Finch

Finch is an adaptable compiler for loop nests over sparse or otherwise structured arrays. Finch supports general sparsity as well as many specialized sparsity patterns, like clustered nonzeros, diagonals, or triangles. In addition to zero, Finch supports optimizations over arbitrary fill values and operators, even run-length-compression.

At it's heart, Finch is powered by a domain specific language for coiteration, breaking structured iterators into units we call Looplets. The Looplets are lowered progressively, leaving several opportunities to rewrite and simplify intermediate expressions.

Installation:

julia> using Pkg; Pkg.add("Finch")

Basics:

To begin, the following program sums the rows of a sparse matrix:

using Finch
A = sprand(5, 5, 0.5)
y = zeros(5)
@finch begin
    y .= 0
    for i=_, j=_
        y[i] += A[i, j]
    end
end

The @finch macro takes a block of code, and compiles it using the sparsity attributes of the arguments. In this case, A is a sparse matrix, so the compiler generates a sparse loop nest. The compiler takes care of applying rules like x * 0 => 0 during compilation to make the code more efficient.

You can call @finch on any loop program, but it will only generate sparse code if the arguments are sparse. For example, the following program calculates the sum of the elements of a dense matrix:

using Finch
A = rand(5, 5)
s = Scalar(0.0)
@finch begin
    s .= 0
    for i=_, j=_
        s[] += A[i, j]
    end
end

You can call @finch_code to see the generated code (since A is dense, the code is dense):

julia> @finch_code for i=_, j=_ ; s[] += A[i, j] end
quote
    s = (ex.bodies[1]).body.body.lhs.tns.bind
    s_val = s.val
    A = (ex.bodies[1]).body.body.rhs.tns.bind
    sugar_1 = size((ex.bodies[1]).body.body.rhs.tns.bind)
    A_mode1_stop = sugar_1[1]
    A_mode2_stop = sugar_1[2]
    @warn "Performance Warning: non-concordant traversal of A[i, j] (hint: most arrays prefer column major or first index fast, run in fast mode to ignore this warning)"
    for i_3 = 1:A_mode1_stop
        for j_3 = 1:A_mode2_stop
            val = A[i_3, j_3]
            s_val = val + s_val
        end
    end
    result = ()
    s.val = s_val
    result
end

We're working on adding more documentation, for now take a look at the examples!