<alyssa>
The next step will be getting "advanced effects" setup
<alyssa>
Currently it's failing on gl_VertexID
<alyssa>
So next step will be to finish up my gl_VertexID series and get that landed and we should make some progress
<alyssa>
Though at some point inevitably we'll be hitting the broken transform feedback / MRT support
<alyssa>
I could probably rebase on top of lfrb's series for xfb
<alyssa>
Finishing up MRT will happen when it needs to ... right now we're decently close but not *quite* there, although it's possible for STK it might not be a huge issue.
<anarsoul>
alyssa: do you have any kind of explicit synchronization between tiler and vertex shader?
<alyssa>
anarsoul: Hardware scoreboarding.
<anarsoul>
what's that?
<anarsoul>
we have some kind of semaphores in utgard
<alyssa>
Multiple jobs are submitted to the kernel at once in a job chain (a linked list)
<alyssa>
Each job in the chain has an index assigned
<alyssa>
And each job may specify up to two dependencies on other jobs within the chain
<anarsoul>
I see
<alyssa>
So the TILER job specifies an explicit dpeendency on the corresponding VERTEX job and the preceding TILER job
<anarsoul>
for lima tiler + vs is a single job
<anarsoul>
but there're semaphores in command stream
<anarsoul>
and I'm actually not sure what it's supposed to synchronize :(
<alyssa>
The main caveat is that dependencies have to occur earlier in the list, so we can't just throw an arbitrary dep graph at the hw (need to do a topological sort every frame which is annoying but shrug)
<alyssa>
But beyond that the scheduling is handled in hardware so we don't think about it.