No allocations during optimization

[benruijl]

I'd like to run totsu on problems in a tight inner loop of my program. As such, performing allocations is too costly. It would be nice if there is a way to provide a pre-allocated output vector. For example:

fn solve_socp<L>(
    &mut self,
    param: &PDIPMParam,
    log: &mut L,
    vec_f: &Mat,
    mat_g: &[Mat],
    vec_h: &[Mat],
    vec_c: &[Mat],
    scl_d: &[FP],
    mat_a: &Mat,
    vec_b: &Mat
    x: &mut Mat,
) -> Result<Info, String>

where x is used to store the variable values and Info is some struct that exists purely on the heap and provides info about the precision of the solution, etc.

Does totsu allocate vectors during the solving?

[convexbrain]

Unfortunately I think allocations occur while solving (e.g. during a matrix-matrix multiplication), although it is possible to make a x: &mut Mat version of solve_socp.

[convexbrain]

At version https://github.com/convexbrain/Totsu/releases/tag/rust_conic_v0.6.0, the core solver doesn't rely on dynamic heap allocation.