[Mesa-dev] [PATCH 3/5] panfrost: Disable PIPE_CAP_TGSI_TEXCOORD

[Connor Abbott]

On Fri, Mar 15, 2019 at 5:00 PM Ilia Mirkin <imirkin at alum.mit.edu> wrote:
>
> On Fri, Mar 15, 2019 at 11:52 AM Connor Abbott <cwabbott0 at gmail.com>
wrote:
> >
> > On Fri, Mar 15, 2019 at 3:46 PM Ilia Mirkin <imirkin at alum.mit.edu>
wrote:
> > >
> > > On Fri, Mar 15, 2019 at 10:29 AM Alyssa Rosenzweig <
alyssa at rosenzweig.io> wrote:
> > > >
> > > > > This is needed if you can only handle point sprites on certain
> > > > > varyings but not others. This is the case for nv30- and nvc0-based
> > > > > GPUs, which is why the cap was introduced.
> > > > >
> > > > > If your GPU does not have such restrictions, you can safely
remove the cap.
> > > >
> > > > Ah-ha, I see, thank you.
> > > >
> > > > I can't handle point sprites on any varyings (they all have to get
> > > > lowered to gl_PointCoord), so.. :)
> > >
> > > Yeah, that's not extremely surprising. Point sprites weren't a thing
> > > in GLES 1, and only available as gl_PointCoord in GLES 2. With GL 1.x
> > > + multitexture, you can have up to 8 tex coords used in your fragment
> > > pipeline, and so need a fixed-function way of setting them to point
> > > coords when drawing points. Hence the current TEXCOORD semantics of
> > > just bumping generic varyings by 8, so that we can be sure that any
> > > shaders with true gl_TexCoord[] usage get the low 8 varyings assigned
> > > (and which, on nvc0+ go through "special" shader outputs, subject to
> > > the replacement).
> > >
> > > Note that e.g. Adreno A3xx+ is capable of doing the replacement
> > > everywhere, despite being a GLES-targeted GPU. So it's not out of the
> > > realm of possibility that you just haven't figured out the full
> > > mechanism for doing it yet.
> >
> > Can you replace more than one varying and output multiple points vertex
shader, or something crazy like that? On Mali gl_PointCoord is just a
normal varying whose descriptor points to a special point-coord buffer that
gets fed to the tiler, so we should be able to make an arbitrary varying
"turn into" gl_PointCoord just by changing its descriptor at draw time.
>
> Yeah, any (gl_TexCoord) varying may get replaced by a point sprite
> coord (in GL 1.x / GL 2.x). Note that there's additional complication
> from coordinate inversion, when you're drawing on the winsys fb vs an
> fbo. So you have to be able to generate both S,T and S,1-T. [Or
> perhaps there's an explicit control for it, I forget.] With GLES2+,
> it's just gl_PointCoord though.
>
> The way gallium handles this is that you get a mask of varyings to
> replace in
https://cgit.freedesktop.org/mesa/mesa/tree/src/gallium/include/pipe/p_state.h#n188
> + whether to invert or not (sprite_coord_mode).

Oh, whoops... I guess I was thinking of gl_PointSize. gl_PointCoord in the
fragment shader does indeed come through an extra-special varying, at least
the way the blob does it.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20190315/dd3a6dcf/attachment.html>