fix: Can't detect librocm 6.0.x

[imwints]

librocm reports it's version as 7.0.0.0 in the 6.0.x release series.

Closes: #759

[aristocratos]

@imwints
Well, that's not confusing at all...

But are you sure the minor version won't change until version 7?

Wouldn't it make sense that version 7 would be 8.0.0.0 internally?

[aristocratos]

Why is that not confusing?

That was irony...

But I was asking why you are checking version.minor == 0, we are only concerned with a major release change, since AMD hopefully won't do any breaking changes until next major release.

And by the current logic that would be 8.0.0.0 (or 9.0.0.0 if they are skipping one again).

[imwints]

:) went over my head

And by the current logic that would be 8.0.0.0 (or 9.0.0.0 if they are skipping one again).

I just thought if they change their scheme again and release a 7.x with the version 7.1.0.0 or similar this is save for now and would only accept the real 6.x for now. I just wanted to limit the allowed version range so this just wouldn't allow any new release. But I'm happy to remove that

[aristocratos]

Hm, there is also a rsmi_version_str_get() function that returns a string.

Descriptions:

rsmi_version_get(): Get the build version information for the currently running build of RSMI

rsmi_version_str_get(): Get the driver version string for the current system.

Might be an idea to test rsmi_version_str_get() instead and check if it's correct then if so check if first char is 5 or 6.

@imwints Do you have a AMD card to test with?

The definition is at: https://rocm.docs.amd.com/projects/rocm_smi_lib/en/docs-6.0.0/.doxygen/docBin/html/group__VersQuer.html#ga6d4f034afcef2e50b4e8392c5dfd7a57

[imwints]

This will just give you the output of uname -r

[alerque]

c.f. Upstream issue filed by @imwints.

[aristocratos]

This will just give you the output of uname -r

Huh, rsmi_version_str_get() gives you the output of uname -r?

That makes even less sense to me...

[imwints]

Here is the implementation

[aristocratos]

Will merge and create a new release later today, since I'm guessing it's unlikely anything will come from the issue in the ROCm repository in any reasonable time.