Help porting C to CUDA and then move branching to CPU.

[cyberspacemonkey]

I'd like to learn how to Mod ccminer to improve performance by moving the branching part to the CPU and run only the hashing algos on GPU, can someone guide me step by step? or can at least tell me where can I find a guide for that? I understand it's difficult but I'd appreciate any help.

[sp_]

Ccminer is already doing all the hashing code on the gpu, but when it finds a solution it verifies the result on the CPU.

[djm34]

I'd like to learn how to Mod ccminer to improve performance by moving the branching part to the CPU and run only the hashing algos on GPU, can someone guide me step by step? or can at least tell me where can I find a guide for that? I understand it's difficult but I'd appreciate any help.

what branching part ?
as sp_ wrote, everything is already done on gpu except the (unnecessary) cpu validation ( where the result to gpu algo is compared to the "original" cpu algo)

By branching if you mean to merge together all the kernel into one big entity then it will kill the performance (not enough registers)

[cyberspacemonkey]

I'd like to learn how to Mod ccminer to improve performance by moving the branching part to the CPU and run only the hashing algos on GPU, can someone guide me step by step? or can at least tell me where can I find a guide for that? I understand it's difficult but I'd appreciate any help.

what branching part ?
as sp_ wrote, everything is already done on gpu except the (unnecessary) cpu validation ( where the result to gpu algo is compared to the "original" cpu algo)

By branching if you mean to merge together all the kernel into one big entity then it will kill the performance (not enough registers)

How would I go about modifying my ccminer to improve performance then?

[djm34]

I'd like to learn how to Mod ccminer to improve performance by moving the branching part to the CPU and run only the hashing algos on GPU, can someone guide me step by step? or can at least tell me where can I find a guide for that? I understand it's difficult but I'd appreciate any help.

what branching part ?
as sp_ wrote, everything is already done on gpu except the (unnecessary) cpu validation ( where the result to gpu algo is compared to the "original" cpu algo)

By branching if you mean to merge together all the kernel into one big entity then it will kill the performance (not enough registers)

How would I go about modifying my ccminer to improve performance then?

rewrite the algos

[cyberspacemonkey]

I'd like to learn how to Mod ccminer to improve performance by moving the branching part to the CPU and run only the hashing algos on GPU, can someone guide me step by step? or can at least tell me where can I find a guide for that? I understand it's difficult but I'd appreciate any help.

what branching part ?
as sp_ wrote, everything is already done on gpu except the (unnecessary) cpu validation ( where the result to gpu algo is compared to the "original" cpu algo)

By branching if you mean to merge together all the kernel into one big entity then it will kill the performance (not enough registers)

How would I go about modifying my ccminer to improve performance then?

rewrite the algos

Sounds difficult, but is it possible? if so how would I start? thanks for replying by the way.