I chose to just convert unpaired 32 bit length instructions
after parsing all instructions, although it might be possible
to determine beforehand whether there would be any lone ones,
and then even do some swapping to bring them together ...
VP outputs that should be loadable in the FP are mapped to
interpolant indices by HPOS, COL0 etc.; of course HPOS is
always written, so the highest byte of 1988 is a bitmask that
selects which components of HPOS are used for interpolants,
i.e. the FP inputs in COL0 start at index POPCNT(1988[24:28]).
Record interpolation mode for attributes while parsing declarations,
and also remember the indices of FP color inputs and FP depth output,
which has to end up in the highest output register.
We now inspect the TGSI instructions in tx_prep to determine where
temps and FP attrs are last accessed.
This will enable us to reclaim some temporaries early and we also
use it to omit pre-loading FP attributes that aren't used.
In spu_tri.c:setup_sort_vertices() triangles are culled after the
vertices are sorted. This patch moves the check a little earlier
and performs the actual check a little faster through intrinsics and
a little trickery.
Reduced code size and less work is done before a triangle is deemed
OK to skip.
It was taking approximately 50 cycles to extract the vertex indices,
calculate the vertex_header pointers and call tri_draw() for each
three vertices - .
Unrolled, it takes less than 100 cycles to extract, unpack,
calculate pointers and call tri_draw() eight times. It does have a
nasty jump-tabled switch. I'm sure that there's a better way...
Code size of spu_render.o gets larger due to the extra constants and
work in the inner loop, there are extra stack saves and loads
because there are more registers in use, and an assert. spu_tri.o
gets a little smaller.
Less briefly... Shaders need to be recompiled if their constantbuf
offsets have changed. However, since we only change them from shaders if
immediates need to be emitted, we shouldn't bother if the shader doesn't
use immediates.
Before you get all excited, this is *not* to be construed as actual support
for GLSL shaders. The GL version is still 1.3, and stuff still sucks. Just
flicking it on so that it can be tested and developed a bit easier.
This accompanies kernel patches that make GB_TILE_CONFIG's various members
completely controlled in DRM.
GB_TILE_CONFIG has the following controls:
- The number of GB (pixel) pipes enabled
- The size and style of tiling
- Subpixel precision (either 1/12 or 1/16)
Per airlied and glisse, userspace and kernel will now agree (always) on
a subpixel precision of 1/12, and tiling will always be kernel-controlled.
