Previously, the NLM kernels would be launched once per offset with one thread per pixel.
However, with the smaller tile sizes that are now feasible, there wasn't enough work to fully occupy GPUs which results in a significant slowdown.
Therefore, the kernels are now launched in a single call that handles all offsets at once.
This has two downsides: Memory accesses to accumulating buffers are now atomic, and more importantly, the temporary memory now has to be allocated for every shift at once, increasing the required memory.
On the other hand, of course, the smaller tiles significantly reduce the size of the memory.
The main bottleneck right now is the construction of the transformation - there is nothing to be parallelized there, one thread per pixel is the maximum.
I tried to parallelize the SVD implementation by storing the matrix in shared memory and launching one block per pixel, but that wasn't really going anywhere.
To make the new code somewhat readable, the handling of rectangular regions was cleaned up a bit and commented, it should be easier to understand what's going on now.
Also, some variables have been renamed to make the difference between buffer width and stride more apparent, in addition to some general style cleanup.
* Remove some unnecessary SSE emulation defines.
* Use full precision float division so we can enable it.
* Add sqrt(), sqr(), fabs(), shuffle variations, mask().
* Optimize reduce_add(), select().
Differential Revision: https://developer.blender.org/D2764
This commit contains the first part of the new Cycles denoising option,
which filters the resulting image using information gathered during rendering
to get rid of noise while preserving visual features as well as possible.
To use the option, enable it in the render layer options. The default settings
fit a wide range of scenes, but the user can tweak individual settings to
control the tradeoff between a noise-free image, image details, and calculation
time.
Note that the denoiser may still change in the future and that some features
are not implemented yet. The most important missing feature is animation
denoising, which uses information from multiple frames at once to produce a
flicker-free and smoother result. These features will be added in the future.
Finally, thanks to all the people who supported this project:
- Google (through the GSoC) and Theory Studios for sponsoring the development
- The authors of the papers I used for implementing the denoiser (more details
on them will be included in the technical docs)
- The other Cycles devs for feedback on the code, especially Sergey for
mentoring the GSoC project and Brecht for the code review!
- And of course the users who helped with testing, reported bugs and things
that could and/or should work better!
The idea is to make include statements more explicit and obvious where the
file is coming from, additionally reducing chance of wrong header being
picked up.
For example, it was not obvious whether bvh.h was refferring to builder
or traversal, whenter node.h is a generic graph node or a shader node
and cases like that.
Surely this might look obvious for the active developers, but after some
time of not touching the code it becomes less obvious where file is coming
from.
This was briefly mentioned in T50824 and seems @brecht is fine with such
explicitness, but need to agree with all active developers before committing
this.
Please note that this patch is lacking changes related on GPU/OpenCL
support. This will be solved if/when we all agree this is a good idea to move
forward.
Reviewers: brecht, lukasstockner97, maiself, nirved, dingto, juicyfruit, swerner
Reviewed By: lukasstockner97, maiself, nirved, dingto
Subscribers: brecht
Differential Revision: https://developer.blender.org/D2586
Use fast-math friendly version of this function.
We should probably avoid unsafe fast math, but this is to be done with
real care with all the benchmarks properly done.
For now comitting much safer fix.
There are following benefits:
- Modifying intersection algorithm will not cause so much re-compilation.
- It works around header dependency hell and allows us to use vectorization
types much easier in there.
I can no longer reproduce crash with neither of the files where
the crash was originally visible. This is something where other
changes (light threshold, sampling) had an effect and made code
to work as it is supposed to. Could have been optimizator issue
or something like that.
Let's see if we hit same issue again.
In scenes with many lights, some of them might have a very small contribution to some pixels, but the shadow rays are traced anyways.
To avoid that, this patch adds probabilistic termination to light samples - if the contribution before checking for shadowing is below a user-defined threshold, the sample will be discarded with probability (1 - (contribution / threshold)) and otherwise kept, but weighted more to remain unbiased.
This is the same approach that's also used in path termination based on length.
Note that the rendering remains unbiased with this option, it just adds a bit of noise - but if the setting is used moderately, the speedup gained easily outweighs the additional noise.
Reviewers: #cycles
Subscribers: sergey, brecht
Differential Revision: https://developer.blender.org/D2217
When using the Normal output of the Texture Coordinate node on Point and Spot lamps, the coordinates now depend on the rotation of the lamp.
On Area lamps, the Parametric output of the Geometry node now returns UV coordinates on the area lamp.
Credit for the Area lamp part goes to Stefan Werner (from D1995).
Oh man, is it a compiler bug? Is it something we do stupid?
For now more crap to prevent crashes. During the conference will talk to
Maxyn about how can we troubleshoot such weird issues.
Basically don't use rcp() in areas which seems to be critical after
second look. Also disabled some multiplication operators, not sure
yet why they might be a problem.
Tomorrow will be setting up a full test with all cases which were
buggy in our farm to see if this fix is complete.
The idea here is to avoid if statements which could cause wrong
branch prediction.
Gives a bit of measurable speedup up to ~1%. Still nice :)
Inspired by Maxym Dmytrychenko, thanks!
This was giving some speedup but made intersection tests to fail
from watertight point of view.
