[SLOF] [PATCH 3/4] fbuffer: Implement MRMOVE as an accelerated primitive

[Thomas Huth]

On 03/08/15 17:53, Thomas Huth wrote:
> On 03/08/15 12:37, Nikunj A Dadhania wrote:
>> Thomas Huth <thuth at redhat.com> writes:
>>
>>> The character drawing function fb8-draw-character uses "mrmove"
>>> (which moves main memory contents to IO memory) to copy the data
>>> of the character from main memory to the frame buffer. However,
>>> the current implementation of "mrmove"  performs quite badly on
>>> board-qemu since it triggers a hypercall for each memory access
>>> (e.g. for each 8 bytes that are transfered).
>>> But since the KVMPPC_H_LOGICAL_MEMOP hypercall can transfer bigger
>>> regions at once, we can accelerate the character drawing quite a
>>> bit by simply mapping the "mrmove" to the same macro that is
>>> already used for the "rmove". For keeping board-js2x in sync,
>>> this patch also transforms the "mrmove" for js2x into primitives.
>>>
>>> Signed-off-by: Thomas Huth <thuth at redhat.com>
>>
>> I dont have a js2x handy, did you test this on js2x?
> 
> No, sorry, unfortunately, I also was not able to test this on js2x yet.
> I still have a YDL PowerStation somewhere in a corner of my flat, but
> it's currently not set up ... will do that one day when I got enough
> spare time again, but that won't happen within the next few weeks (KVM
> forum's ahead!). So if there are problems, I'll fix them up as soon as I
> got the PowerStation running again (and I guess there might be other
> problems, too, since board-js2x hardly got any testing within the last
> months/years, I think).

FWIW, I've recently dusted off my PowerStation and gave it a try...
the HEAD of the SLOF master branch is unfortunately quite broken on js2x
nowadays (I'll try to send some first fixes later), but I was able to
cherry-pick the fbuffer acceleration patches to an older level of SLOF
from 2012 which was still working nicely.

So the screen acceleration patches are working fine on board-js2x, and
also result in a nice speedup - the "etest" benchmark that I've posted
in an earlier mail went down from 1081 ms to 101 ms, and a simple
"erase-screen" now takes 11 ms instead of 135 ms.

 Thomas