Explore scrypt8k

[JayDDee]

Scrypt8k uses parameters: N=8192, R=8, p=1.
The current scrypt code can handle different N values but R and P are hard coded at 1.

The following is an unoptimized version of scrypt with a variable R parameter.

https://git.xmr.pm/wownero/node-multi-hashing/src/branch/master/scryptn.c

Explore whether the optimized Pooler code in cpuminer-opt can be modified to support
different R values.

[JayDDee]

As expected the Pooler code has lost a lot of structure due to its extreme optimization.
Whether there is enough structure left to implement variable R parameter is still unknown.
Even if possible it may require some restructuring that would likely affect performance.
Hopefully it only affects buffer sizes and loop counters that exist in both implementations.
More analysis is required.

I've identified the following locations that use the R parametrer.

• the size of the scratchpad
• PBKDF2 SHA256 block length
• scrypt core/smix

[JayDDee]

Things are not looking good for using the existing Pooler scrypt code with a variable R.
It is optimized for a fixed R value. Suppporting multiple R values woudl require seperate
copies of the code, each one optimized for one R value. That would not be acceptible.

Modifying the Pooler code to support variable R would likely have a performance impact and
would also be unacceptible, even if possible.

At this time the only option appears to be to import the Wonero scrypt code to support
scrypt variations where R != 1 at significantly reduced performance.

Although importing code without optimizing is not consistent with the primary goal of cpuminer-opt
it does offer some forward compatibility with future variations of scrypt using different R values
and is worth considering.

[JayDDee]

More problems.

I can't get the Wonero code to work with scrypt (1024, 1, 1). I replaced the Pooler hash function
with Wonero and all I got was invalid shares. Since the 2 functions are so dissimilar there are
no common test points to collect debug data for comparison.

At this point I'm stuck.

[JayDDee]

I'm thinking of abandoning this enhancement due to lack of interest. I can't find a single pool
supporting this algo. Also without a working reference the chances of success are very low.

Unless these problems are resolved shortly this isue will be closed.

[JayDDee]

Closed.