SNA Continues To Be Far Better Than UXA, GLAMOR

[ickle]

Doesn't SNA work under the assumption that it has an integrated gfx card with shared memory? If yes, these numbers don't say much in the dedicated gfx card world.
Are there benchmarks comparing GLAMOR to EXA/UXA on radeon?

There are some older results for glamor vs EXA using radeon and a HD5770 (my only card), I have some new ones in the works, but suffice to say there isn't much improvement despite the advances in DPM. The outcome of those is that for a few synthetics tests Glamor is much faster than EXA on r600, but for actual desktop usage the difference is marginal and EXA notches ahead. Glamor also has far more dire fallbacks than EXA, and explodes more frequently.

[ickle]

Are there benchmarks comparing GLAMOR to EXA/UXA on radeon?

Finished the benchmark run ony my HD5770 (r600g) in a Sandybridge i5-2500. Glamor (as of today) is on average 50% faster than EXA, but is still 2x slower than the CPU for 2D graphics.

[fscan]

Finished the benchmark run ony my HD5770 (r600g) in a Sandybridge i5-2500. Glamor (as of today) is on average 50% faster than EXA, but is still 2x slower than the CPU for 2D graphics.

Oh, nice .. thx for your work. I'm wondering how it affects cpu load .. so it wouldn't be that bad if its 2x slower than cpu rendering if the cpu is free to do other stuff. 2D rendering is mostly Fast Enough (TM) anyways.

[ickle]

Oh, nice .. thx for your work. I'm wondering how it affects cpu load .. so it wouldn't be that bad if its 2x slower than cpu rendering if the cpu is free to do other stuff. 2D rendering is mostly Fast Enough (TM) anyways.

Indeed that is a very good point, something which is completely missed by the throughput benchmarking. The most important part of the display server is making sure it does not impact upon the rest of the system preventing it doing useful work and making sure that any output is presented without (noticeable) delay. On the other hand if the useful work of the system is to display an animation...

Anyway I have an idea for a different test that I've never tried before - so it is likely to so some contention points that I have so far failed to measure...

[Tyler_K]

We look forward to your patches.

(Source: http://ickle.wordpress.com/2013/10/0...h/#comment-729)

I'm curious whether your new round incorporated some of the "low hanging fruit", as it would appear that grigori did follow through: http://lists.freedesktop.org/archive...er/000403.html

[ickle]

(Source: http://ickle.wordpress.com/2013/10/0...h/#comment-729)

I'm curious whether your new round incorporated some of the "low hanging fruit", as it would appear that grigori did follow through: http://lists.freedesktop.org/archive...er/000403.html

No, I haven't tested those yet. I have to leave some excitement for tomorrow.

[Tyler_K]

No, I haven't tested those yet. I have to leave some excitement for tomorrow.

LOL, okay thanks ... and it is indeed with bated breath that we shall await upon those future results!