Fixed shift magics with 800KB lookup table

[Teemu Pudas]

Because the attack bitboards are incompatible: if a1=0, one is b1|a2|a3|a4 and the other is (rank1^b1)|b2.

[Gerd Isenberg]

Because the attack bitboards are incompatible: if a1=0, one is b1|a2|a3|a4 and the other is (rank1^b1)|b2.

I see, you already added the sqIndecies, which I initially missed and only focused on the magic products/shifts.
Anyway, if I calculate your values I get different products as well as indices. What did I miss?

Code: Select all

0x000001000100001A * 0x00A801F7FBFEFFFF => 0x080C312F96E5FFE6
0x0000020000000200 * 0xFFB000810113FFFF => 0x8801000227FFFE00
0x080 + 5630 => 0x167E
0x880 + 3582 => 0x247C


[Teemu Pudas]

0x080 + 5630 => 0x167E
0x880 + 3582 => 0x247C

0x247C - 3582 == 0x167E, not 0x880.

[Gerd Isenberg]

0x080 + 5630 => 0x167E
0x880 + 3582 => 0x247C

0x247C - 3582 == 0x167E, not 0x880.

oups, you are right, i mixed up some values. Too early in the morning.

Code: Select all

0x080 + 5630 => 0x167E
0x880 + 3582 => 0x167E


[Gerd Isenberg]

Because the attack bitboards are incompatible: if a1=0, one is b1|a2|a3|a4 and the other is (rank1^b1)|b2.

Solved!
Grant still uses variable shifts but no fixed as the title suggests

http://www.talkchess.com/forum/viewtopi ... 05&t=39123

Gerd

[rkalyankumar]

How to generate the magic factor & the (overlapping) index?

Thanks,
Kalyan