Ideas to compute packed positions in a chess game

[devc1]

I see that every chess engine relies on move generation and evaluation speed which is almost the same between every engine. So any engine can only lookup a max of 4-6 depth without eliminating any moves.

I don't find any effective usage of SIMD in chess engines.

I guess so if you use AVX, how can we compute multiple positions. This may let us break the max of depth 6 to depth 8 or more (without eliminating moves).

Can we solve this ?
If we get a solution, it may be an evolution to the chess engines since it will search faster, eliminate less moves maybe and go deeper.

It is like using the "vpaddb" instruction to compute 32 moves at once !

[Octocontrabass]

I can't say I'm familiar with chess programming, but you haven't looked very hard if you haven't found anyone using SIMD for chess.

[devc1]

Everyone uses SIMD in chess but they only use them to compute scalar moves faster, make permutes. Not packed ones.
I want to find how to compute packed moves, which means calculating multiple moves in parallel.
I need some way to use bit manipulation in an SIMD register.

I have already looked into that page,

[Octocontrabass]

I need some way to use bit manipulation in an SIMD register.

It sounds like you're looking for SWAR.

[devc1]

Is this an instruction set or a library of functions ?

[Octocontrabass]

It's a set of techniques for operating on multiple elements within a single register, usually in ways the instruction set wasn't intentionally designed for.

For example, if AVX doesn't have an instruction for the operation you're looking for, you might look up some SWAR techniques to figure out how you can combine several instructions to get the operation you want.

[devc1]

Nice, it is still not called SIMD haha : )
Thanks anyway.

[Voldemort]

https://www.chessprogramming.org
As pointed by others usually have resources for what you are looking for. But I am lazy to dig deep enough to read engine source code see if they force use of SIMD instructions underneath.
I must point out that :
https://github.com/lantonov/asmFish is written in ASM, you may be change the logic to use the instructions you want by changing few functions here and there. But then I have not went far ahead reading its code.