nanocurrency / nano-pow Goto Github PK

Use Google's HighwayHash or AVX2 implementation of Siphash.

Our SipHash implementation is a fast and portable drop-in replacement for the reference C code. Outputs are identical for the given test cases (messages between 0 and 63 bytes).
Interestingly, it is about twice as fast as a SIMD implementation using SSE4.1 (https://goo.gl/80GBSD). This is presumably due to the lack of SIMD bit rotate instructions prior to AVX-512.

Would it be possible to publish a dart implementation of compute pow ? It could use ffi to use this

Due to the stepping-loop used this should be relatively simple, however cpp_driver::fill_impl () does not use steps so we must ensure there are no performance losses when using them.

Argument --verbose prints out the fill/search time separately.

From my tests:

4 bytes:

• lookup = floor(difficulty/2) + 1
• GPU threads = 8192 (higher for Vega)

5 bytes:

• lookup = floor(difficulty/2)
• GPU threads = 6144 (unknown for Vega)