AMD ROCm 5.1.1 Released

[Soul_keeper]

Experimental_ROC is not an AMD project ?
I'd imagine you have something similar internally right ?
It stopped at version 2.0 and a lot has changed since.

[bridgman]

Experimental_ROC is not an AMD project ?
I'd imagine you have something similar internally right ?
It stopped at version 2.0 and a lot has changed since.

It was created and maintained by an AMD developer until he ran out of free time, but don't think it was ever an official project.

Good point re: what we have internally - I'll find out what we use for our CI systems and see if there is anything we could leverage there to simplify customer builds. Failing that will try to get a discussion going about either reviving the "experimental" tree or doing something else to simplify building.

I picked up a 6500XT so I could keep an eye on support for our entry level GPUs, so I imagine I might need to start building the stack again too.

Why do we need 50 sub projects ? this has to be the most complicated project to install that exists ...
Pretty much impossible to use for most people. Absolutely guessing which cmake options to use and which order to install these things (and which things I can get away with not installing).

It's probably comparable to building the full CUDA stack, but NVidia doesn't let you do that

Either way, I think we need to make it easier to build if only to ensure it can be picked up by distro packagers.

[Soul_keeper]

Thanks.
I just managed to hack together 10 working slackware packages and get a basic ROCm setup working.
But i'm sure there are things not being done right (ie: wrong file locations compared to the official rpm/deb).

I mostly just need to verify the cmake invocations are proper.

Some things that took me awhile to figure out:
ROCclr (didn't exist before, then was standalone, now standalone is not supported).
hipamd instead of HIP must be compiled.
-DAMD_OPENCL_PATH now needed for ROCm-OpenCL-Runtime (along with many others for each subproject).