It's not obvious that they use the mouse coordinate of the initial key-press event (behavior since rBf1f33ba7be2d), so added comment.
Also corrected other comments.
OpenCL seems to work fine here, and for some reason that comparison was
giving compilation error on OpenCL here.
Better to compile OpenCL kernel than to be fully robust to weird corner
cases.
Newly computed custom normals were forgotten during poly flipping, leading
to wrong custom normals being assigned to wrong loop...
Dead simple, but was tough to track down this one!
At the moment light shading in Blender is produced in viewspace. Apparently, that's why
shader nodes work with normals in camera space. But it is not convenient for artists.
The more convenient approach is implemented in Cycles where normals are represented in world space.
Blend4Web Team designed the engine keeping in mind shader parameters readability,
so normals are interpreted in world space as well. And now our users have to use some tweaks, like
empty node group with the name "Replace", which is replacing one input by another on the engine side
(replacing working configuration in Blender Viewport by the configuration that has the same behavior in the engine).
This patch adds the ability to switch to world space for normals and lamp vector in BI and Viewport.
This patch is very important to us and we crave to see this patch in Blender 2.7 because
it will significantly simplify Blend4Web material creation workflow.
{F315547}
{F315548}
Reviewers: campbellbarton, brecht
Reviewed By: brecht
Subscribers: homyachetser, Evgeny_Rodygin, AlexKowel, yurikovelenov
Differential Revision: https://developer.blender.org/D2046
Was using O(n^2) lookup on ID's with undo.
This caused undo to hang with 1000's of data-blocks
(especially with heavy scenes & outliner-space, which doesn't even need to be visible to cause a slow-down).
Internally this uses a ghash per id-type, which is lazy-initialized.
Each key uses the name and library since there may be name collisions between libraries.
Developer Notes:
- Adds small `BKE_main_idmap_*` API.
- Needed to change linking order for this to build.
The original quad intersection test works by just testing against the two triangles that define the quad.
However, in this case it's actually faster to use the same test that's also used for portals: Determining
the distance to the plane in which the quad lies, calculating the hitpoint and checking whether it's in the
quad by projecting onto the sides.
Reviewers: brecht, sergey, dingto
Reviewed By: dingto
Differential Revision: https://developer.blender.org/D2045
Being granular means we need to re-build depsgraph a bit more often..
The issue was caused by rigidbody requiring some special nodes to
handle physics which were not created with just tagging object for
update.
The issue was caused by non-watertight nature of intersection, which is now addressed.
Hopefully it doesn't cause any regression caused by uninitialized precalculated storage.
Makes behavior of proxy_from backlink working similar to the old dependency graph.
it's nasty, but needed here in the studio to get proxies fixes ASAP.
In certain types of animation it's possible to have some objects
scaling to zero. In this case we can save render times by avoid
traversing such instances.
Better to do ti ahead of a time, so traversal stays simple.
Reviewers: lukasstockner97, dingto, brecht
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D2048
Currently for windows only, this is an initial commit towards native
support of NUMA.
Current commit makes it so Cycles will use all logical processors on
Windows running on system with more than 64 threads.
Reviewers: juicyfruit, dingto, lukasstockner97, maiself, brecht
Subscribers: LazyDodo
Differential Revision: https://developer.blender.org/D2049
FCurve evaluation depended on FCurve.curval, which isn't threadsafe.
Now only use this value for debug display,
and pass the value instead of storing in the FCurve for all but debug-display.
This covers the following workflow. Animator will disable Update All 3D views
in order to get quick playback in a particular 3D editor. However, it also
handy to see playback in all editors and image editors to see comparison of
animation and reference footage.
So the idea here is to refresh reasonable editors when playback is invoked
from animation editors.
Commit to make Hjalti happy.
Still not sure how to properly solve the issue, needs some trickery to get
actual optimized values from intersection function (using printf() avoids
some optimization and makes stuff render correct).
For the time being let's just simplify check.
- the name of the enumerator `SNAP_NOT_OBEDIT` was changed to `SNAP_NOT_ACTIVE`.
- the parameter `snap_to_flag` was moved to outside `SnapObjectParams`.
