This matches behavior of Multiscatter GGX and could become handy later on
when/if we decide it would be beneficial to replace on closure with another.
Reviewers: lukasstockner97, brecht
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D2413
Mostly this is making inlining match CUDA 7.5 in a few performance critical
places. The end result is that performance is now better than before, possibly
due to less register spilling or other CUDA 8.0 compiler improvements.
On benchmarks scenes, there are 3% to 35% render time reductions. Stack memory
usage is reduced a little too.
Reviewed By: sergey
Differential Revision: https://developer.blender.org/D2269
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).
In the triangle intersection refinement code, rays that are parallel to the triangle caused a divide by zero.
These rays might initially hit the triangle due to the watertight intersection test, but are very rare - therefore, just skipping the refinement for them works fine.
Also, a few remaining issues in the MultiGGX code are fixed that were caused by rays parallel to the surface (which happened more often there due to smooth shading).
- In fresnel_dielectric, the differentials calculation sometimes divided by zero.
- When the normal map was (0.5, 0.5, 0.5), the code would try to normalize a zero vector. Now, it just uses the regular normal as a fallback.
- The approximate error function used in Beckmann sampling sometimes overflowed to inf while calculating r^16. The final value is 1 - 1/r^16, however,
so now it just returns 1 if the computation would overflow otherwise.
Also, this fixes a numerical issue where A would be inf.
Since later G is set to 1 if A is larger than 1.6, the code now checks the reciprocal of A for being smaller than 1/1.6 - same effect, but no inf involved.
Make sure we don't perform any implicit address space conversion.
A bit annoying, but less intrusive approaches (like using temp private
variable in .cl kernel) do not work correct here.
Using generic address space will help from code side here, but will
be somewhat slower due to extra things happening as far as i know.
As far as I can see, the second issue there was that the functions receive a pointer to a member variable of the
ShaderData, which is stored in global memory. However, this means that the pointer points to global memory as well,
therefore OpenCL requires the ccl_addr_space "keyword" in front of the pointer.
With this commit, the OpenCL kernels build on Linux with the Intel CPU OpenCL runtime - however, they already did
without the change and I don't have an AMD card, so I can't really test whether the AMD runtime is happy as well now.
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
There are several fixes in here, which hopefully will make the shader
working correct without too much magic in there.
First of all, this commit brings BURLEY_TRUNCATE down from 30 to 16
which reduces noise a lot. It's still higher than original truncate
from Brecht, but this reduces PDF value at a cutoff distance by an
order of magnitude (now it's 0.008387, previously it was 0.063521
for the albedo of 0.8 and radius 1.0). This should converge to a
proper result faster and don't have artifacts.
This kind of reverts fix for T47356, but after additional thinking
came to conclusion Burley is not being totally smooth, it is about
giving less waxy results which it's kind of doing in the file.
Second of all, this commit fixes burley_eval() to use normalized
diffusion reflectance. This matches the way we calculate CDF and
solves numeric instability close to 0, making PDF profile looking
closer to other SSS profiles:
https://developer.blender.org/F282355https://developer.blender.org/F282356https://developer.blender.org/F282357
Reviewers: brecht
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D1792
After the clamping commit we need to bump BURLEY_TRUNCATE
constant a bit, otherwise mean free path does not really
match the disk radius needed for importance sampling.
The value was too high, causing bad Newton iteration step.
Now the value is not so good, but it's still within 9 iterations
and those high number of iterations are only happening in
approx 1% of input values.
The idea is simply to pre-compute fitting and parameterization
in the bssrdf_setup() function and re-use the values in both
sample() and eval().
The only trick is where to store the pre-calculated values and
the answer is inside of ShaderClosure->custom{1,2,3}. There's
no memory bump here because we now simply re-use padding fields
for the pre-calculated values. Similar trick we can do for other
BSDFs.
Seems to give nice speedup up to 7% here on my desktop with
Core i7 CPU, SSE4.1 kernel.
This is an alternate fix for T40964 which resolves bad handling of
caustics reported in T45609.
There were too much transmission rays being discarded by the original
fix, which caused by caustic light being totally disabled. There is
still some room for investigation why exactly original paper didn't
work that well, could be caused by the way how the pdf is calculated.
In any case current results seems rather correct now.
This simplification is safe, as the call to volume_phase_eval() is guarded behind a CLOSURE_IS_PHASE check, which is equal to
CLOSURE_VOLUME_HENYEY_GREENSTEIN_ID. I don't think we will add more phase functions anytime soon, if at all.
This commit contains all the work related on the AMD megakernel split work
which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus
some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely
someone else which we're forgetting to mention.
Currently only AMD cards are enabled for the new split kernel, but it is
possible to force split opencl kernel to be used by setting the following
environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1.
Not all the features are supported yet, and that being said no motion blur,
camera blur, SSS and volumetrics for now. Also transparent shadows are
disabled on AMD device because of some compiler bug.
This kernel is also only implements regular path tracing and supporting
branched one will take a bit. Branched path tracing is exposed to the
interface still, which is a bit misleading and will be hidden there soon.
More feature will be enabled once they're ported to the split kernel and
tested.
Neither regular CPU nor CUDA has any difference, they're generating the
same exact code, which means no regressions/improvements there.
Based on the research paper:
https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf
Here's the documentation:
https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit
Design discussion of the patch:
https://developer.blender.org/T44197
Differential Revision: https://developer.blender.org/D1200
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
Our own implementation is in fact the same performance as in fast_math from
OpenShadingLanguage, but implementation from fast_math is using explicit madd
function, which increases chance of compiler deciding to use intrinsics.
This patch is based on some work done in D788 and re-formulation from Beckmann
implementation in OpenShadingLanguage.
Skipping texture lookup helps a lot on GPUs where it's more expensive to access
texture memory than to do some extra calculation in threads.
CPU code still uses lookup-table based approach since this seems to be still
faster (at least on computers i've got access to).
This change gives about 2% speedup on BMW scene with GTX560TI.
It did not preserve stratification too well and lookup-table approach was
working much better. There are now also some more interesting forumlation
from Wenzel and OpenShadingLanguage which should work better than old code.
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.
Issue was introduced in 01ee21f where i didn't notice *_setup()
function only doing partial initialization, and some of parameters
are expected to be initialized by callee function.
This was hitting only some setups, so tests with benchmark scenes
didn't unleash issues. Now it should all be fine.
This is to go to the 2.74 branch and we actually might re-AHOY.
Fix T44007: Cycles Volumetrics: block artifacts with overlapping volumes
The issue was caused by uninitialized parameters of some closures, which
lead to unpredictable behavior of shader_merge_closures().
