Re: kern/57059 (amdgpu graphics ring test failing)

[Taylor R Campbell <riastradh%NetBSD.org@localhost>]

> Date: Wed, 12 Jul 2023 14:35:07 +0000
> From: Bruno Melo <bmelo%protonmail.com@localhost>
> 
> I tried removing the volatile keyword but the error persists the
> same. Any reason for putting this volatile?

The reason for the volatile keyword is to ensure the floating-point
operations that lead into the computation of bw_limit_passed have all
completed before DC_FP_END().

However, evidently there's other issues like this -- as I suspected
from code inspection -- so I'm working on another patch to make sure
DC_FP_START/END is consistently called around all the entry points
rather than trying to rely on using it in-line.

The volatile hack was just a quick way to test for progress --
confirmed; now to do it properly.  If you want to try a little more
progress, you could try putting DC_FP_START/END around, say, the body
of dc_create_resource_pool (and maybe around the calls to
dc->hwss.prepare_bandwidth and dc->hwss.validate_bandwidth and
dc->hwss.optimize_bandwidth).  Need to think more about where exactly
this needs to go.