[Mesa-dev] [Bug 84566] Unify the format conversion code

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=84566

--- Comment #23 from Iago Toral <itoral at igalia.com> ---
(In reply to Jason Ekstrand from comment #21)
> (In reply to Iago Toral from comment #19)
> > Jason, for conversions where we cannot use a fast path in the master
> > converter (that is, where we need to unpack the src to rgba and then pack
> > from rgba to dst) you differentiate between 3 paths: uint (integer), float
> > (is_signed || bits > 8) and ubyte (otherwise).
> > 
> > While the last two paths seem to be always valid (I think we have pack and
> > unpack functions for all relevant formats for these types), I am hitting an
> > assertion in the first path because there are no uint unpack functions for
> > some types. Indeed, _mesa_unpack_uint_rgba_row only supports unpacking from
> > a limited set of uint formats. I am hitting this path in cases where dst is
> > an integer type but src is not. The code will choose the integer even though
> > the implementation requires that both src and dst are integer, since we need
> > an uint unpack function for the src in this case.
> > 
> > A solution for this would be to forget about the uint path and always go
> > with the float path or the ubyte path, but I imagine that the uint path has
> > benefits when it can be used... I guess because we could avoid
> > uint->{float,byte} transformations.
> > 
> > So, I think we have two options:
> > 1) Auto-generate uint unpack/pack functions for all required mesa formats.
> > 2) Only use the integer path if both src and dst are integer types.
> > 
> > I think 1) is only useful when src is not an integer type (so we would be
> > doing a type conversion anyway), so maybe 2) is the most sensible approach
> > in this case. What do you think?
> 
> 2 sounds like the better plan.  Can you give an example of when we are
> hitting that assertion?  I thought the GL spec forbid you from doing
> normalized -> integer conversions so I'm not seeing how this happens.

Sure, at least this conversion is hitting that:

src: MESA_FORMAT_B10G10R10A2_UNORM (GL_UNSIGNED_NORMALIZED)
dst: MESA_FORMAT_R10G10B10A2_UINT (GL_UNSIGNED_INT)

This comes from this piglit test:
bin/ext_texture_integer-texture_integer_glsl130

I think what is going on is that the src argument here is defined as
GL_RGBA_INTEGER and GL_UNSIGNED_INT_10_10_10_2, which I map to
MESA_FORMAT_A2B10G10R10_UNORM instead of MESA_FORMAT_A2B10G10R10_UINT (we don't
have MESA_FORMAT_A2B10G10R10_UINT, so that's why). This works because they have
the same pixel layout, so the conversion is the same, but then we hit this
situation.

Should we add the mesa_formats for MESA_FORMAT_A2B10G10R10_UINT and
MESA_FORMAT_A2R10G10B10_UINT too? (currently we only have
MESA_FORMAT_B10G10R10A2_UINT and MESA_FORMAT_R10G10B10A2_UINT).

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20141016/997d3aa2/attachment.html>