Hello guys, I've been working lately on the @WanderingPhilosopher KeyHuntCudaClient version, stripped everything to keep only the single address search mode for the sake of searching puzzle 66 only.

Now the problem, as usual, no mater what ideas I try, there's no possibility to adapt it to what I want to increase the speed.

Without this set of controlling the public keys too to reduce the search complexity is not possible with the current resources, even if we know the priv key ranges.

For e.g. now I tried to modify the getGPUStartingKeys function, on the part where the keys are computed, to compute only the pub keys with the expected prefix, let's say 02 and save them to the p array instead of computing everything and store there.
Normaly this a good thing because you reduce almost 50% of keys, but as nbThread > filtredKeys, the rest of the threads are filled with zeros and I end up processing that garbage too when setKeys is called, so not a solution, even if I force the nbThread = filtredKeys.

Maybe the _GetHash160Comp I said, to filter there the keys just before the SHA256Initialize(s); an if condition to check if (publicKeyBytes[0] & 0xFF) != 0x02, to return early, otherwise to process.

The reason of considering this is as you know hash160 transformation is an expensive process, like the _ModInvGrouped from ComputeKeysSEARCH_MODE_SA and it that way I wanted to save processing.

Oh and the final goal was to have a new cmd argument defined as --pubKeyStartsWith to control the key, only the expected prefix and maybe the second byte, let' say --pubKeyStartsWith "02b7", I think that search will be more refined.

Any ideas?