- the member `use_object_edit` was renamed to `use_object_edit_cage`.
- added the arg `params` in `ED_transform_snap_object_project_ray`.
- simplifications in the loop of the function `snapObjectsRay`.
That was a nice and funny hunt, albeit rather time consumming!
To summarize, so far code was using a static global gpu_buffer for pbvh vbo drawing
of 'grid' types (multires mostly?).
There were two issues here:
1) Global gpu buffer was assigned to GPU_PBVH_Buffers->index_buf, but then nearly no
check was done when freeing that buffer, to ensure we were not freeing the global one
(not totally sure this one was actually causing any issue, but was bad and unsafe anyway).
Was solved by adding a flag to GPU_PBVH_Buffers to indicate when we are using some
'common' buffer here, which freeing is handled separately.
2) Main issue: if several multires objects in sculpt mode with different grid size
were present simultaneously, the global gpu buffer had to be resized for each object draw
(i.e., freed and re-allocated), but then the pbvh nodes from other objects storing freed reference
to that global buffer had no way to know that it had been freed, which was causing the segfault & crash.
Was solved by getting rid of that global buffer, and instead allocating one 'grid_commmon_gpu_buffer' per pbvh.
Told ya baby, globals are *PURE EVIL*!
This seems to be illegal and not productive anyway. Do it ahead of
a time now, which solves shading issues in edit mode and prevents
assert from happening.
Opening a DeckLink card for output with the DeckLink object
sometimes failed because some types of card report no support for BGRA pixel
format, yet it works. The test is maintained but a warning is printed
instead of throwing an exception.
Optional cache size suffix added in Decklink output format string: may be
needed for some card that needs more than the default 8 frames.
Issue here is that for Blender, any pass which name starts with 'Diffuse' is considered
a diffuse pass - and it does not really support several passes of the same type in renderresult.
So for now, passtype_from_name() is now checking whether a pass of same type already exists
in render layers, and return 0 (uknown passtype) in this case.
Some adjustments to how bevel edge 'profiles' are adjusted in some
cases. For the bug fix, wanted to handle cases of middle of three
coplanar beveled cases to make profile curve rather than linear
interpolate.
Also undid an earlier decision to make profile plane be perpendicular
to beveled edge i the non-coplanar case.
- add_vertex() can be called from split_vertex() which does not guarantee
to have properly pre-allocate arrays.
- Need to check whether Cycles is compiled with OSL in XML reader.
A bit tricky, need to pass additional information about what the attribute
is and how to deal with it.
BI path stays unchanged, just to make things simplier for now.
Fixes T48555: Cycles GLSL- Incorrect Vertex Color results from Attribute node
Some of these values can get quite large and are hard to read, adding this
makes it easy to read them at a glance.
Reviewed By: sergey
Differential Revision: https://developer.blender.org/D2039
We had transparent shadows disabled for some time because they were causing
drivers to crash. Can't reproduce that issue anymore with current drivers,
so will enable them and see how it goes.
Although there is a "UI limit", it's necessary to have the hard limit on the
property too. I noticed this bug first hand just now, when, after accidentally
setting the size to 8188 (due to a combination of typos and sluggish UI refresh)
my machine locked up completely when trying to jump to that frame. It got so bad
that I had to do a hard reset to fix it - so, it's possible that even 1000 or 2000
as used now are actually way too large still.
It was possible to have issues in cases when several child dependencies
goes to IDs with different layers. In this case order of flushing was not
really well defined, which could lead to cases when indirect dependency
via invisible object wouldn't work.
Need some sort of barrier to prevent scheduling of parent nodes for until
all children are done, but that's becoming quite nasty thing to implement.
Added a temp field to component for now. maybe it's not so crazy actually
and we might use it for evaluation as well, so we wouldn't flush updates
to components which does not affect visible stuff.
This patch enables Tube, Sphere and Box projections in GLSL for the image texture node.
Reviewers: sergey
Projects: #nodes, #opengl_gfx
Differential Revision: https://developer.blender.org/D2036
Previously a whole mesh was stored between undo steps,
This commit uses BLI_array_store to de-duplicate memory use between undo steps.
Memory saving depends entirely on kinds of edits performed,
in own tests 5x-15x less memory use is common.
