| Age | Commit message (Collapse) | Author |
|---|
|
|
|
|
|
|
|
|
|
This is a better place than in u_rect.c
|
|
|
|
Disabled as it doesn't make VS/PSPrecision DCT happy, and it would
unnecessarily slow some cases where it is not needed.
|
|
|
|
|
|
|
|
|
|
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>
|
|
supplied for tgsi translation.
|
|
PIPE_FORMAT_R10G10B10X2_USCALED, half floats, were not supported, so
just rely on u_format for (almost) universal format support.
|
|
Unneeded since we code generate the vertex fecthes.
|
|
|
|
It was only used for D3D's REP/END/BGNFOR/ENDFOR. D3D's aL register is
just like another address register now.
|
|
|
|
Saves time and trouble.
|
|
we were resetting the mask on each new break/continue statement within
the same scope. we always need to and the current execution mask
with the current break/continue mask to get the correct result (the
masks are always ~1 initially)
|
|
Re-add commit 2d65a7caf97684aa654088c76a74b632fbd685fa
Signed-off-by: Xavier Chantry <chantry.xavier@gmail.com>
Signed-off-by: José Fonseca <jfonseca@vmware.com>
|
|
If we're using a wrap mode in which border color sampling is possible
it means that texcoords may be outside of the texture image bounds.
Fetching the texel may result in a segfault.
Use the 'use_border' variable to catch such texcoords and replace
the texel offset with zero (which will be in bounds).
Fixes segfault in Lightsmark demo, fd.o bug 27877.
|
|
|
|
|
|
In no particular order:
* Make list const
* Add function comments
* Clearly state that demo lists are not complete
* Fix whitespace
* Use __FUNCTION__ instead of __func__
* Add unimplemented check which always fail
Thanks Brian and Keith.
|
|
|
|
|
|
if fetch_count % 4 != 0 then on the last iteration we fetch garbage.
this patch makes sure we stay within bounds
|
|
|
|
|
|
|
|
|
|
|
|
add function to set sample mask, and state for alpha-to-coverage and
alpha-to-one. Also make it possible to query for supported sample count
with is_msaa_supported().
Use explicit resource_resolve() to resolve a resource. Note that it is illegal
to bind a unresolved resource as a sampler view, must be resolved first (as per
d3d10 and OGL APIs, binding unresolved resource would mean that special texture
fetch functions need to be used which give explicit control over what samples
to fetch, which isn't supported yet).
Also change surface_fill() and surface_copy() to operate directly on resources.
Blits should operate directly on resources, most often state trackers just used
get_tex_surface() then did a blit. Note this also means the blit bind flags are
gone, if a driver implements this functionality it is expected to handle it for
all resources having depth_stencil/render_target/sampler_view bind flags (might
even require it for all bind flags?).
Might want to introduce quality levels for MSAA later.
Might need to revisit this for hw which does instant resolve.
|
|
|
|
|
|
max_index must be observed to prevent crashes due to bad index data.
I've been using this patch for some time without regressions.
Some places, where we use internal vertex buffer, it is not entirely
clear what max_index should be, so passing just ~0 to avoid regressions
for now.
|
|
|
|
The POSIX function pthread_cond_wait can have spurious wakeups when
waiting on a condition variable.
Add a 64-bit counter that is incremented whenever the barrier becomes
full. A woken thread checks the counter. If the counter has not changed
then it has been spuriously woken and goes back to sleep. If the counter
has changed then it was properly signaled and exits the barrier.
Tested on Mac OS X.
This patch was based on ideas from Luca Barbieri.
|
|
This allows u_sampler_view_default_dx9_template to do its magic on DX9.
|
|
Everybody should respect max_index, specially llvm generated code, which
likes to eat vertices 4 at a time, so it may end up chew a bit a bit more
than actually exists.
|
|
Avoid hiding existing variable already named info in outer scope.
|
|
info cannot be NULL at the call to debug_printf. emit_instruction
dereferences info, so at debug_printf it is either not NULL or the
program has already crashed.
|
|
Fixes glean glsl1 test: texture2D(), with bias.
|
|
|
|
|
|
|
|
Otherwise drivers that bake the full blend state in a key end up having
uninitialized memory in their key.
Courtesy of valgrind.
|
|
As the argument in general might not be a constant.
|
|
This silences Coverity assign_where_compare_meant warnings.
|
