AMD Navi Support Makes It Into DRM-Next For Linux 5.3, AMDGPU Hits Two Million Lines

[phoronix]

Phoronix: AMD Navi Support Makes It Into DRM-Next For Linux 5.3, AMDGPU Hits Two Million Lines

With the Linux kernel driver support for the upcoming "Navi" graphics cards only having been sent out last week for AMDGPU/AMDKFD, given it was more than 450 patches and more than 400 thousand lines of code (granted much of that automated header files), there was some risk it could be postponed given the imminent cut-off of new material to DRM-Next for Linux 5.3 given the rigid release cycle. Fortunately, that pull request has been honored...

http://www.phoronix.com/scan.php?page=news_item&px=AMD-Navi-DRM-Next-Lands

[hotaru]

That combined 2,038,797 lines is spread currently across 1,179 lines of code.

I'm pretty sure whatever was intended there is not what was written.

[Michael]

I'm pretty sure whatever was intended there is not what was written.

Yep files for the second one. Thanks.

[chuckula]

So they finally throw some driver code over the wall less than two weeks before the product is supposed to launch?

Not exactly what I call same day support. Considering Navi has been delayed from its original launch schedule this support should have been out late last year.

[bridgman]

So they finally throw some driver code over the wall less than two weeks before the product is supposed to launch?

Yep, although in fairness this is the first time we will have had full pre-launch upstream support for a significantly new GPU architecture.

Not exactly what I call same day support.

Right. Same day support would be two weeks from now

Seriously though, we were targeting 5.2 but the last few parts of IP review and approvals took quite a bit longer than expected... which was actually a bit of a surprise because it has been fairly predictable for the last few years.

Considering Navi has been delayed from its original launch schedule this support should have been out late last year.

Not sure what you are talking about here... are you suggesting that development was finished last year and that we've just been sitting on the product waiting for the right trade show to come along ?

[ms178]

Seriously though, we were targeting 5.2 but the last few parts of IP review and approvals took quite a bit longer than expected... which was actually a bit of a surprise because it has been fairly predictable for the last few years.

The press was highly critical with the Radeon VII launch so I hope the concerned teams at AMD spent that extra time to polish the user experience a bit better. Let's wait and see for the launch reviews soon.

[andrei_me]

bridgman how is the feeling of the team when a new hardware is launched? Do you guys feel like mission accomplished or afraid that the drivers might not work or something like that?

For example, the launch of RX 590, which supposedly was working from your side but later the OEM changed some stuff and break the support, I imagine the team was "Wat? How can it not be working? We tested it before commiting the code!"

[bridgman]

The press was highly critical with the Radeon VII launch so I hope the concerned teams at AMD spent that extra time to polish the user experience a bit better. Let's wait and see for the launch reviews soon.

Can you be more specific ? I thought the Radeon VII launch went pretty well on Linux.

[bridgman]

bridgman how is the feeling of the team when a new hardware is launched? Do you guys feel like mission accomplished or afraid that the drivers might not work or something like that?

There's always some concern about scenarios like RX 590, but in general the comfort level is a function of how much the HW changed relative to previous products. It's probably fair to say that everyone wishes for a couple of extra weeks, despite being able to start months earlier than previous generations. The changes in Navi are pretty significant so "of course I'm being paranoid but am I being sufficiently paranoid ?" is pretty much the order of the day.

For example, the launch of RX 590, which supposedly was working from your side but later the OEM changed some stuff and break the support, I imagine the team was "Wat? How can it not be working? We tested it before committing the code!"

Yeah, that was not a happy time. The drivers were heavily tested, full QA signoff, with a few layers of developer and integration testing for good measure, and everything looked good until the production boards started arriving.

I'm hoping we can arrange for at least one board partner to start officially supporting Linux or at least testing their boards with Linux... that seems like the best way to close the loop and avoid any accumulation of configuration differences between internal and production HW.

[schwarzman]

I'm hoping we can arrange for at least one board partner to start officially supporting Linux or at least testing their boards with Linux... that seems like the best way to close the loop and avoid any accumulation of configuration differences between internal and production HW.

If you manage to do that please encourage this board partner to make these efforts public somehow. For example Dell was one of the first big vendors supporting LVFS which encouraged me to buy machines from them as LVFS support is pretty convenient.