Compacting the memory does give some overhead however its done in a background thread
so its not blocking in most cases.
New behavior and threading can be ifdef'd out to check for regressions.
See D2026 for details.
Not replacing with some BLI_task_stuff here, tests show this is pointless
(in absolute best case - i.e. single huge mesh in scene - parallelizing here switches
from 0.8ms to 0.5ms for that piece of code - with something like 750ms per frame update...).
Sometimes the autokeying flags don't get set correctly (i.e. the "mode" flags
used for 'Add + Replace' vs 'Replace Only' aren't set), meaning that the old logic
would always fall through to the "replace only" case. When this happens, the resulting
behaviour can be quite strange and hard to debug. This fix prevents problems like
this from continuing to be an issue...
Not sure why, but ANIM_animdata_update() totally ignored NLAStrip type of bAnimListElem.
For now only added support for ANIM_UPDATE_DEPS type of update, don't know whether
others are needed in this case or not... time will say.
This is mainly a maintenance commit which was aimed to make work with
this module more pleasant and solve such issues as:
- Annoyance with looong files, which had craftload in them
- Usage of STL for the data structures we've got in BLI
- Possible symbol conflicts
- Not real clear layout of what is located where
So in this commit the following changes are done:
- STL is prohibited, it's not really predictable on various compilers,
with our BLI algorithms we can predict things much better.
There are still few usages of std::vector, but that we'll be
solving later once we've got similar thing in BLI.
- Simplify foreach loops, avoid using const_iterator all over the place.
- New directory layout, which is hopefully easier to follow.
- Some files were split, some of them will be split soon.
The idea of this is to split huge functions into own files with
good documentation and everything.
- Removed stuff which was planned for use in the future but was never
finished, tested or anything.
Let's wipe it out for now, and bring back once we really start using
it, so it'll be more clear if it solves our needs.
- All the internal routines were moved to DEG namespace to separate
them better from rest of blender.
Some places now annoyingly using DEG::foo, but that we can olve by
moving some utility functions inside of the namespace.
While working on this we've found some hotspot in updates flush, so
now playback of blenrig is few percent faster (something like 96fps
with previous master and around 99-100fps after this change).
Not saying it's something final, there is still room for cleanup and
API simplification, but those might happen as a regular development
now without doing any global changes.
Gives over 50% faster scope update (from 4.5ms to 2.2ms here with SD shot)!
Probably mostly due to more clever usage of thread-local data (which avoids any lock,
when OMP code had a rather stupid critical section for minmax)...
Bevel had assumed that when rebuilding a face that touches
a vertex with beveled edges, the edges of the face at that vertex
would be adjacent in internal order. That is not necessarily true
if there are edges with more than two faces attached.
We could just prohibit beveling any edges that touch a vertex
where this happens (we already don't bevel non-manifold edges)
but the use case in the model of T47257 seems reasonable.
Also had to fix the edge-ordering code, and the face reconstruction
code to take care of cases where the face normal may not be as expected.
New code using loop/finalize model is about 45% faster (from 4.4ms tp 2.4ms per frame,
overall playback of single shot in sequencer in this preview mode goes from 40 to 45fps).
For now, just validate new pointcache names as if they were filnames.
Ideal solution would be rather to validate names in `ptcache_filename()`, but it would
likely break some existing caches - and we'd also have to ensure we still have
uniqueness of name after it has been cleaned up, wich would not be trivial at this stage.
So let's go for simple solution for now, especially since this part of code is to be
fully rewritten in 2.8...
See T48453 for details and test scenes
Reviewers: a.romanov, sergey
Reviewed By: a.romanov, sergey
Projects: #opengl_gfx, #nodes
Differential Revision: https://developer.blender.org/D2011
It's not really clear at this moment how we can detect cases
when attribute needs linearization. For now added a comment
so we don't forget about this, hopefully.
Some closures were missing from calculation, leading to an array
under-allocation, presumable causing memory corruption issues with
emission shaders on OpenCL and was causing issues with Volume 3D
textures with CUDA.
The issue was identified by Thomas Dinges, the patch is different
from the original D2006. See the brief discussion there. Current
approach is similar (or the same) as Brecht suggested.
