Oswin Krause

all program generation is cached and the cache size is limited (i think 32 or 64 programs). load_program is also called only in the constructor.

In my current application I am going to use it for type deduction (e.g. it could mess up when I want to count the number of elements in a vector...

don't worry i can likely only pick it up with the next boost releas eincluding it. So no stress!

Hi, Is there any news on this? this issue becomes more severe for me as some unit tests currently generate > 10000 kernels instead of maybe 10 or so. I...

Oh i forgot my idea on how to solve the unique name problem. The best way to make the argument names unique is to tell the kernel which variables belong...

Any news on this? This is a _real_ showstopper for me. The biggest issue I have is that all kinds of indexing into a buffer leads to recompilation of the...

If we have a working proposal, i can invest the time to add and thest the changes in the meta_kernel as well as some of the iterators. I am stuck...

I worked myself a bit on a solution, see lines 581-607 https://github.com/Shark-ML/Remora/blob/1b6aba317eaa381215e9c273247cbf73c8ce9f6a/include/remora/gpu/traits.hpp#L581 This is a solution were functions with arguments are transformed into functions with a registered variable name. Line...

Hi, the issue with meta_kernel is that it is in namespace detail, which by boost convention is recognized as "implementation detail, subject to change, do not touch". As i got...