Terry Cojean

Thanks for pinging me @upsj. When I saw the title I wanted to comment already :). In general, I think this is a good idea or at least it might...

Just for completeness, `jacobi_kernels.cu` and `zero_array.cu` also use the namespace `kernel`.

The default `CudaExecutor` was here before, because automatical in the previous releases was purely a Cuda strategy. Changing this would definitely break interface. We can consider this a bugfix maybe...

#969 implemented a non-interface breaking way of achieving the same result. But `Csr::automatical` and strategies in general still need to be fixed.

@hartwiganzt is this not done after the recent lecture? In which case, maybe we could put the slides somewhere and/or convert it to a wiki page?

Here is some more feedback on tools. I feel that we should be able to understand things reasonably without tools but it's true they help life a lot. What I...

I might add if there is any emacs user, I found [rtags](https://github.com/Andersbakken/rtags/tree/v2.18) which does works quite well: 1) it finds macro generated names properly and 2) deals decently with templates....

@flipflapflop this issue discusses the code generation macros. You will find some explanations of this and tentative solutions which looked more annoying than what we have. In general, we are...

So as a summary, what is left after the fix of #166 is that we still have a copy of the full residual vector even when the method has no...

I uploaded [here](https://drive.google.com/open?id=14dPbjh1izCHMW_GxlYwjSwpiHty0MkhK) a trace taken on a V100 on Juwels to figure out the copy time for BiCGSTAB on thermal2. We extracted the timing by hand with @hartwiganzt for...