In Cycle standalone, the xml import of uv's is done in a way that
a vertex could only have one unique uv coordinate. In practice it
is most of the time not the case.
Patch by Laurent Boiron, thanks!
This fixes a rare case where NaNs could exist inside Cycles.
When certain invalid meshes were passed in, Cycles would try too normalize
a zero length normal during its setup stage. While it does check against
division by zero, it still returns a zero length normal and passes it on to
the path tracing kernel. The kernel then operates under the assumption that
normals are valid, and in the case of such a zero length normal, would
eventually create NaNs that propagate through and result in black pixels.
Reviewers: #cycles
Subscribers: brecht, sergey
Projects: #cycles
Differential Revision: https://developer.blender.org/D2008
Previously, the VideoDeckLink module was rejecting combinations of
video modes and pixel formats that cause padding. For example HD720
(1280 pixels per line) and 10BitYUV (encodes by blocks of 48 pixels) =>
16 padding pixels at the end of each line.
This format is now accepted and the padding pixels are sent to the BGE.
It is sufficient to edit the UV map so that these pixels are not displayed.
The title says it all actually.
Added special custom data type, because we don't know in advance
whether we're referencing UV or Color layer. Also made it so vertex
attributes are normalized.
TODO: Border render in viewport ignores the normalization of the
attribute array for some reason, will be looked into still.
Reviewers: mont29, brecht, campbellbarton
Reviewed By: brecht, campbellbarton
Differential Revision: https://developer.blender.org/D2022
Use loadImage() method whenever possible to transfer video frame
directly from GL driver to Decklink frame. This is possible when
alpha is true on ImageRender and render size matches exactly video size.
Best results with offscreen render.
At some point the idea was that we could have an optimization where we could
render multiple render layers without re-exporting the scene, by just updating
the layer bits. We are not doing this now and in practice with the available
render layer control like exclude layers it's not always possible anyway.
This makes it easier to support an arbitrary number of layers in the future
(hopefully this summer), and frees up some useful bits in the kernel.
Reviewed By: sergey, dingto
Differential Revision: https://developer.blender.org/D2020
This commit implements Bump node in GLSL, making it possible to
see previews of bump mapping in viewport without need to render.
Nothing really fancy going on here, just uses internal dFdx/dFdy
functions to get derivatives of the surface and map itself.
Quite basic but seems to behave correct-ish.
This commit also makes Displacement material output to affect
viewport shading by re-linking unconnected Normal input to a
node which was used for displacement output (via Bump node).
Intention of all this is to make it really easy to do bump map
painting with Cycles as an active render engine.
Reviewers: campbellbarton, mont29, brecht, psy-fi
Reviewed By: brecht
Subscribers: Blendify, eyecandy
Differential Revision: https://developer.blender.org/D2014
Since version 6 G++ switched to C++11 by default, which breaks some logic
around WITH_CXX11 checks in out CMake files, leading to compilation errors.
This is easy to solve by explicitly enabling older C++ standard when C++11
was not explicitly enabled by CMake options.
However, G++-6 will also use new ABI by default even if older standard was
specified in the compiler options. This is being addressed by a special
define flag.
This tricks made it possible to use new G++-6 without need to recompile
any of pre-compiled libraries.
However, this might break compilation with existing system libraries, which
might already be using new ABI. We can't address this automatically, so
now we simply default WITH_C11 and WITH_CXX11 options to whatever defaults
of the current compiler are. This means, for G++-6 we'll set WITH_CXX11 to
truth. This should make linking with system libraries working just fine,
but to make pre-compiled libraries we still might need to disable CXX11.
This should work fine work for a new environments with G++-6 and install_deps
script run from scratch there, because C++ standard will be the same for
both Blender dependencies and Blender itself.
This commit makes Dynamicpaint modifier evaluation (during playback) a few percents quicker.
However, it makes dynapaint's 'image sequence' baking about 33% quicker (from 119 to 77 seconds
in own heavy test), partly due to switch to BLI_task itself (about 20%), and partly due to
optimizations (remaining ~13%).
