Domagoj Šarić

Shouldn't this be high priority given the prevalence of devices w/o an L3 cache in the mobile and embedded world? ;) Likewise - would a better 'fix' (then pretending l3...

cool thanks will try it...once works and i can compile ;D

tried it with the current code with a simple example: ``` kwk::shape shape_d; kwk::shape shape_s; auto z = kumi::map( []( auto a, auto b ) { return a + b;...

> Based on the upstream bug, this is not a regression from prior releases, correct? Just a (painful) behavior mismatch between LTO and non-LTO builds? I would suppose so -...

waiting for this :) but do consider (ct) efficiency, i.e. overhead of slicers for just, in effect, reduced fold range (than the one in numel() ;P

> The idea is sound as we discussed during the conference, how tobkeep those info as IIRC typzdef and attributes dont mix well. You can use 'decorating pass-through' functions instead...

> I was thinking maybe use the decorators directly when passing the pointer to the view and let the deduction guide do its job. We're moving toward a more DG,...

Cool...something to show off with to the ML python engineers 😁 Just a 'tiny' leftover - restricted (pardon the pun) automatic conversion of view types (only) in the less constrained...

Since it is inlined it should not be needed but hey it's MSVC so ... 🤦 😅 ... best to simply slap [[ gnu::returns_nonnull, gnu::malloc, msvc::restrict ]] on it :)

heads up 'bout the other ugly corporation - apple's stl still does not have std::assume_aligned yet msvc does oddly support __builtin_assume_aligned so go with that for the time being...