Matrix of SVectors allocating

General Usage Performance
3

This is probably not what you’re looking for, but an obvious way to get around the problem on 1.10 is to write the matrix multiplication yourself:

function mymul!(C, A, B)
    for j = axes(C, 2), i = axes(C, 1)
        C[i, j] = sum(A[i, k] * B[k, j] for k=axes(A, 2))
    end
end

julia> @btime mymul!($B, $Dr, $A)
  26.432 ns (0 allocations: 0 bytes)

This is likely a bit slower than the 1.11 mul! version, but is much faster than the allocating version on 1.10

julia> @btime mul!($B, $Dr, $A);
  296.651 ns (3 allocations: 14.14 KiB)
2 Likes