Needs deeper investigation, but need to quickly get it fixed for
the studio.
This gives about 5% speedup on AVX2 kernels (other kernels still
have SSE disabled for math operations) and this solves the slowdown
of koro scene mention in the previous commit.
The title says it all actually. This commit also contains
changes to pass float3 as const reference in affected functions.
This should make MSVC happier without breaking OpenCL because it's
only done in areas which are ifdef-ed for non-OpenCL.
Another patch based on inspiration from Maxym Dmytrychenko, thanks!
All the changes are mainly giving explicit tips on inlining functions,
so they match how inlining worked with previous toolkit.
This make kernel compiled by CUDA 8 render in average with same speed
as previous kernels. Some scenes are somewhat faster, some of them are
somewhat slower. But slowdown is within 1% so far.
On a positive side it allows us to enable newer generation cards on
buildbots (so GTX 10x0 will be officially supported soon).
This adds support for ngons and attributes on subdivision meshes. Ngons are
needed for proper attribute interpolation as well as correct Catmull-Clark
subdivision. Several changes are made to achieve this:
- new primitive `SubdFace` added to `Mesh`
- 3 more textures are used to store info on patches from subd meshes
- Blender export uses loop interface instead of tessface for subd meshes
- `Attribute` class is updated with a simplified way to pass primitive counts
around and to support ngons.
- extra points for ngons are generated for O(1) attribute interpolation
- curves are temporally disabled on subd meshes to avoid various bugs with
implementation
- old unneeded code is removed from `subd/`
- various fixes and improvements
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D2108
This commit adds a new distribution to the Glossy, Anisotropic and Glass BSDFs that implements the
multiple-scattering microfacet model described in the paper "Multiple-Scattering Microfacet BSDFs with the Smith Model".
Essentially, the improvement is that unlike classical GGX, which only models single scattering and assumes
the contribution of multiple bounces to be zero, this new model performs a random walk on the microsurface until
the ray leaves it again, which ensures perfect energy conservation.
In practise, this means that the "darkening problem" - GGX materials becoming darker with increasing
roughness - is solved in a physically correct and efficient way.
The downside of this model is that it has no (known) analytic expression for evalation. However, it can be
evaluated stochastically, and although the correct PDF isn't known either, the properties of MIS and the
balance heuristic guarantee an unbiased result at the cost of slightly higher noise.
Reviewers: dingto, #cycles, brecht
Reviewed By: dingto, #cycles, brecht
Subscribers: bliblubli, ace_dragon, gregzaal, brecht, harvester, dingto, marcog, swerner, jtheninja, Blendify, nutel
Differential Revision: https://developer.blender.org/D2002
The original quad intersection test works by just testing against the two triangles that define the quad.
However, in this case it's actually faster to use the same test that's also used for portals: Determining
the distance to the plane in which the quad lies, calculating the hitpoint and checking whether it's in the
quad by projecting onto the sides.
Reviewers: brecht, sergey, dingto
Reviewed By: dingto
Differential Revision: https://developer.blender.org/D2045
Seems particular CUDA implementations has some precision issues,
which made integer coordinate (which was expected to always be
positive) to go negative.
This patch adds the "Hilbert Spiral", a custom-designed continuous space-filling curve, as a tile order for rendering in Cycles.
It essentially works by dividing the tiles into tile blocks which are processed in a spiral outwards from the center. Inside each
block, the tiles are processed in a regular Hilbert curve pattern. By rotating that pattern according to the spiral direction,
a continuous curve is obtained, which helps with cache coherency and therefore rendering speed.
The curve is a compromise between the faster-rendering Bottom-to-Top etc. orders and the Center order, which is a bit slower,
but starts with the more important areas. The Hilbert Spiral also starts in the center (unless huge tiles are used) and is still
marginally slower than Bottom-to-Top, but noticeably faster than Center.
Reviewers: sergey, #cycles, dingto
Reviewed By: #cycles, dingto
Subscribers: iscream, gregzaal, sergey, mib2berlin
Differential Revision: https://developer.blender.org/D1166
Recent changes to kernel broke compilation of the kernels again, need some
other kind of solution for this issue.
Don't have much time for this currently, but will be addressed before the
release.
Meanwhile it's better to have some buildbot builds instead of totally failing
one.
This more a workaround for CUDA optimizer which can't optimize clamp(x, 0, 1)
into a single instruction and uses 4 instructions instead.
Original patch by @lockal with own modification:
Don't make changes outside of the kernel. They don't make any difference
anyway and term saturate() has a bit different meaning outside of kernel.
This gives around 2% of speedup in Barcelona file, but in more complex shader
setups with lots of math nodes with clamping speedup could be much nicer.
Subscribers: dingto
Projects: #cycles
Differential Revision: https://developer.blender.org/D1224
This inconsistency drove me totally crazy, it's really confusing
when it's inconsistent especially when you work on both Cycles and
Blender sides.
Shouldn;t cause merge PITA, it's whitespace changes only, Git should
be able to merge it nicely.
OpenCL doesn't let you to get address of vector components, which
is kinda annoying. On the other hand, maybe now compiler will have
more chances to optimize something out.
