bupticybee / TexasSolver

🚀 A very efficient Texas Holdem GTO solver :spades::hearts::clubs::diamonds:
https://bupticybee.github.io/texassolver_page
GNU Affero General Public License v3.0
1.65k stars 295 forks source link

CUDA #181

Open zikkuratti opened 9 months ago

zikkuratti commented 9 months ago

If i considering rewriting the calculation module to use CUDA cores, could you please advise from which file the calculation begins?

bupticybee commented 9 months ago

Can it really be done? I remember reading some peper says it's impossible.

Anyway you should check https://github.com/bupticybee/TexasSolver/blob/master/src/solver/PCfrSolver.cpp ,that's where calculation begin.

zikkuratti commented 9 months ago

if you provide me some extra architectural info i can try for a 6 month +- Now i go to research some CUDA algorythm

bupticybee commented 9 months ago

if you provide me some extra architectural info i can try for a 6 month +- Now i go to research some CUDA algorythm

Yeah I can provide anything you want. Please feel free to contact me anytime.

zikkuratti commented 9 months ago

Why on check not null moves he leave solve nodes empty sometimes?

bupticybee commented 9 months ago

Why on check not null moves he leave solve nodes empty sometimes?

What do you mean by this? Can you refer to some specific part of the code?

zikkuratti commented 9 months ago

Untitled

bupticybee commented 9 months ago

Untitled

Oh you need to double click this node to expand it. It's a QT bug.

DiaMyt commented 6 months ago

Hello @zikkuratti ! I noticed that you have started working on implementing the solver using GPU cores. Could you please share information about the current progress? Do you have an idea about the feasibility of successfully implementing this functionality?

zikkuratti commented 4 months ago

not yet but soon

bupticybee commented 4 months ago

not yet but soon

Glad to know this is still on going! Did you manage to get some speedup?

zikkuratti commented 4 months ago

now im fixing some bugs on another project after that i push on CUDA

zikkuratti commented 4 months ago

I need fast instruction how to make-build this code. No makefile here. I resarch some cuda stuff and need to recompile your code from scratch.

zikkuratti commented 4 months ago

Also i check console solver it boot #####100% 20-30 sec its too loong i wanna try too boost it, need to find it src folder comands list.

zikkuratti commented 4 months ago

And i think about can we speed up it if after console solving flop to json, we not rebuild turn and river, as new tree (like GUI) but add combinations of turn in json, cuz flop is 22100 turn ~270 000 river is 2,6 billions of extra

bupticybee commented 4 months ago

I need fast instruction how to make-build this code. No makefile here. I resarch some cuda stuff and need to recompile your code from scratch.

https://github.com/bupticybee/TexasSolver/tree/console

Please check this branch. Console branch has a makefile, and code structure is much simpler.

bupticybee commented 4 months ago

Also i check console solver it boot #####100% 20-30 sec its too loong i wanna try too boost it, need to find it src folder comands list.

There are ways too, could you contact me somehow? Do you have a telegram account? You can find me at https://t.me/bupticybee ,I'll provide whatever help you need.

bupticybee commented 2 months ago

Hi a small update, do you want to check on this? https://github.com/bupticybee/TexasSolver/pull/194

You could work with him on this feature? Or did you already do that? @zikkuratti