Blender Git Commits

Viewport: Make lights in local view behave like BI and Cycles

Ignore that lights are not included in local view, always render them if they are on the correct layers. BI and Cycles do this as well.

Fix T49003: Cycles volumes have wrong results after recent microdisp commits

Problem was that sd->prim can be -1 for volumes and was causing check in subd
code to access out of bounds

Fix T43220, T47551: collider scaling or rotation causes smoke to explode.

The problem happens because smoke collides only with the surface of the
collider and uses incompressible fluid solver. This means that scaling
the collider tries to compress or decompress fluid within the volume of
the collider, which can't be handled by the simulation. Fast rotation
likely also causes transient scaling due to emulation of arcs by chords.

This can be fixed by finding compartments completely isolated by obstacles
from the rest of the domain, and forcing total divergence within each one
to be zero so that equations are solvable. Physical validity is somewhat
dubious, but without this the solver simply breaks down.

From the physics point of view, the effect of the correction should be
similar to opening a hole from every cell to another dimension that lets
an equal amount of air to pass through to balance the change in volume.

Reviewers: miikah, lukastoenne

Reviewed By: lukastoenne

Subscribers: dafassi, scorpion81, #physics

Maniphest Tasks: T43220, T47551

Differential Revision: https://developer.blender.org/D2112

Fix T43782: smoke simulation problems with multiple touching colliders.

When the colliders are joined, each cell that touches them only gets
velocity contribution once. When there are multiple objects, velocities
are summed, which causes some cells to get 2x, 3x or more than the actual
object velocity.

Fix this by using the average velocity of all colliders touching a cell.

Reviewers: miikah, lukastoenne

Reviewed By: lukastoenne

Subscribers: dafassi, scorpion81, #physics

Maniphest Tasks: T43782

Differential Revision: https://developer.blender.org/D2112

Cycles: log how many nodes were deduplicated for use in tests.

To make the number more meaningful, also skip deduplicating
obviously unused nodes with no outgoing links.

Cycles: add folding for redundant A to B to A conversions.

As a result of other folding simplifications it may happen that
two type conversion nodes end up directly connected. In some
cases it may be possible to then remove both. A realistic case
might be an optimized out Mix RGB node used to blend vectors.

It seems it's safe to optimize when B is a float3 type
(color, vector), and A is float3 or float.

Reviewers: #cycles, sergey

Reviewed By: #cycles, sergey

Subscribers: sergey

Differential Revision: https://developer.blender.org/D2134

Fix T48986: Incorrect normals or tangents when using custom split normals.

(Re)-setting custom normals could cause some unwanted splitting of some of them, leading
to slightly different tangent space. Simply enlarged slightly the threshold detecting
similar normals as identical ones for now, afarid this is the kind of issue that cannot
get a full complete solution for until we drop floats...

Buildbot: Use annoying hybrid setup of two CUDA toolkits

This is for until we'll solve issues with toolkit 8.0.

Cycles: Some more inline policy tweaks for CUDA 8

Makes it so toolkit does exactly the same decision about what to inline,
but unfortunately it has really barely visible difference on GTX-980.

Cycles: Cleanup, trailing whitespace

Cycles: Some cleanup in CUDA device file

Cycles: Include NVCC compiler flags into md5 hash

This way we can easily switch between toolkits without worrying
whether some kernel was compiled with old or new CUDA toolkit.

It's also now possible to switch machine architecture and have
proper cached kernel detected. Not as if it happens every day,
but i did such a bitness switch back in the days :)

Fix T48998: 'header input' of rotation transform was clamped in [-PI, PI[ range.

Can't see any reason for this behavior (inherited from 2.4x code), so for now just removing it.

Fix T47591: Smoke keeps both .bphys and .vdb cache files

Delete old caches when changing cache type.

Fix T47520: Compositor Node "Filter" Outputs clear image

Make 'Laplace' filter an edge filter operation, since this what it is
typically used for, and such operation does not affect the input's alpha
channel.

Reviewers: sergey, campbellbarton

Reviewed By: sergey

Differential Revision: https://developer.blender.org/D1817

Cycles: tweak some folding tests to avoid unnecessary type conversion.

It's easy to connect a simple value to output without forcing
conversion to color by using the Strength socket of Emission.

Fix tmp screen becoming permanent after going fullscreen

Doubt this is intentional.

Fix T48991: some particles emitted at a distance from object.

The cause seems to be that despite dt_frac being computed as
1/(subframes+1) with integer subframes value, it doesn't always
add up to exactly 1.0 due to precision limitations. If the sum
is similar to 1.00000???, the last subframe is skipped, and all
particles that were supposed to be emitted in that interval are
emitted next frame, with the code working incorrectly due to
skewed time range.

To fix, separate the code from the dynamic timestep feature that
adjusts the last subframe length into a separate function, and
use it even when dynamic timestep is disabled.

Audapsace: Make it work with Jack and no libjack-dev package installed

@neXyon, mind backporting it to your upstream? :)

Buildbot: Enable proper Pascal architecture