[virglrenderer-devel] getting multisample position into the guest

[Gert Wollny]

Am Montag, den 25.06.2018, 16:16 +1000 schrieb Dave Airlie:
> On 23 June 2018 at 00:51, Gert Wollny <gert.wollny at collabora.com>
> wrote:
> > Dear all,
> > 
> > to fix a series of gles31 tests the call to glGetMultisamplevf has
> > to be implemented in virgl. As I understand it, these values
> > returned by this function are usually fixed by the hardware, so
> > would you agree that putting them in the caps would be an option?
> > In total it would be "16 float values and one int (number of actual
> > samples)".
> > 
> 
> Yeah I expect we need to put them in the caps, there might be a semi-
> sane set of defaults we could use in the guest, though don't we need
> more values? One set per number of samples? 
Indeed, that's correct now that I looked closer at how other drivers do
it, I saw it too. 

> 
> It might also help to advertise how many samples we support, at the
> moment we end up a bit broken here (those 6 sample tests.)
Yes, of course, I meant to write this. 

In summary, if we support a maximum of 16 then we have to take the
samples for 2, 4, 8 , and 16 into account. If we want to save space we
could encode it like the Intel driver does. They need in total only 8
uint32_t values to encode all sample postions. I think I'll test how
well this works out on r600 regarding the tests.  

Best, 
Gert