31 lines
1.5 KiB
Nim
31 lines
1.5 KiB
Nim
import std/[random, stats, times, math], sha256_64B{.all.}, cligen/strUt, cligen
|
|
|
|
proc benchK(a: Algo, x: seq[uint64], bN=256, mN=5, aN=10, cold=false) =
|
|
var o = newSeq[Digest](bN) # Output hash/digests
|
|
var s = 0u64
|
|
var ra: RunningStat
|
|
for aj in 0 ..< aN: # Average loop
|
|
var rm: RunningStat
|
|
for m in 0 ..< mN: # Minimum loop
|
|
let t0 = epochTime()
|
|
sha256_64Bs a, cast[pua Digest](o[0].addr),
|
|
cast[pua Block64](x[8*(mN*(m + aN*aj))].unsafeAddr), bN
|
|
rm.push epochTime() - t0 # Note time & below accum chksum
|
|
for b in 0 ..< bN: s += cast[ptr uint64](o[b][0].addr)[]
|
|
ra.push (if cold: rm.max else: rm.min) # Avg(Most(Times))
|
|
let scale = 1e9/bN.float # Want amortized nanosec per hash
|
|
let sigDt = ra.standardDeviation*scale/ra.n.float.sqrt
|
|
echo a, " ", fmtUncertain(ra.mean*scale, sigDt), " ", s
|
|
|
|
proc bench*(algos: seq[Algo], bN=256, mN=5, aN=10, cold=false) =
|
|
## Small bench CLI for Potuz' asm hashtree lib largely to get at auto (but
|
|
## likely static unless we go ATLAS route) method choice on various CPUs.
|
|
var x: seq[uint64] # Pre-generated rand data
|
|
for j in 1 .. 8*bN*mN*aN: x.add randState.next
|
|
for a in algos: benchK(a, x, bN, mN, aN)
|
|
|
|
when defined(release): randomize()
|
|
when defined(linux): import cligen/osUt; setAffinity() # lessen CPU migration
|
|
dispatch bench, help={"bN":"block count", "mN":"most loop", "aN":"avg loop",
|
|
"cold": "average max times not min"}
|