| Age | Commit message (Collapse) | Author |
|---|
|
verts per primitive, not total count
|
|
spotted by Keith
|
|
Keith came up with a new way of running the pipeline which involves passing
a few info structs around (for fetch, vertices and prims) and allows us
to correctly handle cases where we endup with multiple primitives generated
by the pipeline itself.
|
|
|
|
|
|
|
|
|
|
done to handle strips correctly as inputs - we need to decompose
the strips
|
|
don't overwrite the inputs and make sure the correct primitive
is used on entry
|
|
|
|
|
|
|
|
|
|
of vertices
lots and lots of fixes for geometry shaders. in particular now we work when the gs
emits a different primitive than the one the pipeline was started with and also
we work when gs emits more vertices than would fit in the original buffer.
|
|
|
|
plus used the correctly specified stride
|
|
|
|
interface wise we have everything needed by d3d10 and gl transform feedback.
the draw module misses implementation of some corner cases (e.g. when stream
output wants different number of components per output than normal rendering
paths)
|
|
aka transform feedback
|
|
They aren't implemented for llvm.
|
|
|
|
|
|
|
|
|
|
Use front/back instead of cw/ccw throughout.
Also, use offset_point/line/fill instead of offset_cw/ccw.
Brings gallium representation of this state into line with its main
user, and also what turns out to be the most common hardware
representation.
This fixes a long-standing bias in the interface towards the
architecture of the software rasterizer.
|
|
|
|
|
|
|
|
Note that this also requires X86 for llvm, if llvmpipe/draw_llvm works
on PPC then the condition should be extended to include && x86.
Signed-off-by: Török Edwin <edwintorok@gmail.com>
Signed-off-by: José Fonseca <jfonseca@vmware.com>
|
|
This fixes edge flags for polygons and provoking vertex for
filled quads.
|
|
|
|
Now we have draw_elements() and draw_arrays() to be consistent with the
pipe_context drawing functions.
|
|
Builds on commit ddb0e18f6c5582d4d2cc59ffd16ad9c4639ed059 and fixes
regressions in glean clipFlat test.
We assume that Gallium drivers observe flatshade_first for all triangles
and that all the assorted per-triangle calls in the 'draw' module also
follow flatshade_first. Everything else builds on those rules.
Gallium does not use follow flatshade_first for GL quads, quad strips
and polygons; the "last" vertex is always the provoking vertex for those
prims. So now there are separate QUAD_FIRST_PV and QUAD_LAST_PV macros
in the draw primitive decomposition code instead of one QUAD macro.
|
|
|
|
Based on John Lawless' feedback.
|
|
|
|
This gives a ~30% shader optimization time improvement on blender.
Tested by comparing the dumped LLVM modules.
Current ordering:
time ~/llvm-git/obj/Release-Asserts/bin/opt l.bc -constprop -instcombine
-mem2reg -gvn -simplifycfg
real 0m1.126s
user 0m1.108s
sys 0m0.012s
With this patch:
time ~/llvm-git/obj/Release-Asserts/bin/opt l.bc -mem2reg -constprop -instcombine -gvn -simplifycfg
real 0m0.885s
user 0m0.880s
sys 0m0.000s
The overall improvement in blender is ~15%.
Blender without the patch takes 1m13s:
edwin 5934 87.6 11.5 729440 458296 pts/5 SLl+ 17:35 1:13 blender
Blender with the patch takes 1m3s:
edwin 5726 94.2 11.2 716424 446168 pts/5 SLl+ 17:32 1:03 blender
It is still slow with the patch, but better (most of the optimization time is
taken up by GVN, see LLVM PR7023).
Signed-off-by: Török Edwin <edwintorok@gmail.com>
Signed-off-by: José Fonseca <jfonseca@vmware.com>
|
|
Unneeded since we code generate the vertex fecthes.
|
|
Saves time and trouble.
|
|
|
|
|
|
if fetch_count % 4 != 0 then on the last iteration we fetch garbage.
this patch makes sure we stay within bounds
|
|
|
|
|
|
|
|
|
|
|
|
|
|
we were only running the llvm paths when the input elts were linear,
now we can handle abritrary fetch elts arrays. we do this by generating
two paths - linear and fetch_elts one and just selecting the right one
at run time.
|
|
|
