This improves stability and convergence speed of Workbench Temporal AntiAliasing.
This adds a filtering kernel (blackmann-haris, same as EEVEE/Cycles) to the
temporal antialiasing sampling. We also gather neighbor pixels since they might
end up in the pixel footprint.
We use a 1px radius for the filter window which is a bit less than the 1.5 default
of cycles and EEVEE since it does blur quite a bit more than what we have now.
Another improvement is that the filtering is now in log space which improves
AntiAliasing around highlights.
Theses improvement may not be very useful for every day case but it was an
experiment to try to make TAA usable for GPencil.
Test file used :
{F9798807}
|filtered+logspace|filtered|original|
|{F9798847}|{F9798848}|{F9798849}|
Reviewed By: jbakker
Differential Revision: https://developer.blender.org/D10414
This is a complete refactor over the old system. The goal was to increase quality
first and then have something more flexible and optimised.
|{F9603145} | {F9603142}|{F9603147}|
This fixes issues we had with the old system which were:
- Too much overdraw (low performance).
- Not enough precision in render targets (hugly color banding/drifting).
- Poor resolution near in-focus regions.
- Wrong support of orthographic views.
- Missing alpha support in viewport.
- Missing bokeh shape inversion on foreground field.
- Issues on some GPUs. (see T72489) (But I'm sure this one will have other issues as well heh...)
- Fix T81092
I chose Unreal's Diaphragm DOF as a reference / goal implementation.
It is well described in the presentation "A Life of a Bokeh" by Guillaume Abadie.
You can check about it here https://epicgames.ent.box.com/s/s86j70iamxvsuu6j35pilypficznec04
Along side the main implementation we provide a way to increase the quality by jittering the
camera position for each sample (the ones specified under the Sampling tab).
The jittering is dividing the actual post processing dof radius so that it fills the undersampling.
The user can still add more overblur to have a noiseless image, but reducing bokeh shape sharpness.
Effect of overblur (left without, right with):
| {F9603122} | {F9603123}|
The actual implementation differs a bit:
- Foreground gather implementation uses the same "ring binning" accumulator as background
but uses a custom occlusion method. This gives the problem of inflating the foreground elements
when they are over background or in-focus regions.
This is was a hard decision but this was preferable to the other method that was giving poor
opacity masks for foreground and had other more noticeable issues. Do note it is possible
to improve this part in the future if a better alternative is found.
- Use occlusion texture for foreground. Presentation says it wasn't really needed for them.
- The TAA stabilisation pass is replace by a simple neighborhood clamping at the reduce copy
stage for simplicity.
- We don't do a brute-force in-focus separate gather pass. Instead we just do the brute force
pass during resolve. Using the separate pass could be a future optimization if needed but
might give less precise results.
- We don't use compute shaders at all so shader branching might not be optimal. But performance
is still way better than our previous implementation.
- We mainly rely on density change to fix all undersampling issues even for foreground (which
is something the reference implementation is not doing strangely).
Remaining issues (not considered blocking for me):
- Slight defocus stability: Due to slight defocus bruteforce gather using the bare scene color,
highlights are dilated and make convergence quite slow or imposible when using jittered DOF
(or gives )
- ~~Slight defocus inflating: There seems to be a 1px inflation discontinuity of the slight focus
convolution compared to the half resolution. This is not really noticeable if using jittered
camera.~~ Fixed
- Foreground occlusion approximation is a bit glitchy and gives incorrect result if the
a defocus foreground element overlaps a farther foreground element. Note that this is easily
mitigated using the jittered camera position.
|{F9603114}|{F9603115}|{F9603116}|
- Foreground is inflating, not revealing background. However this avoids some other bugs too
as discussed previously. Also mitigated with jittered camera position.
|{F9603130}|{F9603129}|
- Sensor vertical fit is still broken (does not match cycles).
- Scattred bokeh shapes can be a bit strange at polygon vertices. This is due to the distance field
stored in the Bokeh LUT which is not rounded at the edges. This is barely noticeable if the
shape does not rotate.
- ~~Sampling pattern of the jittered camera position is suboptimal. Could try something like hammersley
or poisson disc distribution.~~Used hexaweb sampling pattern which is not random but has better
stability and overall coverage.
- Very large bokeh (> 300 px) can exhibit undersampling artifact in gather pass and quite a bit of
bleeding. But at this size it is preferable to use jittered camera position.
Codewise the changes are pretty much self contained and each pass are well documented.
However the whole pipeline is quite complex to understand from bird's-eye view.
Notes:
- There is the possibility of using arbitrary bokeh texture with this implementation.
