Another comment:
What would solve your DRI_PRIME issues would also be
dma-buf fences.
While I believe thread_submit should be a bit better (because
it avoids a card stall waiting for another card to finish), dma-buf
fences make the two cards synchronize rendering properly.
I don't know much of the kernel details, but as I understand radeon and
amdgpu support
the kernel interface for dma-buf fences, but intel not. You should be
looking
to the intel team to implement the feature.
Axel
On 12/05/2016 07:46, Axel Davy wrote:
On 12/05/2016 04:41, Mike Lothian wrote:
Hi Axel
Is the thread_submit=true only for nine or does it work with all of DRI3?
I'm keen to get rid of the tearing on my Skylake/Tonga setup
Thanks
Mike
It is gallium nine only for now.
In the case of gallium nine, it as a reasonnable assumption all
accesses to the rendered content (read/write) is done through the api.
Besides the window messages are independent of whether there is
presentation or not (and when). Thus you can delay the presentation.
I believe in the case of GL, it is too much to do these assumptions.
Though perhaps the option could still be added and user test if for
their app it works or not. But it's not trivial work to add support
for the option.
Axel
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
