On 07/17/2014 12:04 PM, Jason Ekstrand wrote:
This is the first installment of some work I've been doing over the past
couple of weeks to refactor mesa's texture conversion/storage code. There
is more to be done and more that I have done but have not included in this
series. This is the first mailing-list-ready fruits of my efforts. The
important bits here include:
1) Using a human-readable CSV file to describe texture formats similar to
the way it is currently don in gallium. This is much easier to
read/edit than the structure in formats.c. The guts of formats.c is
then autogenerated from this CSV file.
I'm kind of on the fence about this. Some of us have been hoping that
we'd eventually consolidate some of the Mesa and gallium code so we
wouldn't have duplicated/parallel code. The format code is a good
example. In theory, the MESA_FORMAT_ stuff could be replaced by the
gallium PIPE_FORMAT_ code.
I kind of like the hand-written format code since it's simple to
understand and update. Now you've got 600 lines of (uncommented) Python
code to figure out and maintain. When something breaks it's important
that the code is easily maintainable if you're not around.
2) Adding a very generic yet efficient _mesa_swizzle_and_convert function
that is capable of performing the vast majority of texture format
conversions in one function. It has also been fairly carefully tuned
to be even faster than the _mesa_swizzle_ubyte_image special-case that
we had before for ubyte textures only it also works on the other
datatypes and can even do type conversions as it swizzles.
3) Refactoring of texstore.c including the use of the above
_mesa_swizzle_and_convert function along with the already-existing
packing functions to remove a lot of hand-written special-case code.
This part generally looks good. Comments already posted.
Thanks to the format CSV file, there's even more that we can now do.
Things I hope to accomplish in the future include:
1) Autogenerate the bulk of main/format_pack.c, main/format_unpack.c, and
main/pack.c from CSV files. There's some refactoring that will be
required first, but it shouldn't be that hard and I already have the
python code to do the generation; it's just not part of this patch
series.
Again, this kind of thing was already done for gallium. It's kind of
sad to effectively duplicate it.
I'm interested in hearing other opinions.
-Brian
2) Find a general way to do depth-stencil formats. I'm a bit dubious as
to whether or not this will turn out to be practical, but I haven't had
a chance to look into it too much yet.
2) Do similar refactors for GetTexImage, ReadPixels, and DrawPixels.
Happy Reviewing!
--Jason Ekstrand
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
