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.
Goal is to reduce OpenCL kernel recompilations.
Currently viewport renders are still set to use 64 closures as this seems to
be faster and we don't want to cause a performance regression there. Needs
to be investigated.
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D2775
This is a prequisite for getting host memory allocation to work. There appears
to be no support for 3D textures using host memory. The original version of
this code was written by Stefan Werner for D2056.
* Remove tex_* and pixels_* functions, replace by mem_*.
* Add MEM_TEXTURE and MEM_PIXELS as memory types recognized by devices.
* No longer create device_memory and call mem_* directly, always go
through device_only_memory, device_vector and device_pixels.
This change affects CUDA GPUs not connected to a display or connected to a
display but supporting compute preemption so that the display does not
freeze. I couldn't find an official list, but compute preemption seems to be
only supported with GTX 1070+ and Linux (not GTX 1060- or Windows).
This helps improve small tile rendering performance further if there are
sufficient samples x number of pixels in a single tile to keep the GPU busy.
* Use common TextureInfo struct for all devices, except CUDA fermi.
* Move image sampling code to kernels/*/kernel_*_image.h files.
* Use arrays for data textures on Fermi too, so device_vector<Struct> works.
The work size is still very conservative, and this doesn't help for progressive
refine. For that we will need to render multiple tiles at the same time. But this
should already help for denoising renders that require too much memory with big
tiles, and just generally soften the performance dropoff with small tiles.
Differential Revision: https://developer.blender.org/D2856
Image textures were being packed into a single buffer for OpenCL, which
limited the amount of memory available for images to the size of one
buffer (usually 4gb on AMD hardware). By packing textures into multiple
buffers that limit is removed, while simultaneously reducing the number
of buffers that need to be passed to each kernel.
Benchmarks were within 2%.
Fixes T51554.
Differential Revision: https://developer.blender.org/D2745
Some of the functions might have been inlined, but others i don't see
how that was possible (don't think virtual functions can be inlined here).
In any case, better be explicitly optimal in the code.
The previous outlier heuristic only checked whether the pixel is more than
twice as bright compared to the 75% quantile of the 5x5 neighborhood.
While this detected fireflies robustly, it also incorrectly marked a lot of
legitimate small highlights as outliers and filtered them away.
This commit adds an additional condition for marking a pixel as a firefly:
In addition to being above the reference brightness, the lower end of the
3-sigma confidence interval has to be below it.
Since the lower end approximates how low the true value of the pixel might be,
this test separates pixels that are supposed to be very bright from pixels that
are very bright due to random fireflies.
Also, since there is now a reliable outlier filter as a preprocessing step,
the additional confidence interval test in the reconstruction kernel is no
longer needed.
Seems re-loading module invalidates memory pointers by the looks of it,
which gives an error on the next kernel call.
Not sure how to move memory pointer from one CUDA module to another one,
so for now simply disabling kernel re-load for CUDA devices. Not ideal,
but better than failing render.
Feature-selective option for CUDA is not an official feature anyway.
- Some arguments were inapproriatry tagged as unused
using (void)foo semantic.
Only use such semantic in tricky casses, when something
needs to be ignored in release builds or something is
dependent on tricky ifndef policy.
For rest of the cases just use void foo(int /bar*/)
semantic, which ensures variable is not used. Solves
confusion and code running out of sync with later
development.
- Used proper unused semantic to some arguments.
- Added braces to make code easier to follow, tricky
indentation with ifdef, uh.
Extremely bright pixels in the rendered image cause the denoising algorithm
to produce extremely noticable artifacts. Therefore, a heuristic is needed
to exclude these pixels from the filtering process.
The new approach calculates the 75% percentile of the 5x5 neighborhood of
each pixel and flags the pixel if it is more than twice as bright.
During the reconstruction process, flagged pixels are skipped. Therefore,
they don't cause any problems for neighboring pixels, and the outlier pixels
themselves are replaced by a prediction of their actual value based on their
feature pass values and the neighboring pixels.
Therefore, the denoiser now also works as a smarter despeckling filter that
uses a more accurate prediction of the pixel instead of a simple average.
This can be used even if denoising isn't wanted by setting the denoising
radius to 1.
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
By calculating the size of the state buffer in the kernel rather than the host
less code is needed and the size actually reflects the requested features.
Will also be a little faster in some cases because of larger global work size.
This is to help debug and track memory usage for generic buffers. We
have similar for textures already since those require a name, but for
buffers the name is only for debugging proposes.