As usual, did a lot of tests here to ensure nothing is broken, but a lot more users' testing would definitively
be welcome too! ;)
Note that some quite meaningless omp forloops have been removed (parallelizing thousands of vec copy does
make it two or three times quicker, but the few hundreds of microseconds gained do not make any difference
in a hundreds millisecond process).
Also, this code could still use a lot more cleanup (naming etc.), the way it (tries to) handle malloc faults
is also totally flacky and makes the code horribly verbose and convoluted in some places - without actually
catching all possible faults (memarena could make it more easy to handle here), etc.
Usage:
* D+X - Works anytime, anywhere
* Shift-X - Works in EditMode only
* Via Delete Menu - EditMode only
Often doing video tutorials or perhaps during dailies/shot review you want to
quickly get rid of a quick scribble you made for making a point, without having
to undo (i.e. maybe you edited some objects in between) and/or without having
to use the eraser (i.e. it'd take too long to cover the entire area).
On light coloured backgrounds (especially on white), it was impossible to see
where the cursor was. This commit adds a second ring (black) to the cursor so
that on light backgrounds, even if the light ring isn't visible the black one
will be.
Gives 3-4% speedup in pre-bake step (from 112 to 108 seconds with own heavy test file).
Note that here we have a huge potential performance boost if we replace the flat
`Bounds2D *faceBB` array of UV tris bounding boxes by a real 2D AABB tree (right now,
we can loop over all UV triangles times the number of pixels of the surface times 5...).
It gives some slight differences on the plane corners, but can't
really figure out source of the issue here yet.
It's still better than fully white texture for the previews anyway.
At this point we should perhaps ifdef chunks of the code in order
to have faster GLSL compilation and probably even faster compiled
code. Will look into this shortly.
Compile time per kernel increased alot after recent image commits, re-shuffle some code to fix this.
Patch by "LazyDodo".
Differential Revision: https://developer.blender.org/D2012
In addition to the original bug report, I've gone through cleaning up a range of
related bugs which only became clear when hunting around the code...
* Custom Handle References weren't getting cleared when the bones they used got
deleted. But, neither was the custom bone shape location/transform reference.
* Various places where posebone settings are copied around were also missing code
to handle the new Bendy Bone properties.
(WHY DO WE HAVE SO MANY VARIATIONS OF COPYING POSE DATA!?!?)
* If duplicating a Bendy Bone with custom references, and the custom references
are also selected/duplicated, the new Bendy Bones will use the corresponding
duplicated bones
This commit adds Peak Memory to the stamp options, the value is the same one that is already shown in the image viewer.
Requested by @nutel.
Reviewers: campbellbarton
Subscribers: campbellbarton, nutel
Differential Revision: https://developer.blender.org/D1989
Once and for all - I hope!
So, for the records: Xinput1 events' valuators have a first_axis and axis_count defining the 'range'
of valid axes for that event - but valid data in the valuator's array **always** start at zero
(which means that, if event reports first axis as being axis 2, axis 2's value is in axis_data[0]).
Was not so far, because this effect is not modifying its 'own' PaintPoint, which means
it's not threadsafe. Since a global lock (mutex or spinlock) would not be much efficient
(we need to lock a given point pretty much all the computaion cycle), and since locking
a same PaintPOint from different threads at the same time is *very* unlikely,
solution here is to use an 'array of locks', one for each PaintPoint (same thing as BLI_bitmap,
using atomic ops to set/clear bits).
Here in own test (complex dynapaint over a huge sphere combining all dynapaint types), it gives
20% speedup of the whole dynapaint simulation!
Note: maybe we'd want to move that kind of bitlock into BLI lib some day - not totally sure how,
so let's keep it local for now...
Vertex's normal is always normalized - and its conversion from short to float should not change that!
On the other hand, linear interpolation of three normals **does not give a normalized vector**
(unless all three inputs are exactly the same).
Also, minor optimization, avoid recomputing that interpolated normal twice for each PaintUVPoint.
This adds support for CUDA Texture objects (also known as Bindless textures) for Kepler GPUs (Geforce 6xx and above).
This is used for all 2D/3D textures, data still uses arrays as before.
User benefits:
* No more limits of image textures on Kepler.
