[Xf86-video-armsoc] n-buffering support causes too many flips on Mali-400

Daniel Drake drake at endlessm.com
Fri Dec 19 04:57:08 PST 2014


On Mon, Nov 17, 2014 at 12:11 AM, Sunny Sun <Sunny.Sun at arm.com> wrote:
> Hi Daniel
> As I replied to you two months ago, this issue has been fixed with mali400 release r5p0-01rel0.

Just to close this thread: I retested with r5p0, and indeed, FPS is
now 60 (same as vsync).
What's a little strange is that if I go back to a previous commit,
before n-buffering support was implemented, FPS goes up to 90. Maybe
nothing to worry about given that the latest code works fine...

Thanks,
Daniel

> -----Original Message-----
> From: Xf86-video-armsoc [mailto:xf86-video-armsoc-bounces at lists.freedesktop.org] On Behalf Of Daniel Drake
> Sent: Saturday, November 15, 2014 1:34 AM
> To: xf86-video-armsoc at lists.freedesktop.org
> Cc: Eskil Sund
> Subject: [Xf86-video-armsoc] n-buffering support causes too many flips on Mali-400
>
> Hi,
>
> I am experiencing an issue with the following commit:
>
> commit 1d13da045ced12a123db184fff677b0de41d80e6
> Author: Eskil Sund <eskil.sund at arm.com>
> Date:   Tue Apr 8 13:11:06 2014 +0000
>
>     Enabling n-buffering support when using page flip events.
>
>
> For the test below I run armsoc with this commit as HEAD, on ODROID-X2, Mali-400 driver version r4p0:
> VARIANT=mali400-r1p1-gles11-gles20-linux-monolithic-no_profiling-x11-dma_buf
>
> My HDMI display is running at 60Hz vertical refresh.
>
> I start X, then the simple metacity window manager (doesn't use OpenGL, no compositing) and then I run the following test app:
> https://gist.github.com/dsd/6d17d596501a01c4aa0e
>
> This app updates the screen using eglSwapBuffers, which should be synced to vblank interrupts.
> However, on this driver version, the app reports 120 frames per second (i.e. twice too many), and there is a lot of tearing visible on screen.
>
> If I go back to the previous commit, before n-buffering, the test case now runs fine, 60fps and no visual problems.
>
> Is this a known issue, has it been investigated?
>
> Thanks
> Daniel


More information about the Xf86-video-armsoc mailing list