On Fri, Mar 3, 2017 at 9:46 AM, Michel Dänzer <[email protected]> wrote:
> From: Michel Dänzer <[email protected]>
>
> This reverts commit 950ffb8d6fd1480f305e38c571bda44f247f1de2 (and parts
> of commit 4b5326aeba539249fcded91bf7806a708eeca651).
>
> It halved (or worse) the performance with some x11perf tests on
> radeonsi, even though radeonsi has a BO cache.
>
> Signed-off-by: Michel Dänzer <[email protected]>
> ---
>
> I suspect the reason why radeonsi's BO cache can't make up for the lack of
> the glamor FBO cache is that the former can only return idle BOs. So
> while the glamor FBO cache allows re-using the same FBO over and over
> for a series of operations using temporary pixmaps of the same size, the
> radeonsi BO cache has to keep using different BOs until a command stream
> is flushed to the GPU and processed by it.

I don't think the radeonsi cache is sufficient here. It only hides the
kernel overhead. It doesn't hide any overhead of Mesa itself.

Marek
_______________________________________________
[email protected]: X.Org development
Archives: http://lists.x.org/archives/xorg-devel
Info: https://lists.x.org/mailman/listinfo/xorg-devel

Reply via email to