[meta-freescale] GPU tests on master next - core-image-x11 - sfp

[Eric Bénard]

Le Wed, 28 Aug 2013 15:29:29 -0300,
Daiane Angolini <daiane.angolini at freescale.com> a écrit :

> On 08/28/2013 03:00 PM, Eric Bénard wrote:
> > Le Wed, 28 Aug 2013 14:20:25 -0300,
> > Daiane Angolini <daiane.angolini at freescale.com> a écrit :
> >
> >> On 08/28/2013 01:43 PM, Daiane Angolini wrote:
> >>> I've been running some tests on master-next.
> >>>
> >>> I used the following commands. Do you know any other I could run?
> >>>
> >>>
> >>
> >> The past email, was for SOFT fp
> >>> RESULTS:
> >>> $ DISPLAY=:0 glxgears
> >>> 108 frames in 5.0 seconds = 21.508 FPS
> >>>
> >>> $ DISPLAY=:0 es2gears_x11
> >>> 1494 frames in 5.0 seconds = 298.740 FPS
> >>>
> >>> $ DISPLAY=:0 glmark2-es2
> >>> glmark2 Score: 212
> >>>
> >>> $ DISPLAY=:0 glmark2-es2 --fullscreen
> >>> glmark2 Score: 229
> >>
> >> The results for HARD fp:
> >>
> >>
> >> $ DISPLAY=:0 glxgears
> >> 107 frames in 5.0 seconds = 21.395 FPS
> >>
> >> $ DISPLAY=:0 es2gears_x11
> >> 1451 frames in 5.0 seconds = 290.200 FPS
> >>
> >> $ DISPLAY=:0 glmark2-es2
> >> glmark2 Score: 213
> >>
> >> $ DISPLAY=:0 glmark2-es2 --fullscreen
> >> glmark2 Score: 228
> >>
> > so hardfp brings lower results than softfp ... interesting ;-)
> 
> I took only measure. One time. I haven't calculated the standard 
> deviation or error from measures.
> 
> I would say the results is just "equal".
> 
> And, I'm not sure how much hfp and sfp impacts the result of a task 
> executed mainly by hardware (gpu).
> 
> 
> Do you think, by the results, that there is still any bug?
> 
on the GPU point of view no.

I find interesting that the hardfp gives lower results than the
softfp : maybe that would be interesting to have true real life
benchmarks on this side (not GPU centric) before defaulting hardfp in
meta-fsl-arm for i.MX6 platforms.

Eric