We had 5 float4 and 145 byte4 slots there before, now we have 1024 float4 and 1024 byte4.
This can be extended further if we need to (just change the define).
* Single channel textures slots (byte and float) are now supported on Kepler as well (1024 slots for each type).
ToDo / Issues:
* 3D textures don't work yet, at least don't show up during render. I have no idea whats wrong yet.
* Dynamically allocate bindless_mapping array?
I hope Fermi still works fine, but that should be tested on a Fermi card before pushing to master.
Part of my GSoC 2016.
Reviewers: sergey, #cycles, brecht
Subscribers: swerner, jtheninja, brecht, sergey
Differential Revision: https://developer.blender.org/D1999
It was totally useless to multiply diffuse color with the vertex color
when doing texture painting. It was masking actual texture and only was
forcing artists to create an empty vertex color layer to work this around.
The assembler version returns the previous value of the variable while
all the other versions return the new value.
This was fixed before but somehow it came back when the atomic module
was refactored.
Previously, cache size was only 3 frames, which caused problems with older
types of Decklink cards (was working with the 4K Extreme).
The default is now 4 frames and it is possible to specify a different cache
size when opening the card:
VideoDecklink("<mode>/<pixel>[/3D][:<cachesize>]", cardIdx)
Ex with 8 frames in cache:
VideoDecklink("pal/2vuy:8", 0)
The idea of pole merge is to fade interocular distance after a certain
altitude to zero when altitude goes closer to a pole. This should prevent
annoyances looking up in the sky or down to the bottom.
Works for both panorama and perspective cameras when Spherical Stereo
is enabled.
Reviewers: dfelinto, brecht
Reviewed By: brecht
Subscribers: sebastian_k
Differential Revision: https://developer.blender.org/D1998
This commit makes it so malloc() is only happening once per volume and
once per transparent shadow query (per thread), improving scalability of
the code to multiple CPU cores.
Hard to measure this with a low-bottom i7 here currently, but from quick
tests seems volume sampling gave about 3-5% speedup.
The idea is to store allocated memory in kernel globals, which are per
thread on CPU already.
Reviewers: dingto, juicyfruit, lukasstockner97, maiself, brecht
Reviewed By: brecht
Subscribers: Blendify, nutel
Differential Revision: https://developer.blender.org/D1996
When using Left Click select, it wasn't possible to sculpt using E+LMB.
I've changed the order of things in the keymap so that the select operator won't
end up catching and blocking all these events.
Change the order of the bending controls ("Curve XY Offsets") so the user can activate both InX and OutX by holding down the left mouse button. This way, it's easy to bend symmetrically on X or Y.
This commit/patch/branch brings a bunch of powerful new options for B-Bones and
for working with B-Bones, making it easier for animators to create their own
rigs, using fewer bones (which also means hopefully lighter + faster rigs ;)
This functionality was first demoed by Daniel at BConf15
Some highlights from this patch include:
* You can now directly control the shape of B-Bones using a series of properties
instead of being restricted to trying to indirectly control them through the
neighbouring bones. See the "Bendy Bones" panel...
* B-Bones can be shaped in EditMode to define a "curved rest pose" for the bone.
This is useful for things like eyebrows and mouths/eyelids
* You can now make B-Bones use custom bones as their reference bone handles,
instead of only using the parent/child bones. To do so, enable the
"Use Custom Reference Handles" toggle. If none are specified, then the BBone will
only use the Bendy Bone properties.
* Constraints Head/Tail option can now slide along the B-Bone shape, instead of
just linearly interpolating between the endpoints of the bone.
For more details, see:
* http://aligorith.blogspot.co.nz/2016/05/bendy-bones-dev-update.html
* http://aligorith.blogspot.co.nz/2016/05/an-in-depth-look-at-how-b-bones-work.html
-- Credits --
Original Idea: Daniel M Lara (pepeland)
Original Patch/Research: Jose Molina
Additional Development + Polish: Joshua Leung (aligorith)
Testing/Feedback: Daniel M Lara (pepeland), Juan Pablo Bouza (jpbouza)
Together with the extended loop callback and userdata_chunk, this allows to perform
cumulative tasks (like aggregation) in a lockfree way using local userdata_chunk to store temp data,
and once all workers have finished, to merge those userdata_chunks in the finalize callback
(from calling thread, so no need to lock here either).
