RuntimeError: radix_sort: failed

[robomike]

When investigating this issue, the problem seems to be with thrust cupy/cupy#5846.

Can you please take a look at let us know why this problem with thurst occurs?

[robomike]

It also seems to be related to #936

[alliepiper]

We've been trying to find a repro for #936 for a while, but haven't been able to replicate / debug it to figure out what's going on. If you can find a thrust-only C++ minimal reproduction we can take a look.

I also suspect that this was fixed in CTK 11.4 (Thrust/CUB 1.12) by NVIDIA/cub@63e2ad4, which fixed a lot of overflows that may result in the InvalidValue error you're seeing.

[elstehle]

Thanks for reporting the issue, @robomike. I assume that the root cause of this issue is related to NVIDIA/cub#545, as there are multiple shared symptoms. I assume that this issue has been addressed by PR NVIDIA/cub#547. So, I will close the issue for now. Please feel free to reopen, should you still experience the issue, @robomike.