gfx1100 added to CI
Hey all,
On 2024-03-06 23:54, Cordell Bloor wrote:
> Unless there are concerns raised, I will migrate rocm-hipamd tomorrow.
I think we need results for gfx110n ASAP, so I went ahead and
temporarily repurposed a box of mine, added the W7800 (gfx1100), and
activated a podman-based worker.
I hooked up the worker to ci-test.rocm.debian.net. If the host proves
stable enough, I'll switch it over to ci.rocm.debian.net tomorrow and
let the scheduler pick it up.
I'll try a QEMU-based worker before doing that.
Feel free to submit jobs for gfx1100 on ci-test! But note that packages
must have at least one test in a "pure" release for the results to show
up in the UI. This is by (debci's) design.
For example, when testing rocrand, I immediately pinned src:rocrand and
src:rocm-hipamd to experimental, but never saw results. To reach the
test history, I needed to schedule a test for rocrand in unstable
without any pins.
The above should resolve itself once rocm-hipamd reaches unstable.
Best,
Christian
PS: When loading, the amdgpu driver complains
> [drm:amdgpu_ras_eeprom_init [amdgpu]] *ERROR* Failed to read EEPROM table header, res:-5
I haven't investigated yet, but it did not seem to affect the rocrand
test, which passed. Haven't tried others yet.
Reply to: