chethega

Use clang -Ofast xoshiro256plusplus.c -march=native -shared -fpic to compile.

Use gcc -Ofast xoshiro256plusplus.c -march=native -S or clang -Ofast xoshiro256plusplus.c -march=native -emit-llvm -S to see the generated code, and weep.

Output of julia ./xoshiro256plusplus.jl on my admittedly anemic machine:

julia dsfmt 1024 x UInt64
  2.209 μs (0 allocations: 0 bytes)
ref impl, 1 x interleaved, 1024 x UInt64
  1.579 μs (0 allocations: 0 bytes)

chethega

module AbstractAlgebra

export ZZ, FreeModule, Submodule, Generic, test_submodule

abstract type RingElem end
abstract type Ring end
abstract type FPModule{T} end
abstract type FPModuleElem{T} end

mutable struct Integers{T <: Integer} <: Ring

chethega

# This file is a part of Julia. License is MIT: https://julialang.org/license

function _truncate_at_width_or_chars(str, width, chars="", truncmark="…")
    truncwidth = textwidth(truncmark)
    (width <= 0 || width < truncwidth) && return ""

    wid = truncidx = lastidx = 0
    for (idx, c) in pairs(str)
        lastidx = idx
        wid += textwidth(c)

chethega

using Random


println("start...", Base.VERSION)

function gentape(Nsz, Ntape, rf)
    init = [rf() for i=1:Nsz]
    unique!(init)
    Nsz = length(init)
    state = copy(init)

chethega

#fallback aes via mbedtls.
#officially, mbedtls is a binary dependency of julia, so this should always work?
#also, results are (same as other versions) dependent on system endianness.
#factor 5 slower than native on my system, and causes no allocations. Probably fast enough.

module AESmbed

using Libdl
export Aes_wrap

chethega

if VERSION >= v"1.0.2"
    import Base._blsr
else
    @inline _blsr(x) = x & (x-1)
end

struct BitVec{N} <:AbstractVector{Bool}
    chunks::NTuple{N,UInt64}
end

chethega

if VERSION >= v"1.0.2"
    import Base._blsr
else
    @inline _blsr(x) = x & (x-1)
end

struct SBitSet{N} <:AbstractSet{Int64}
    chunks::NTuple{N,UInt64}
end

chethega

@enum hilb_state _down _up _left _right

function pairwise_recurse_multithreaded!(res,d,A,B, thresh, nthreads)
   # thresh = threshcompute(A);
   jobs = Vector{UnitRange{Int}}(nthreads)
   resize!(jobs, 0)
   si = round(Int,size(A,2)/nthreads)
   el = 0
   for i in 1 : nthreads-1
       push!(jobs, 1 + el : el+si)