps has more sense to prefer clarity to teorethical speedup.

[lech]

Thanks for answers.
Another questions for root:
Why there is no threat detection in root (to avoid reduction if true).

[zamar]

Thanks for answers.
Another questions for root:
Why there is no threat detection in root (to avoid reduction if true).

Nobody ever put it there Shouldn't be a big deal, but still one of those little things which could be tested...

[lech]

Thanks for answers.
Another questions for root:
Why there is no threat detection in root (to avoid reduction if true).

Nobody ever put it there Shouldn't be a big deal, but still one of those little things which could be tested...

I thought about the mate threat detection in root only.
I was watching how works Houdini 1.5
In this position
[d] rk6/p1p5/Bp2p3/1P6/q2B1P2/4K3/8/8 b - - 0 1
after 21 iterations Houdini on 3rd and 4th places analyzing Qxd4 and Qxa6 (!) and it is a long analyze (like PV search).
It looks like Houdini use root to set (select) moves by increase (?).
Next example:
[d] 8/5pKN/5Pp1/5kP1/1B6/8/b7/7q b - - 0 1
Very quick (26 depth) the move Qxh7 is on 3rd place (!!!) Amazing !