Only happened when rendering second time with OSL enabled.
Was caused by fix for T47201. Hopefully now all the issues
are solved (meaning, we don't keep files opened and don't
crash either).
This way we avoid passing structures which could be up to
few hundred bytes by value to the utility functions.
Ideally we'll also have to add `const` qualifier in majority
of the calls, but C++ RNA does not allow us to do that because
it does not know if some function modifies contents or not.
The idea is to switch from allocating separate buffers for shader data's
structure of arrays to allocating one huge memory block and do some index
trickery to make it accessed as SOA.
This saves quite reasonable amount of lines of code in device_opencl and
also makes it possible to get rid of special declaration of ShaderData
structure.
As a side effect it also makes it easier to experiment with SOA vs. AOS
for split kernel.
Works fine here on NVidia GTX580, Intel CPU amd AMD Fiji cards.
Reviewers: #cycles, brecht, juicyfruit, dingto
Differential Revision: https://developer.blender.org/D1593
Seems i was the only one who was really up to using it and
i do have gcc-5 finally backported and installed here so
such a fine-tune flags are no longer needed.
It was mainly unfinished code for volume in a split kernel which
should be done differently anyway to avoid such a code copy-paste.
The code didn't really work, so likely nobody will cry.
Use KernelGlobals to access all the global arrays for the intermediate
storage instead of passing all this storage things explicitly.
Tested here with Intel OpenCL, NVIDIA GTX580 and AMD Fiji, didn't see
any artifacts, so guess it's all good.
Reviewers: juicyfruit, dingto, lukasstockner97
Differential Revision: https://developer.blender.org/D1736
The issue was caused by different AABB used by Cycles and texture sampler.
Instead of trying to keep this two functions in sync we now do have an
utility call in the point density node to query the AABB.
This changes the following defaults:
- Render settings:
* Samples: 100
* Preview Samples: 50
* Filter: Blackmann-Harris
* Tile Order: Hilbert Spiral
- Lamp settings:
* Use MIS: On
- Material settings:
* Volume Sampling: Multiple Importance
Old files are not affected, I tested the versioning code back and forth.
More changes are to come (World, BVH...) but that needs a bit more work.
Fix T47213.
There was actually no real bug here, just clarify now in the UI that Mesh, World and Lamp samples only have an effect if we sample all lights (direct or indirect).
The issue was discontinuity in logic when importing vertices from blender
and then importing data layers regardless of how we split the face. Quite
interesting we didn't notice this issue before.
Thanks Bastien for the investigation, based on D1742 but redid it to make
patch a bit more clear to follow.
There is no function pointers in OpenCL specification. For as long
as we want to support this platform we should follow the specifications.
While the code is not totally optimal now, it should not be that huge
of performance issue on CPU since it does jump tables just nicely, so
it's not that much extra computation here.
Now image will be opened for while render session is active, this is
needed to keep image cache working correctly. But stopping render
should now release all files descriptors.
Displacement shader was not updating motion vertex positions.
Current solution is not totally correct because it applies same offset
for all time steps. Ideally we'll need to evaluate displacement shader
for every time offset separately, but currently we don't have subframe
image access.
For the time being will consider this a TODO.
Compiling OSL scripts with errors in them would cause Blender to crash since the OSL version
bump to 1.6.9 instead of printing the error to the console as it did before.
With version 1.6.2, OSL added a pointer to an OpenImageIO ErrorHandler as an argument to the
OSLCompiler constructor. However, since it defaults to the NULL pointer, Blender still compiled
fine after the OSL version bump.
It turns out, though, that this pointer is used without further checks inside the OSL code, which
makes it crash when it tries to report an error unless a valid ErrorHandler pointer is specified.
Therefore, this commit simply passes a pointer to the static default handler that OIIO offers,
which prints the error to the console just like OSL did before.
Using this feature for a more advanced error handling and displaying from the Blender side would
be possible and seems reasonable, but for now it's not really relevant for fixing this bug.
The combined pass is built with the contributions the user finds fit.
It is useful for lightmap baking, as well as non-view dependent effects
baking.
The manual will be updated once we get closer to the 2.77 release.
Meanwhile the new page can be found here:
http://dalaifelinto.com/blender-manual/render/cycles/baking.html
Reviewers: sergey, brecht
Differential Revision: https://developer.blender.org/D1674
This commit removes the experimental CUDA kernel, making SSS and CMJ
regular features.
Several improvements have been made in the past few
weeks (thanks Sergey!) which make SSS render several times faster (2-3x
compared to 2.76b) on the GPU, and the increased VRAM usage has also been
fixed. Therefore the experimental kernel is no longer needed.
Differential Revision: https://developer.blender.org/D1726
Manual has been updated: too:
https://www.blender.org/manual/render/cycles/features.html
The goal is to make Experimental kernel closer in performance to the
official kernel, avoiding spills and such.
There should not be big impact on official kernel, own tests showed
few percent performance drop on laptop's GPU. CPU was always the
same speed on AVX, AVX2 and SSE4.1 CPUs i've been testing here.
This seems to be the last essential step before we can get rid of
Experimental kernel and enable SSS officially on GPU without causing
some major performance issues.
Surely some more tweaks are possibly required, but that we can do
for until cows go home anyway.
Should be no functional changes at all, just speeds up re-compilation
when some features needs to be disabled for development purposes.
For example, when running lots of Valgrind it's handy to disable any
GPU devices because otherwise you'll be wasting quite some time in
the driver while enumerating devices.
Reviewers: dingto, lukasstockner97, brecht, juicyfruit
Differential Revision: https://developer.blender.org/D1730
Previously several areas were calling TEST_SHARED_PTR_SUPPORT and
TEST_UNORDERED_MAP_SUPPORT which isn't that bad on it's own but
was causing some quite verbose output with same information line
printed multiple times. additionally, what's more worse, define flags
for Ceres were duplicated in main CMakeLists and Ceres's CMakeLists.
Now we've got a single place where checks for those classes are
happening and other areas are simply checking for variables set by
those check macros, keeping CMake output clean and nice.
The main purpose of such linking is to make Blender compatible with
NVidia's debuggers and profilers which are doing some LD_PRELOAD
magic to intercept some function calls. Such magic conflicts with
our CUDA wrangler magic and causes segmentation faults.
The option is disabled by default, so there's no affect on any of
artists.
In order to make Blender linked directly against CUDA library use
the WITH_CUDA_DYNLOAD CMake option (it's marked as advanced).
This panel is only visible when debug_value is set to 256 and has no
affect in other cases. However, if debug value is not set to this
value, environment variables will be used to control which features
are enabled, so there's no visible changes to anyone in fact.
There are some changes needed to prevent devices re-enumeration on
every Cycles session create.
Reviewers: juicyfruit, lukasstockner97, dingto, brecht
Reviewed By: lukasstockner97, dingto
Differential Revision: https://developer.blender.org/D1720
Although the code made it impossible to use time_start_ uninitialized, at least GCC did
still produce multiple warnings about it.
Since time_dt() is an extremely cheap operation and functionality does not change in any way when
removing the check in the constructor, this commit removes the check and therefore the warning.