However implementation is a bit involved.
- Gathering max sample count is hardcoded to avoid to deal with shader variations. The actual
max sample count is already quite high but samples are not evenly distributed due to the
ring binning method.
- While this implementation does not need 32bit/channel textures to render correctly it does use
many other textures so actual VRAM usage is higher than previous method for viewport but less
for render. Textures are reused to avoid many allocations.
- Bokeh LUT computation is fast and done for each redraw because it can be animated. Also the
texture can be shared with other viewport with different camera settings.
The output layout was wrong and it's a mistery why it works on most
implementations since it's clearly a wrong usage.
Thanks @sebbas for helping narrowing down the issue.
All the changes made in the branch `soc-2020-fluid-tools` are included in this patch.
**Major changes:**
=== Viewport Display ===
- //Raw voxel display// or //closest (nearest-neighbor)// interpolation for displaying the underlying voxel data of the simulation grids more clearly.
- An option to display //gridlines// when the slicing method is //single//.
==== Grid Display ====
- Visualization for flags, pressure and level-set representation grids with a fixed color coding based on Manta GUI.
==== Vector Display ====
- //**M**arker **A**nd **C**ell// grid visualization options for vector grids like velocity or external forces.
- Made vector display options available for external forces.
==== Coloring options for //gridlines// ====
- Range highlighting and cell filtering options for displaying the simulation grid data more precisely.
- Color gridlines with flags.
- Also, made slicing and interpolation options available for Volume Object.
Reviewed By: JacquesLucke, sebbas
Differential Revision: https://developer.blender.org/D8705
On MacOS + Intel Iris Graphics 6100 (may affect other config too), the
texelFetch operation bypass the base mip setting of the texture object.
Using textureLod with lod = 0.0 ensure the lowest (after clamping) mip will
always be selected.
Also disable the texture filtering for this sampler to avoid unecessary
fetches.
This should fix T78653 Blender 2.83 broken Depth of Field in Viewport
- add the use of DRWShaderLibrary to EEVEE's glsl codebase to reduce code
complexity and duplication.
- split bsdf_common_lib.glsl into multiple sub library which are now shared
with other engines.
- the surface shader code is now more organised and have its own files.
- change default world to use a material nodetree and make lookdev shader
more clear.
Reviewed By: jbakker
Differential Revision: https://developer.blender.org/D8306
Also includes outline overlays. Removes the temp overlay drawing
We make the geometry follow camera like billboards this uses less
geometry. Currently we use half octahedron for now. Goal would be
to use icospheres.
This patch also optimize the case when pointcloud has uniform radius.
However we should premultiply the radius prop by the default radius
beforehand to avoid a multiplication on CPU.
Using geometry instead of pseudo raytraced spheres is more scalable as
we can render as low as 1 or 2 triangle to a full half sphere and can
integrate easily in the render pipeline using a low amount of code.
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D8301
This removes some fragment shader hacks and improve the support of
different repeat & filtering modes.
This fix T77453 Image texture not repeating in viewport
All textures in workbench are using linear interpolation. The fragment
shader modifies the uv coordinates to sample always in the center of a
texel. In rare conditions the GPU could sample an incorrect value due to
rounding errors making some rendering artifacts.
This patch skips the interpolation in the fragment shader to remove
these render artifacts.
Reviewed By: Clément Foucault
Differential Revision: https://developer.blender.org/D7816
Only the volume drawing part is really finished and exposed to the user. Hair
plugs into the existing hair rendering code and is fairly straightforward. The
pointcloud drawing is a hack using overlays rather than Eevee and workbench.
The most tricky part for volume rendering is the case where each volume grid
has a different transform, which requires an additional matrix in the shader
and non-trivial logic in Eevee volume drawing. In the common case were all the
transforms match we don't use the additional per-grid matrix in the shader.
Ref T73201, T68981
Differential Revision: https://developer.blender.org/D6955
This patch is (almost) a complete rewrite of workbench engine.
The features remain unchanged but the code quality is greatly improved.
Hair shading is brighter but also more correct.
This also introduce the concept of `DRWShaderLibrary` to make a simple
include system inside the GLSL files.
Differential Revision: https://developer.blender.org/D7060
This is more in line with standard grids and means we don't have to make
many special exceptions in the upcoming change for arbitrary number of volume
grids support in Eevee.
The workbench shader was also changed to fix bugs where squared density was
used, and the smoke color would affect the density so that black smoke would
be invisible. This can change the look of smoke in workbench significantly.
