r600/shader: emit tessellation factors to GDS at end of TCS.

[Dieter Nützel]

Am 15.01.2016 21:04, schrieb Dave Airlie:
> On 15 January 2016 at 12:25, Dieter Nützel <Dieter at nuetzel-hh.de> 
> wrote:
>> Hello Dave,

>> Second:
>> What could be differend between r6xx/r7xx and EG+ (NI, here) that the 
>> below
>> fixes the 'gsraytrace' GPU hang on the later but NOT on the former?
>> 
>> Any ideas, hints where I could start?
>> r6xx/r7xx do not have any TESS support as we now.
>> So where to start?
>> Disabling some lines of your commit and try?
> 
> This commit just fixed things by luck on EG I suppose,

I called it by 'accident' ;-)
Look, here:
Bug 91865 - [r600g] GPU hang in 'gsraytrace' - NI/Turks (6670)
https://bugs.freedesktop.org/show_bug.cgi?id=91865#c10

> it must be the NOP
> at the end of the shader, and I think on r600/700 that could should
> still happen,
> 
> I'd probably need to spend some time staring at fglrx dumps to work out 
> why
> we crash, does R600_DEBUG=nosb help at all?

Not at all, for me on EG and RV730.

With the commit it crash on RV730 for 'gsraytrace':
Bug 83319 - [r600g] GPU lockup in gsraytrace (Mesa-demo-8.2.0) - RV730
https://bugs.freedesktop.org/show_bug.cgi?id=83319#c15

For all other cases with R600_DEBUG=nosb
vsraytrace and fsraytrace on RV730,
vsraytrace, fsraytrace and gsraytrace on EG it show the damages like
below (see attachments):
Bug 93715 - [r600g] rendering issues with R600_DEBUG=nosb for all three 
'raytrace' variants

BTW gsraytrace on EG is now better than with fglrx (15.11?)

Let me know when I should test sth on RV730.
Thanks for looking at it.

-Dieter