gallium: reorganize fragment shader execution, unbreak sse
This is probably going to get further reworked in the near future.
Right now there's a new interface wrapped around each shader execution
mode - exec, sse2, llvm. The llvm code was disabled already and has
just been moved as-is to a new file, whereas the sse2 and exec code is
actually enabled.
The way the interfaces has turned out suggests to me that the correct
approach is to actually have each shader include a pointer to a quad stage
which will do a better job of encapsulating the execution environment than
what I have here -- that's a second step however.
Update the Makefiles and includes for the new paths.
Note that there hasn't been no separation of the Makefiles yet, and make is
jumping all over the place. That will be taken care shortly. But for now, make
should work. It was tested with linux and linux-dri. Linux-cell and linux-llvm
might require some minor tweaks.
gallium: remove some debug assertions in vertex format validation
If a fragment shader references an input for which there's no vertex
shader output (ex: texcoord3), use vertex output 0 by default.
Basically, the attribute's value will be undefined. The shader writer
should never rely on undefined fragment shader inputs anyway.
Delay validation until someone really needs the vertex layout (vbuf alloc
vertex buffer or point/line/tri setup/rendering).
This will allow the vertex size to change depending on whether we're
drawing points, lines or triangles.
Several routines use shuffle patterns that are stored in memory. For
code gen, it is difficult to directly access the data segments. The
routines have been modified to be passed a pointer to a global table
of shuffle patterns.
This *should* be the last change to this file before switching over to
code gen.