Note that this changes how userdata_chunk is handled (now fully from 'main' thread,
which means a given worker thread will always get the same userdata_chunk, without
being re-initialized anymore to init value at start of each iter chunk).
New code is actually much, much better than first version, using 'fetch_and_add' atomic op
here allows us to get rid of the loop etc.
The broken CAS issue remains on windows, to be investigated...
Some variants of gcc compilation were reporting 'control reaching end of non-void function' error
in this switch/case maze. Either use break everywhere or not at all (which is simpler, since we
only always return anyway...).
There are some serious issues under windows, causing deadlocks somehow (not reproducible under linux so far).
Until further investigation over why this happens, better to revert to previous
spin-locked behavior.
This reverts commits a83bc4f597 and 98123ae916.
Reading the shared state->iter value after storing it in the 'reference' var could in theory
lead to a race condition setting state->iter value above state->stop, which would be 'deadly'.
This **may** be the cause of T48422, though I was not able to reproduce that issue so far.
When writing temp blenbuffer file, libraries of linked datablocks where not tagged correctly, which
means they were not put in the temp Main used to write the buffer, resulting in implicit localization
of linked data.
Previous to 2.77, this used to be default behavior, was changed in rB591f4549c958b.
However, in most append cases, you do want a full localization of your data, so this new behavior
is kept by default, but there is now an option in append operator to only localize the 'first level'
of data (i.e. datablocks from linked library itself, and not those from other 'sub-libraries').
On big and complex rigs like blendrig or koro, it can give up to ~10% more FPS in best cases.
Hard to tackle all cases in tests though, so please report any unexpected slowdown
in armature animation playback!
Line lengths, monolined 'if' statements, int -> bool, etc.
Also, replaced some internal cooked stuff by BLI helpers (most notably, the
'is inside UV triangle' code in `dynamicPaint_createUVSurface()`), and some
other minor optimizations.
Previously if image only had single channel only z buffer value was displaying.
This isn't handy for cases when you've got single channel buffer which is not
a z buffer.
Also fixed possible read past the array.
install_deps can fail due to conflict between gcc (referenced by base-devel) and
gcc-multilib if the latter is installed. This avoids the conflict by filtering
the contents of base-devel when needed.
Reviewers: mont29
Differential Revision: https://developer.blender.org/D1944
This way, we also save 3/4th of memory for single channel byte textures (e.g. Bump Maps).
Note: In order for this to work, the texture *must* have 1 channel only.
In Gimp you can e.g. do that via the menu: Image -> Mode -> Grayscale
Until now, single channel textures were packed into a float4, wasting 3 floats per pixel. Memory usage of such textures is now reduced by 3/4.
Voxel Attributes such as density, flame and heat benefit from this, but also Bumpmaps with one channel.
This commit also includes some cleanup and code deduplication for image loading.
Example Smoke render from Cosmos Laundromat: http://www.pasteall.org/pic/show.php?id=102972
Memory here went down from ~600MB to ~300MB.
Reviewers: #cycles, brecht
Differential Revision: https://developer.blender.org/D1981
Take advantage of the efficiency provided by the snap_context.
Also fixes errors:
- volume snap fails based on view angle (T48394).
- multiple instances of dupli-objects break volume calculation.
Pass distance argument so its possible to limit the range we get all hits from.
Other changes:
- Use boundbox test before calling callback, avoids redundant calls.
- Remove meaningless return value.
- Add doc string, explaining purpose of this function.
This commit makes use of new taskpool feature (instead of allocating own tasks),
and removes the spinlock used to generate chunks (using atomic ops instead).
In best cases (dynamic scheduled loop with light processing func callback), we
get a few percents of speedup, in most cases there is no sensible enhancement.
This is what old dependency graph was doing and apparently this avoids some updates,
however it's not really clear why those nodes are considering done. Maybe just because
of the way how relations are working. But needs a closer investigation.
