Tobias Ribizel
Tobias Ribizel
I think this is in a good state for a review now. Sadly, since `rmm` is being installed by default inside `rapids-compose`, CPM picks up the installed `rmm` and thus...
yes, this depends on https://github.com/rapidsai/rmm/pull/1088 being merged, so I removed the incorrect label. I'll update it soon™️
@gpucibot merge
+1 on this update, having [`par_nosync`](https://github.com/NVIDIA/thrust/pull/1568) available is also pretty useful to my current work
@cwharris as this no longer applies, do you think the semantics would be useful elsewhere?
As I'm looking into this from the cub side, maybe also a few thoughts: * I think `left_fold` is too specific, since it implies a fixed evaluation order. Also places...
Just to give the reasoning behind my suggestion, I agree with most of the arguments here: `reduce_ordered` describes the algorithm behavior, while `reduce_noncommutative` describes the algorithm assumptions. In my use...
I am having the same kind of issue trying to compile [cuDF](https://github.com/rapidsai/cudf) with clang as device compiler. The specific error I'm observing occurs in type_traits: ``` In file included from...
In case you are using Reference, CUDA or HIP, you could try switching from `ParIlu` (which incurs a certain approximation error), to `Ilu` (which is exact). Can you give some...
That is true, our documentation needs some expansion, we will focus on that before the next release. It should be sufficient to just replace `ParIlu` by `Ilu` if you included...