There should be no user visible change from this, except that tile size
now affects performance. The goal here is to simplify bake denoising in
D3099, letting it reuse more denoising tiles and pass code.
A lot of code is now shared with regular rendering, with the two main
differences being that we read some render result passes from the bake API
when starting to render a tile, and call the bake kernel instead of the
path trace kernel.
With this kind of design where Cycles asks for tiles from the bake API,
it should eventually be easier to reduce memory usage, show tiles as
they are baked, or bake multiple passes at once, though there's still
quite some work needed for that.
Reviewers: #cycles
Subscribers: monio, wmatyjewicz, lukasstockner97, michaelknubben
Differential Revision: https://developer.blender.org/D3108
This feature takes some inspiration from
"RenderMan: An Advanced Path Tracing Architecture for Movie Rendering" and
"A Hierarchical Automatic Stopping Condition for Monte Carlo Global Illumination"
The basic principle is as follows:
While samples are being added to a pixel, the adaptive sampler writes half
of the samples to a separate buffer. This gives it two separate estimates
of the same pixel, and by comparing their difference it estimates convergence.
Once convergence drops below a given threshold, the pixel is considered done.
When a pixel has not converged yet and needs more samples than the minimum,
its immediate neighbors are also set to take more samples. This is done in order
to more reliably detect sharp features such as caustics. A 3x3 box filter that
is run periodically over the tile buffer is used for that purpose.
After a tile has finished rendering, the values of all passes are scaled as if
they were rendered with the full number of samples. This way, any code operating
on these buffers, for example the denoiser, does not need to be changed for
per-pixel sample counts.
Reviewed By: brecht, #cycles
Differential Revision: https://developer.blender.org/D4686
This is the internal implementation, not available from the API or
interface yet. The algorithm takes into account past and future frames,
both to get more coherent animation and reduce noise.
Ref D3889.
Prefiltering of feature passes will happen during rendering, which can
then be used for denoising immediately or written as a render pass for
later (animation) denoising.
The number of denoising data passes written is reduced because of this,
leaving out the feature variance passes. The passes are now Normal,
Albedo, Depth, Shadowing, Variance and Intensity.
Ref D3889.
Textures in 16 bit integer format are sometimes used for displacement, bump and normal maps and can be exported by tools like Substance Painter. Without this patch, Cycles would promote those textures to single precision floating point, causing them to take up twice as much memory as needed.
Reviewers: #cycles, brecht, sergey
Reviewed By: #cycles, brecht, sergey
Subscribers: sergey, dingto, #cycles
Tags: #cycles
Differential Revision: https://developer.blender.org/D3523
I've limited it to just the RGB<->XYZ stuff for now, correct image handling is the next step.
Reviewers: brecht, sergey
Differential Revision: https://developer.blender.org/D3478
This commit contains the minimum to make clang build/work with blender, asan and ninja build support is forthcoming
Things to note:
1) Builds and runs, and is able to pass all tests (except for the freestyle_stroke_material.blend test which was broken at that time for all platforms by the looks of it)
2) It's slightly faster than msvc when using cycles. (time in seconds, on an i7-3370)
victor_cpu
msvc:3099.51
clang:2796.43
pavillon_barcelona_cpu
msvc:1872.05
clang:1827.72
koro_cpu
msvc:1097.58
clang:1006.51
fishy_cat_cpu
msvc:815.37
clang:722.2
classroom_cpu
msvc:1705.39
clang:1575.43
bmw27_cpu
msvc:552.38
clang:561.53
barbershop_interior_cpu
msvc:2134.93
clang:1922.33
3) clang on windows uses a drop in replacement for the Microsoft cl.exe (takes some of the Microsoft parameters, but not all, and takes some of the clang parameters but not all) and uses ms headers + libraries + linker, so you still need visual studio installed and will use our existing vc14 svn libs.
4) X64 only currently, X86 builds but crashes on startup.
5) Tested with llvm/clang 6.0.0
6) Requires visual studio integration, available at https://github.com/LazyDodo/llvm-vs2017-integration
7) The Microsoft compiler spawns a few copies of cl in parallel to get faster build times, clang doesn't, so the build time is 3-4x slower than with msvc.
