Matthew Cordery

1. We were hitting issues because the intel compilers don't have any 'implicit declare target' capability and the fence for the pragma is only for ibm so we basically had...

So I have an updated version of reduce.hpp that modifies the Reduce class reset function. It did solve the ReduceSum problem that I'm seeing and it *should* have solve the...

[AMD Official Use Only - AMD Internal Distribution Only] It would just reduce the time to hipify everything if you didn’t need the repo’s .cu files at all. We’re in...

Yeah, I was trying to avoid that but I'll give it a try. There is nil output from the command that you requested that I run.

[AMD Official Use Only - AMD Internal Distribution Only] Yes but I think it’s ignorable for now. It’s probably just me. Matthew J Cordery, PhD Principal Member of Technical Staff...

Just adding in here that many developers require the fixes that you are putting in to the latest compiler releases and that many of us work on wsl enabled machines...

Just below that. So the two instances where you try to initialize from `hypersparse_non_deg1_key_offsets` from a call to `device_span,` e.g. ``` hypersparse_non_deg1_key_offsets = raft::device_span( std::get(offsets).data(), std::get(offsets).size() - (edge_partition_deg1_hypersparse_key_offset_counts ? (*edge_partition_deg1_hypersparse_key_offset_counts)[j]...

I'm just pointing out something in your code that is not standards compliant and passes when it shouldn't. Clang catches this. Maybe nvcc interprets this differently but I don't understand...

We're porting cuGraph now and this is part of the dependency chain. I noted there is an port of it from a couple of years ago on the internal repo....

[AMD Official Use Only - AMD Internal Distribution Only] This is the error I was seeing before in 6.1. I don’t believe we’re seeing this at this point. In amd_hip_runtime_pt_api.h...