AMDGPU Firmware Updated From 18.20, Vega M Blobs Added

[Divergent1]

sometimes I laugh, if there was anything critical about this firmware update it would be in Debian. The bug report shows one person with an issue, it also shows that the deb sums failed meaning he fiddled with the package,meaning Debian isnt going to fix it.even ubuntu needs it to be confirmed, by at least 2, even then ubuntu wont fix it, I have seen bug reports of up to 12 people ignored by ubuntu. If you are having an issue with Debian complaining here will not get it fixed, file a proper bug report. If the updated firmware works with the 4.18 kernel, they will both hit the debian repo at the same time. Sometime back you could install AMDgpu Pro on Debian from the Steam repo, that package probably has everything needed to take advantage of the new firmware.

[drSeehas]

They skipped all GCN 1.2 parts: fiji, tonga, carrizo, stoney, topaz

Tonga, Fiji, ... are GCN 3.

[parkerlreed]

Heads up. Tried linux-firmware-git on Arch and the Polaris10 firmware causes driver crashes. https://gist.githubusercontent.com/p.../gistfile1.txt

Where exactly would I report this? Would I file a driver bug against AMDGPU or somehow against the linux-firmware file themselves?

[bridgman]

Tonga, Fiji, ... are GCN 3.

They're both - half the media uses "1, 1.1, 1.2..." notation and the other half uses "1, 2, 3". We don't use either (although we tried briefly) - we just talk about IP block versions, where the graphics core in Tonga, Fiji etc... is GFX8.

[drSeehas]

... We don't use either (although we tried briefly) - we just talk about IP block versions, where the graphics core in Tonga, Fiji etc... is GFX8.

But POLARISxx is also GFX8.
How do you differentiate between TONGA and POLARIS12?

[bridgman]

But POLARISxx is also GFX8.
How do you differentiate between TONGA and POLARIS12?

If the programming models were significantly different between the chips we would treat them as different IP block versions, eg 8.0 vs 8.1 vs 8.2, but if the changes were smaller we would just use runtime logic switching on chip type (based in turn on device & revision IDs) to deal with the differences between chips.

In this case we have separate header files for GFX 8.0 and 8.1, but looking at the code I think we only #include the gfx80 headers and do the rest with runtime logic.