8) No openmp support yet. Have not looked at this much, the binary distribution of clang doesn't seem to include it on windows.
9) No ASAN support yet, some of the sanitizers can be made to work, but it was decided to leave support out of this commit.
Reviewers: campbellbarton
Differential Revision: https://developer.blender.org/D3304
Our own implementation was behaving different comparing to OSL and GPU,
namely on the border pixels OSL and CUDA was doing interpolation with
black, but we were clamping coordinate.
This partially fixes issue reported in T53452.
Similar change should also be done for 3D interpolation perhaps, but this
is to be investigated separately.
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.
* 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.
We don't enable global SSE optimizations in regular kernel, and we
keep those disabled on Linux 32bit.
One possible workaround would be to pass arguments by ccl_ref, but
that is quite a few of code which better be done accurately.
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.
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!
Reduce thread divergence in kernel_shader_eval.
Rays are sorted in blocks of 2048 according to shader->id.
On R9 290 Classroom is ~30% faster, and Pabellon Barcelone is ~8% faster.
No sorting for CUDA split kernel.
Reviewers: sergey, maiself
Reviewed By: maiself
Differential Revision: https://developer.blender.org/D2598
This implements branched path tracing for the split kernel.
General approach is to store the ray state at a branch point, trace the
branched ray as normal, then restore the state as necessary before iterating
to the next part of the path. A state machine is used to advance the indirect
loop state, which avoids the need to add any new kernels. Each iteration the
state machine recreates as much state as possible from the stored ray to keep
overall storage down.
Its kind of hard to keep all the different integration loops in sync, so this
needs lots of testing to make sure everything is working correctly. We should
probably start trying to deduplicate the integration loops more now.
Nonbranched BMW is ~2% slower, while classroom is ~2% faster, other scenes
could use more testing still.
Reviewers: sergey, nirved
Reviewed By: nirved
Subscribers: Blendify, bliblubli
Differential Revision: https://developer.blender.org/D2611
This patch allows for an unlimited number of textures in Cycles where the hardware allows. It replaces a number static arrays with dynamic arrays and changes the way the flat_slot indices are calculated. Eventually, I'd like to get to a point where there are only flat slots left and textures off all kinds are stored in a single array.
Note that the arrays in DeviceScene are changed from containing device_vector<T> objects to device_vector<T>* pointers. Ideally, I'd like to store objects, but dynamic resizing of a std:vector in pre-C++11 calls the copy constructor, which for a good reason is not implemented for device_vector. Once we require C++11 for Cycles builds, we can implement a move constructor for device_vector and store objects again.
The limits for CUDA Fermi hardware still apply.
Reviewers: tod_baudais, InsigMathK, dingto, #cycles
Reviewed By: dingto, #cycles
Subscribers: dingto, smellslikedonkey
Differential Revision: https://developer.blender.org/D2650
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
Reduces memory allocation for split kernel.
This allows for faster rendering due to bigger global size,
specially when GPU memory is limited.
Perfromance results:
R9 290 total render time
Before After Change
BMW 4:37 4:34 -1.1 %
Classroom 14:43 14:30 -1.5 %
Fishy Cat 11:20 11:04 -2.4 %
Koro 12:11 12:04 -1.0 %
Pabellon Barcelona 22:01 20:44 -5.8 %
Pabellon Barcelona(*) 15:32 15:09 -2.5 %
(*) without glossy connected to volume
Decoupled ray marching is not supported yet.
Transparent shadows are always enabled for volume rendering.
Changes in kernel/bvh and kernel/geom are from Sergey.
This simiplifies code significantly, and prepares it for
record-all transparent shadow function in split kernel.
This was only needed for the previous implementation of parallel samples. As
we don't have that any more it can be removed.
Real reason for removal tho is this: `per_sample_output_buffers` was being
calculated too small and artifacts resulted. The tile buffer is already
the correct size and calculating the size for `per_sample_output_buffers`
is a bit difficult with the current layout of the code. As
`per_sample_output_buffers` was only needed for `sum_all_radiance`,
removing that kernel and writing output to the tile buffer directly
fixes the artifacts.
