nilsbecker

> You might want to try moving `uu0` into `fip` :/ accessing global data like that is finecky. hmm, i don't think i get it. `fip` is the rhs of...

i tried `uu0 = CuArray(u0)` which resulted in an error involving `jl_apply_generic`. i then also tried replacing `C` with `CuArray(B)` which resulted in `InvalidIRError`. my latest full code is: ```...

you're now building `C` within `fip{2,3}`. does this mean there will be a static matrix generated fresh on every call (on the gpu)?

ah ok. i'm still getting to grips with a mental model for JIT compiled code evaluation. i would have thought that what you described happens only if `C` is statically...

i think i didn't make my point very clearly, sorry about that. `Random.default` is not exposed; you can find it in random.ml only. it is the state that gets used...

@bluddy ah. i thought maybe `include` lets you include the raw module without filtering by signature but i guess that would break abstraction. also apparently it's not possible to blit...

i guess. however, `Random.set_state state` would then not set that one. ah, maybe one could override it: in CCRandom: ```ocaml let default_state = ref (Random.get_state ()) let set_state state =...

i would think that not having > * calling `run g` several times should look random rather than always returning the same thing would be so surprising to qualify as...

so a solution could be to be explicit that `run` without `~st` updates its own private state, and to supply setter and getter functions `set_run_state` and `get_run_state` specifically to handle...

but maybe that's just the way it is, with the optional arguments? maybe it should not be registered for the top level? or with some defaults for the start, stop,...