I was a bit confused how long the compilation took and how much RAM was required, when playing aorund with the develop branch when using the nativ Sha256 libraries. Having done quite a bit of hashing with ZoKrates, it seemed off. Initially I thought this would be a problem with the develop branch, however the same behaviour is found on master.
In zkRelay_val.zok we do 18 rounds of sha256 hashing, validate_block_header() is called 3x: for the 1024bit hash 4 rounds, for the 256bit hash 2 rounds => Total 18 rounds.
However, when compiling 512bitx10.zok (which does 20 sha256 rounds) compilation takes almost 9x as long, and ram usage is up x6.5. While zkRelay_val.zok does contain some other logic, this is very surprising and doesn't really make sense to me. The file is also a bit messy, but I think it will suffice for an example
Interesting is also, that the nativ hashes performs better when a single hash is computed, when its more its a lot worse constraint wise.
| file | compile_opt_microsec | memusg_compile_KiB | constraints |
|---|---|---|---|
| 512Bitx10Embed | 2534824 | 444032 | 277802 |
| 512bitx10 | 410881180 | 26738596 | 363184 |
| 512bitx10_no_loop | 413493618 | 26738504 | 363184 |
| 512Bit | 1795275 | 134032 | 26779 |
| 512BitEmbed | 379155 | 59412 | 28241 |
| getHexLength | 882580 | 39560 | 22612 |
| zkRelay_val | 48289868 | 4066328 | 292357 |
These results come from a server with 60GB ram, locally this will run a lot slower.