each expression type in
xtensor provide the stepper_begin / stepper_end methods so you don't have to instantiate the steppers by yourself
@davidbrochart So SIMD doesn't bring that much performance
It depends a lot on what you are trying to accelerate
Compilers are usually good enough to vectorize code written with loops
so trying to plug XSIMD on something like
for(size_t i = 0; i < s; ++i)
{
res[i] = a[i] + b[i] * c[i] - d[i];
}
where a, b, c and res are tensors, is totally useless
however if a, b, c or d is a more complicated expression (like we have in
xtensor), the cmpiler may not be able to autovectorize it and xsimd is required
@lgonzalezsa do the kernels need to be in different environments? Otherwise (if your setup allows it) all the kernels could be in the same environment, and then yes use the notebook metadata to differentiate between them (voila uses the notebook metadata to know which kernel to start)
Name
xaxis_iterator
Only to iterate on rows. Not other axes.
I had totally forgotten it ;)
@rmporsch so actually you can include
xaxis_iterator.hpp and use axis_begin/axis_end as suggested by @SylvainCorlay
I'm thinking of something like using nginx as a reverse proxy
i want this to persist across reboots
I think it would be great to consolidate the docs so we can provide a guide to achieve something like this
if you would like to open a PR, that would be very welcome!
Where would you place that in the docs?
there is a "Deploying voila" section here: https://voila.readthedocs.io/en/latest/deploy.html
and the source for it is here: https://github.com/QuantStack/voila/blob/7d2fd26e7196ab11a76fe1d73fab431461889561/docs/source/deploy.rst
and the source for it is here: https://github.com/QuantStack/voila/blob/7d2fd26e7196ab11a76fe1d73fab431461889561/docs/source/deploy.rst
for now it's all in one place since there are instructions only for a couple of providers
we might want to slightly reorganize it when the list grows
actually, it did
but it was ripped out at some moment
it sped up dev time, it auto reloaded when any python source, js file, template or notebook changed
but the watches on the filesystem are expensive
we should enable it again and put it under a --dev flag orso
there doesn't seem to be any decent computational geometry package in Python
what's the process of making
scikit-cg?
Just for some primitives alone it would be worth it --
Point2, Line2, Ray2, Segment2 & Vector2 you seldomly find all of them it seems in one package
and it's also very hard to find some cool algorithms from CGAL somewhere else, for example Minkowski sum or Constrained Delaunay Triangulation