The Computer Language
24.11 Benchmarks Game

k-nucleotide Julia #6 program

source code

# The Computer Language Benchmarks Game
# https://salsa.debian.org/benchmarksgame-team/benchmarksgame/
#
# Contributed by Adam Beckmeyer

using Printf

const COUNTSFOR = ("ggt", "ggta", "ggtatt", "ggtattttaatt", "ggtattttaatttatagt")

# This is not reading the file in line-by-line. Doing so is impossible
# in Julia without using FFI because all io is automatically
# buffered. Thus despite not appearing to be buffered reads, all are.
function getseq3(io)
    # First read all the file preceding the 3rd sequence.
    for _=1:3
        readuntil(io, '>')
    end
    readline(io)
    # Then read the third sequence in.
    buf = read(io)

    # In place, remove all newlines from buf and encode nucleotides
    # using only the last 2 bits in each byte.
    i = 1
    for c in buf
        if c != 0x0a # '\n'
            # Gives a -> 0x00, c -> 0x01, g -> 0x03, t -> 0x02
            @inbounds buf[i] = c >> 1 & 0x03
            i += 1
        end
    end
    resize!(buf, i - 1)
end

# Decoding a single encoded nucleotide results in a byte 'a', 'c', 'g' or 't'
decode(c) = c == 0x00 ? 0x61 : c == 0x01 ? 0x63 : c == 0x03 ? 0x67 : 0x74

# Decoding a UInt32 or UInt64 results in a string. This function
# creates the string from the last n encoded nucleotides in c.
function decode(c, n)
    buf = Vector{UInt8}(undef, n)
    for i=1:n
        @inbounds buf[n+1-i] = decode(c % UInt8 & 0x03)
        c = c >> 2
    end
    String(buf)
end

# Count the number of each subsequence of length n present in seq and
# store the result in hashtable d.
function count_subseqs!(d::Dict{T}, seq, n) where T
    # This results in a number where the last 2n bits are 1 and the rest are 0.
    mask = (1 << 2n - 1) % T
    key = zero(T)

    # Fill key with the first n-1 encoded nucleotides
    for i=1:n-1
        key = key << 2 | @inbounds seq[i]
    end

    # Slide key over the remainder of the sequence. Because of mask,
    # only the last n nucleotides are encoded in key.
    for i=n:length(seq)
        key = mask & (key << 2 | @inbounds seq[i])
        inc!(d, key)
    end
end

# If k exists in dictionary, increment it by one. Otherwise set the
# value of k to 1. If julia ever gets an update!/modify! function for
# changing hashtable values with one lookup, this function can be
# removed.
@inline function inc!(d::Dict{T,Int32}, k::T) where T
    index = Base.ht_keyindex2!(d, k)

    if index > 0
        # Positive index means the key already exists in the dictionary
        @inbounds d.vals[index] += 1
    else
        # Negative index means the key doesn't yet exist
        @inbounds Base._setindex!(d, 1, k, -index)
    end
end

# Define a fast hashing function for UInt keys encoded with the
# sequence. Defining a Base function on a Base type is "type-piracy"
# and should be avoided in production code. Instead, Base.hash could
# be defined on a wrapper struct. For this workload, this simple
# xor/bitshift hash speeds things up by over 20% compared to Julia's
# default hash.
Base.hash(x::Unsigned)::UInt = x ⊻ x >> 7

# Create a sorted array with the number of appearances of each
# sequence of length n.
function frequency_table(seq, n)
    d = Dict{UInt32,Int32}()
    count_subseqs!(d, seq, n)

    total = sum(values(d))
    counts = [decode(k, n) => get(d, k, 0) / total for k in keys(d)]
    sort!(counts; lt=freq_isless, rev=true)
end

# This function is used for sorting the frequency table generated by
# the function above. Primary order is by frequency which is second
# in pair. Secondary order is reverse order of string which is first
# in pair.
freq_isless(a, b) = a[2] == b[2] ? isless(b[1], a[1]) : isless(a[2], b[2])

# Count the number of times subseq appears in seq by creating a
# hashtable of all sequences of the same length as subseq appearing in
# seq.
function count_sequence(seq, subseq)
    d = length(subseq) < 16 ? Dict{UInt32,Int32}() : Dict{UInt,Int32}()
    count_subseqs!(d, seq, length(subseq))

    # Construct the key needed for accessing the count of subseq
    key = 0 % UInt
    for c in subseq
        key = key << 2 | (c >> 1 & 0x03)
    end

    get(d, key, 0)
end

function main(io, out)
    seq = getseq3(io)

    # First spawn threads for counting occurrences of subseqs in seq
    # since they take longer than the frequency tables.
    counts = Vector{Int32}(undef, 5)
    counts_task = Threads.@spawn @sync for i=1:5
        Threads.@spawn @inbounds counts[i] =
            count_sequence(seq, codeunits(COUNTSFOR[i]))
    end

    # Because of Julia's threading overhead, it's faster to just
    # calculate both the 1 and 2 nucleotide frequency tables on the
    # main thread while counts_task continues in background.
    for (a, b) in frequency_table(seq, 1)
        @printf(out, "%s %2.3f\n", uppercase(a), 100b)
    end
    println(out)

    for (a, b) in frequency_table(seq, 2)
        @printf(out, "%s %2.3f\n", uppercase(a), 100b)
    end
    println(out)

    wait(counts_task)
    @inbounds for i=1:5
        println(out, counts[i], '\t', uppercase(COUNTSFOR[i]))
    end
end

isinteractive() || main(stdin, stdout)
    

notes, command-line, and program output

NOTES:
64-bit Ubuntu quad core
julia version 1.11.1


 Sun, 10 Nov 2024 19:47:46 GMT

MAKE:
printenv JULIA_NUM_THREADS
4

0.19s to complete and log all make actions

COMMAND LINE:
 /opt/src/julia-1.11.1/bin/julia -O3 --cpu-target=ivybridge --math-mode=ieee  -- knucleotide.julia-6.julia 0 < knucleotide-input25000000.txt

PROGRAM OUTPUT:
A 30.295
T 30.151
C 19.800
G 19.754

AA 9.177
TA 9.132
AT 9.131
TT 9.091
CA 6.002
AC 6.001
AG 5.987
GA 5.984
CT 5.971
TC 5.971
GT 5.957
TG 5.956
CC 3.917
GC 3.911
CG 3.909
GG 3.902

1471758	GGT
446535	GGTA
47336	GGTATT
893	GGTATTTTAATT
893	GGTATTTTAATTTATAGT