[gstreamer-bugs] [Bug 615696] Shaders using GLSL 1.20 without #version.

[GStreamer (bugzilla.gnome.org)]

https://bugzilla.gnome.org/show_bug.cgi?id=615696
  GStreamer | gst-plugins-gl | git

--- Comment #4 from Filippo Argiolas <fargiolas at gnome.org> 2010-04-18 06:38:51 UTC ---
Hey, sorry for the delay but, as always, I have very little spare time...
I tested the optimized filterblur with nvidia proprietary drivers, and the
performance gain was not so high (like 3 seconds unoptimized and 2.3 for the
optimized one). High enough to merge your improvements anyway, given the big
improvement with mesa compiler.

I managed to upgrade to fedora 13 (with experimental nouveau 3d support), and I
can see some shader failing because of array constructor and a couple of errors
about unsupported opcodes (74, IF and 99, BGNLOOP) with xray, sin and glow
effects.
The glow one can be easily solved removing "if kernel[i] != 0" blocks from
convolution shaders the other ones are more tricky and I need more time to
figure out what's happening.

I'll try to work on this (merge your work and fix the other shaders) as soon as
I have a couple of free hours.

I have a doubt: our shaders don't change over time but I always used uniforms
as function parameter to reuse the same shader to do different things basing on
the given uniforms. IIRC the orange book seemed to support this usage, but you
suggest to bake this "constant" parameters in the shader source at compile
time. Is this suggestion just a workaround for the mesa compiler not working
properly or is this a general purpose thing that should always be done with
fragment shaders (i.e. limit the uniforms usage to non constant parameters)? I
can probably easily do it but I feel like losing some part of GLSL coolness
this way.

-- 
Configure bugmail: https://bugzilla.gnome.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the QA contact for the bug.
You are the assignee for the bug.