When using the color grid when smoke has a constant color, the color grid
will no longer be premultiplied by the density. If the color is constant
we want to be able not to store a grid at all. This breaks one test for
Cycles and Eevee, but the setup in that test using a color without density
does not make sense. It suffers from artifacts since the unpremultiplied
color grid by itself will not have smooth boundaries.
Differential Revision: https://developer.blender.org/D6951
This way we remove the need for the srgb boolean uniform and a lot of code complexity. However, mesh update is going to be a bit slower.
I did not benchmark the performance impact.
This also fix a typo in draw_cache_impl_particles.c and fix hair not using vertex color in workbench.
Reviewed By: jbakker
Differential Revision: https://developer.blender.org/D6610
Based on @fclem's suggestion in D6421, this commit implements support for
storing all tiles of a UDIM texture in a single 2D array texture on the GPU.
Previously, Eevee was binding one OpenGL texture per tile, quickly running
into hardware limits with nontrivial UDIM texture sets.
Workbench meanwhile had no UDIM support at all, as reusing the per-tile
approach would require splitting the mesh by tile as well as texture.
With this commit, both Workbench as well as Eevee now support huge numbers
of tiles, with the eventual limits being GPU memory and ultimately
GL_MAX_ARRAY_TEXTURE_LAYERS, which tends to be in the 1000s on modern GPUs.
Initially my plan was to have one array texture per unique size, but managing
the different textures and keeping everything consistent ended up being way
too complex.
Therefore, we now use a simpler version that allocates a texture that
is large enough to fit the largest tile and then packs all tiles into as many
layers as necessary.
As a result, each UDIM texture only binds two textures (one for the actual
images, one for metadata) regardless of how many tiles are used.
Note that this rolls back per-tile GPUTextures, meaning that we again have
per-Image GPUTextures like we did before the original UDIM commit,
but now with four instead of two types.
Reviewed By: fclem
Differential Revision: https://developer.blender.org/D6456
This is the unification of all overlays into one overlay engine as described in T65347.
I went over all the code making it more future proof with less hacks and removing old / not relevent parts.
Goals / Acheivements:
- Remove internal shader usage (only drw shaders)
- Remove viewportSize and viewportSizeInv and put them in gloabl ubo
- Fixed some drawing issues: Missing probe option and Missing Alt+B clipping of some shader
- Remove old (legacy) shaders dependancy (not using view UBO).
- Less shader variation (less compilation time at first load and less patching needed for vulkan)
- removed some geom shaders when I could
- Remove static e_data (except shaders storage where it is OK)
- Clear the way to fix some anoying limitations (dithered transparency, background image compositing etc...)
- Wireframe drawing now uses the same batching capabilities as workbench & eevee (indirect drawing).
- Reduced complexity, removed ~3000 Lines of code in draw (also removed a lot of unused shader in GPU).
- Post AA to avoid complexity and cost of MSAA.
Remaining issues:
- ~~Armature edits, overlay toggles, (... others?) are not refreshing viewport after AA is complete~~
- FXAA is not the best for wires, maybe investigate SMAA
- Maybe do something more temporally stable for AA.
- ~~Paint overlays are not working with AA.~~
- ~~infront objects are difficult to select.~~
- ~~the infront wires sometimes goes through they solid counterpart (missing clear maybe?) (toggle overlays on-off when using infront+wireframe overlay in solid shading)~~
Note: I made some decision to change slightly the appearance of some objects to simplify their drawing. Namely the empty arrows end (which is now hollow/wire) and distance points of the cameras/spots being done by lines.
Reviewed By: jbakker
Differential Revision: https://developer.blender.org/D6296
Background dithering was introduced to solve banding issues on gradient backgrounds.
This patch will enable dithering based on the texture that is used for drawing.
Only when using a GPU_RGBA8 texture the dithering will be enabled.
This disables dithering for final rendering, vertex and texture paint modes.
Reviewed By: fclem
Differential Revision: https://developer.blender.org/D6056
With Blender 2.80 we introduced a more flexible matcap system. One
change we did was to multiply the matcap with the base color that was
shaded. As matcaps contains diffuse and specular lighting in a single
texture this lead to rendering artifacts. Artists were complaining that
everything looked to metalic.
We now support a separate `diffuse` and `specular` pass for matcaps.
`shaded_color = diffuse_light * base_color + specular_light`
For matcaps to support this feature they need to be multilayer openexr
files with 2 renderpasses (named `diffuse` and `specular`). In the future
we can change this to first pass/second pass in stead of this naming
convention.
Reviewed By: fclem, brecht
Differential Revision: https://developer.blender.org/D5335
