@avive ?
my plan is to extend the compiler to compile the more optimal has me
hash code when applicable
regarding the overhead concern, the best way to avoid that is to modify the compiler directly because there are certain compiler directives that you can use when compiling the runtime
also who
there are functions in the commit that return 0 which i presume means unimplemented? panic("unimplemented") is fara
So the idea is to use cpuid() to obtain the runtime CPU capabilities (e.g. AVX2, SHA) and pick the best implementation of sha256 for the supported architecture. So, for example of the SHA intel extension is reported by the CPU then the code-base that uses these instructions will be used....
So I
The bounty is open until the 15th and you are welcome to join
The readme for the project I mentioned above explains what cpu caps are supported. With this bounty, we extended the lib to support the Intel SHA extensions which are only implemented on modern AMD zen-arch cpus....
so these are cpu caps based optimizations. Beyond that there are go-specific optimizations that @svenski123 did to improve the pref on any supported cpu arch...
i just know alot about Go's internal workings and how to implement this solution to get the best performance
as somebody mentioned earlier, this function is likely to be called in rapidly which generates way more garbage than one would initially expect
go assembler is better thought of as a textual representation of the plan 9 compilers IR bitcode
youll see what i mean when i submit my PR
Hi guys. For our kind of use cases, we are interested in sequential hashing of a small amount of data. Please look here: https://github.com/spacemeshos/sha256-simd/blob/8575aafc88c9020b9bb99e4b255fcc729593ffc4/sha256_test.go#L2272
small amount of data = 32 bytes.... (the output size of sha256)
The MHPs benchmarks we published here: https://github.com/spacemeshos/POET/blob/master/benchmarks.md
are computed from TestSequential() linked above....
Please send me a private message with request to access the AMD dev server and I will send you the access key and ssh instructions
Exciting to already see the nice performance improvements
Roger - we need it for the Spacemesh proofs of space time protocol which involved generation of proofs of space and proofs of space-time. Please see https://spacemesh.io for more info
@facundomedica @svenski123 - thank you for your great submissions to the POET contest! As the original contest date was today and you were the first 2 who submitted improved code - we tipped each of you with 5ETH award per the contest terms.
We are giving few more days for a 3rd contestant to have a chance to submit an improvement before announcing 1st place contest winner
If you are here then it means you were interested in proofs of elapsed time at some point and we announce this first here for you! We are happy to announce our POET summer of rust coding contest! See: https://spacemesh.io/summer-of-rust-contest/ - Open if you are a US developer who is interested to work on the Spacemesh protocol full time
