Age | Commit message (Collapse) | Author |
|
|
|
|
|
Was happening when object only had curves (doe example, object with hair
particle system and emitter rendering disabled).
|
|
for QBVH
|
|
|
|
camera is in volume
|
|
OpenSubdiv doesn't like empty meshes, so we need to be careful not to
subdivide when the mesh is empty.
|
|
Changes from microdisplacement work broke previous support for subdivision
meshes, sometimes leading to crashes; this makes things work again. Files
that contain "patch" nodes will need to be updated to use meshes instead, as
specifying patches was both inefficient and completely unsupported by the new
subdivision code.
|
|
AMD platform
Reviewers: sergey
Differential Revision: https://developer.blender.org/D2172
|
|
Fixes part of the performance regressions compared to 2.76b, see T49046, T48876.
|
|
|
|
The if branches were reordered when the original patch was
committed, which broke the implicit non-NULL guarantee on link.
To prevent re-occurrence, add a couple of unit tests.
|
|
For best results use the latest 3Dconnexion driver. But latest is only
supported on Mac OS 10.9+. We go all the way back to Mac OS 10.6 so
have to deal with older driver versions.
See the original dlclose line for my faulty assumption. Waiting to
unload the driver later fixes the crash. Newer drivers don’t seem to
have this issue.
Also removed WITH_INPUT_NDOF guards as NDOFManager.h takes care of
this. Follow-up to b10d005 a few days ago.
|
|
For some reason my NVidia linux driver crashes in
glDeleteTextures when the ID is total garbage.
|
|
|
|
When WITH_INPUT_NDOF is disabled, 3D mouse handling code is removed
from:
- GHOST (was mostly done, finished the job)
- window manager
- various editors
- RNA
- keymaps
The input tab of user prefs does not show 3D mouse settings. Key map
editor does not show NDOF mappings.
DNA does not change.
On my Mac the compiled binary is 42KB smaller after this change. It
runs fine WITH_INPUT_NDOF on or off.
|
|
Accidentally left this in the 3D mouse code. Mac only.
|
|
(Now without the build errors)
|
|
This reverts commit 40b367479c6fe23d6f2b6d822f2d5266485619f3.
Didn't build or solve any known issue. Please don't push changes without
testing them first.
|
|
These latter can cause MSVC debug asserts if the array is empty. With C++11
we'll be able to do this for std::vector later. This hopefully fixes an assert
in the Cycles subdivision code.
|
|
|
|
|
|
Kernels can now be built without patch evaluation when not needed by the
scene (Catmull-Clark subdivision not in use), giving a performance boost
for some devices.
|
|
kernel_textures.h is included in device_opencl.cpp, so we can't check
__KERNEL_OPENCL__ there.
|
|
Enable based on --debug-gpu at the command line. Linux already works
this way.
This commit in master (2.78 timeframe) is the smallest change that gets
the desired result. I did a similar commit in blender2.8. Most ongoing
GL debug work will go into 2.8, not 2.7x.
In support of T49089
|
|
After discussion in IRC, 16bit int should not be converted to half, too much precision loss.
|
|
|
|
Note: This only works for textures loaded from disk via OIIO, not packed textures. That's still a ToDo.
|
|
By calling `tessellate()` from the mesh manager in Cycles we can do pre/post
processing or even threaded tessellation without concerning client side code
with the details.
|
|
This way OpenCL devices can also benefit from a smaller memory footprint, when using e.g. bumpmaps (greyscale, 1 channel).
Additional target for my GSoC 2016.
|
|
their own stdosl.h.
|
|
Atm OpenEXR half files benefit from this and will use only 1/2 of the memory now. More space for HDRs!
Part of my GSoC 2016.
|
|
Now we have the 4 component ones first (float4, byte4, half4) followed by the 1 component ones (float, byte, half).
Makes code a bit more consistent and also reduces code a bit when enabling half support on GPU in next commit.
This also exposed a typo in half CPU images for 3D textures, which wasn't used yet, but good to have that one fixed anyway.
|
|
|
|
This was causing Cycles standalone to fail to build from Blender repo.
Hopefully nothing breaks from removing this.
|
|
|
|
structs"
Using unit tests is a wrong way to control static behavior of the
application. They should only be used for checking dynamic behavior,
all the rest is easily controllable at compile time.
Doing tests at ocmpile time are actually more robust approach since
we don't have strict policy of runnign unit tests before accepting
any change.
Proper alignment control is coming shortly.
This reverts commit 7c3a06c34918567e6b0ab67bded60725ff63073b.
|
|
|
|
|
|
|
|
Since the optimal values depend on the device used, this option doesn't make much sense in the XML.
Therefore, it's now specified via the command line, just like the device itself.
|
|
Commit c96ae81160ad added three data textures and therefore removed three image texture slots, but the value in util_textures.h wasn't updated.
|
|
|
|
Since inputs are user controlled, the node can't assume they aren't zero.
|
|
|
|
To recreate, just delete default cube and render.
Simple fix, but @maiself or @sergey may want to double-check ;)
|
|
changes.
|
|
|
|
|
|
|