[Mesa-dev] [PATCH] virgl: clear render state before submitting clear command

[Gert Wollny]

Am Montag, den 07.05.2018, 18:25 -0700 schrieb Gurchetan Singh:
> In vrend_clear, we already save and restore colormasks and stencils:
> 
> https://cgit.freedesktop.org/virglrenderer/commit/?id=b75e0a1dabdfbda
> 44c310a69026a9dbd7d980294
> https://cgit.freedesktop.org/virglrenderer/commit/?id=252b00d77c30ce3
> 9608c1a9de18523cbdcaca623
> 
> It would be nice if we can put everything on the host side. 
Okay, I can see your point, but may instead of going through all these
states one by one we could something on the host side that does 

 - push the state
 - set clean state 
 - glClear
 - pop state

and especially, factoring out these steps into different functions,
because the host clear function is getting quite messy.

The alternative would be to not do anything about these states on the
host side and just send a clear state :). IMHO the whole clear
implementation is a performance disaster anyway, but I'm afraid that
within the limits of gallium it can't be solved in a better way. 

>  Wouldn't the following code also solve the RASTERIZER_DISCARD
> problem?
> 
> if (ctx->sub->hw_rs_state.rasterizer_discard)
>     glDisable(GL_RASTERIZER_DISCARD)
> 
> ...
> 
> glClear(..)
> 
> ...
> 
> if (ctx->sub->hw_rs_state.rasterizer_discard)
>     glEnable(GL_RASTERIZER_DISCARD)
This was not enough, some tests were still failing (i.e. all the
*rasterizer_discard.fbo.depth* test), and I thought instead of figuring
out one by one which states might have to be reset on the host it is
easier to just send a clear state.

BTW: this also fixed 

dEQP-
GLES3.functional.shaders.indexing.moredynamic.with_value_from_indexing_
expression_fragment

and since I never really looked into that test, I don't yet know which
rasterization state this one might enable and not disable on the host
before clear.

Anyway, If you confirm that we should do it on the host side, then I
will retract this patch and prepare something for virglrenderer. 

Best, 
Gert