Etnaviv Landing Performance Counters For Linux 4.15

[phoronix]

Phoronix: Etnaviv Landing Performance Counters For Linux 4.15

Lucas Stach has sent in the Etnaviv DRM driver changes to DRM-Next for the Linux 4.15 kernel. This is one of the bigger pull requests for this reverse-engineered, open-source Vivante graphics driver...

http://www.phoronix.com/scan.php?page=news_item&px=Etnaviv-Changes-Linux-4.15

[schmidtbag]

Will we be seeing benchmarks of any of this, including against the closed-source drivers? I don't recall ever seeing any benchmarks for these SoCs. Nowadays, i.MX6 devices are relatively cheap.

[Michael]

Will we be seeing benchmarks of any of this, including against the closed-source drivers? I don't recall ever seeing any benchmarks for these SoCs. Nowadays, i.MX6 devices are relatively cheap.

I don't think I have any supported hardware

[schmidtbag]

I don't think I have any supported hardware

If you are interested in testing these, the cheapest board I'm aware of is the UDOO Neo Basic, which is $50:

https://shop.udoo.org/usa/neo/udoo-neo-basic.html

But you might want the Full version since that has ethernet:

https://shop.udoo.org/usa/neo/udoo-neo-full.html

There could be cheaper alternatives, but I'm not aware of anything that offer USB, Ethernet, and HDMI on-board.

Most i.MX6 models with 2+ cores start at around $100, but if you're mostly just tracking the functionality and performance of Etnaviv, I figure these single-core systems ought to be "good enough".

[Lynxeye]

I don't think I have any supported hardware

If I remember correctly you should have an CompuLab Utilite somewhere in your stack of hardware. At least there was an article about this on some 3 years ago. You should be able to install something like Fedora for ARM on that thing, which would allow you to test Etnaviv.

It will probably be a lot harder to get the blob running, so side-by-side tests may not fly.

[Terrablit]

Yeah, he probably won't even be able to run the same kernel version on ARM hardware when comparing drivers, since much of the ARM universe has their closed drivers linked with older kernels, while only the newest kernels have the open drivers. And it's hard to remove other constraints like system memory and I/O. Not to mention there's not going to be nearly as many 3D applications that run on ARM.

But at least he could track performance across builds, I guess.

[caligula]

Will we be seeing benchmarks of any of this, including against the closed-source drivers? I don't recall ever seeing any benchmarks for these SoCs. Nowadays, i.MX6 devices are relatively cheap.

Yes compared to Core i7, but there's quite a bit of competition among ARM boards. I wouldn't call $100 for a dual core cheap. With $100 you can have three RPi3s or 10 Orange Pis.. both quad-core.

[starshipeleven]

Yes compared to Core i7, but there's quite a bit of competition among ARM boards. I wouldn't call $100 for a dual core cheap. With $100 you can have three RPi3s or 10 Orange Pis.. both quad-core.

You get a bit more than just the dual core for 100$.

And Orange Pi is just one of many random shit SBCs with Allwinner SoCs and Mali GPU.

[Guest]

IIRC GCW zero was a gaming handheld with Vivante GPU, they were also using Etnaviv. Of course, it mainly ran emulators for older console games.

[lowflyer]

How do these counters count the "Landing Performance"?