Comment 123 for bug 252094

(In reply to comment #13)
> There's a lot of places where EXA is faster than XAA as well but I'll upload my
> full results from each. I had a system crash during the EXA test (unrelated to
> video) so I don't have full results but hopefully this sheds some light.

Dan,

Thanks for posting the results from this performance testing. There's definitely something interesting to see in the results from the 4 variants of the copy tests (shown below for the 10x10 tests and the 500x500 tests).

For most of the EXA tests the performance is basically unchanged when changing from 10x10 to 500x500. That suggests that there is good GPU acceleration happening. Contrast that with the XAA results where performance generally falls dramatically from 10x10 to 500x500 suggesting that the CPU is getting involved on a per-pixel basis.

The big exception is the pixmap to pixmap test where not only does EXA get dramatically slower from 10x10 to 500x500, but it's also from 2x to 14x slower than XAA for this case. So there's obviously something quite broken there, (perhaps migration for a fallback).

If you've got a recent xf86-video-intel driver (from git) then you can turn on the following option (in the device section of your xorg.conf file):

    Option "FallbackDebug" "true"

and then check your Xorg.#.log file to see if there are fallbacks that correspond to the slow behavior you're seeing.

-Carl

EXA
===
winwin: 10: 1370000
        500: 1330000

pixwin: 10: 1570000
        500: 1550000

winpix: 10: 1420000
        500: 1400000

pixpix: 10: 83700
        500: 1740

XAA
===
winwin: 10: 280000
        500: 1730

pixwin: 10: 206000
        500: 247

winpix: 10: 65100
        500: 52

pixpix: 10: 1210000
        500: 4380