Damien L-G
Damien L-G
Ignoring HIP failures. The issues with SYCL with user-defined functors when deferring an algorithm implementation to oneDPL is already present in `develop` with `Kokkos::sort` and a custom comparator. We decided...
> This will be in 4.3 correct? Would be really nice if it supported `sort_by_key` too... This is the plan yes. Bruno is swamped with something else but will get...
This is an issue with your toolchain. Your Intel compiler uses libstdc++ (the C++ standard library from GCC) version 13.2 which does not seem to be compatible with the CUDA...
> @masterleinad I understand your point of view. It's true that there is no performance gain. > > However, in light of the (IMO very good) SO answer https://stackoverflow.com/a/54545744, I...
FWIW, we run 14.0.0 in the CI and it is passing. I have 15.0.0 on my laptop and it is fine as well. Can anyone reproduce?
Uggg. No I did not expect any of these would have anything to do with that issue...
Using this to get the min/max value in an array is suboptimal. Data transfer twice. The right tool with the standard library parallel algorithms is to use `reduce` (which admittedly...
```C++ auto max_val = std::reduce( first, last, std::numeric_limits::least(), [](const auto& lhs, const auto& rhs){ return std::max(lhs, rhs); });
I will close as we merged Andrey's PR adding the `Experimental::sort_by_key` algorithm. It will be in the upcoming 4.3 release. Please open a new issue if you have feedback, for...
Did you look at the assembly? Adding a reference to #553