Appears mutex was guarateeing number of tasks is not modified at moments
when it's not expected. Removing those mutexes resulted in some hard-to-catch
locks where worker thread were waiting for work by all the tasks were already
done.
This reverts commit a1d8fe052c.
Brain melt here, intention was to reduce number of tasks in case we have not much chunks of data to loop over,
not to increase it!
Note that this only affected dynamic scheduling.
It uses some additional compute power and the evaluation priority is
not even used.
This brings fps 88.2 with blenrig_for_debugging.blend on this desktop.
Simple thing, and apparently fps goes up to 80 with the demo file from jpbouza.
Not sure why at this point fps is so much higher than the old dependency graph
here now. And it's definitely something what others should verify as well.
This reduces stress on the task scheduler and avoids some unwanted overhead
caused by all the threading business in the cases when there's only one
children node. We try to immediately switch to it's evaluation now, keeping
active thread up and running.
This bumps FPS from 58 to 64 on the blenrig test file from jpbouza.
This commit implements new function BLI_task_pool_push_from_thread()
who's main goal is to have less parasitic load on the CPU bu avoiding
memory allocations as much as possible, making taks pushing cheaper.
This function expects thread ID, which must be 0 for the thread from
which pool is created from (and from which wait_work() is called) and
for other threads it mush be the ID which was sent to the thread working
function.
This reduces allocations quite a bit in the new dependency graph,
hopefully gaining some visible speedup on a fewzillion core machines
(on my own machine can only see benefit in profiler, which shows
significant reduce of time wasted in the memory allocation).
Many keyboard layouts (italian, spanish, german...) have direct access to '+' key on main
keyboard area (not the numpad one), ans x11 has own define for this key, so use it instead
of generating an unkown key event.
Note that we most likely have much more missing 'specific' keycodes for non-US keyboard layout,
but think since we already had a 'minus' keyevent, supporting 'plus' one is totally consistent.
And we had a spare space in our defined values just for it even!
This keyevent is only supported/generated by x11 and cocoa Ghost backends for now,
neither SDL nor win32 seem to have matching key events...
Title says it all, this adds OpenCL float4 texture support.
There is a bug in the code still, I get a "Out of ressources error" on nvidia hardware here, not sure whats wrong yet.
Will investigate further, but maybe someone else has an idea. :)
Reviewers: #cycles, brecht
Subscribers: brecht, candreacchio
Differential Revision: https://developer.blender.org/D1983
This commit:
* Removes most of all dirty internal details from public atomi_ops.h file, and move them into /intern private subdir.
* Removes unused 'architectures' (__apple__ and jemalloc).
* Split each implementation into its own file.
* Makes use of C99's limits.h system header to determine pointer and int size, instead of using fix hardcoded list of architectures.
* Introduces new 'faked' atomics ops for floats.
Note that we may add a lot more real and 'faked' atomic operations over integers and floats
(multiplication, division, bitshift, bitwise booleans, etc.), as needs arise.
Reviewers: sergey, campbellbarton
Differential Revision: https://developer.blender.org/D1982
This was originally done for T46320 in order to re-store depsgraph state
after all the constraint modifications.
However, it relations were updated if there's any IK chain the the pose,
which is a bit too annoying.
Set is much slower to iterate through (due to cache misses and such) and
the only advantage of using set is faster removal of link. However, we are
iterating links much much more often than removing them, and even when we
are removing links we don't really need to remove link from nodes which it
connects -- we don't support partial depsgraph updates, so removing links
from nodes on destruction is a waste of time.
If we ever want to support partial updates we can have dedicated function
to remove link from nodes it connects.
This gives a surprising increase of fps from 42 to 56 with test file from
Mr. J.P.Bouza (blenrig_for_debugging.blend). Surprising because old DEG is
actually slower here (52 fps). Didn't see any regressions (and don't see
why they will happen), so let's ask our riggers and animators to perform
further speed tests ;)
Use atomic operations instead, should in theory improve timing of
scheduling. However, probably not so visible yet because actual
task scheduling still have some locks and memory allocations.
Baby steps, what would i say.
This was causing alignment issues which were only visible on a platform
of particular bitness, so it was easy to break stuff for 32bit when
working on 64bit platform and vice versa.
* Fix "Attempt to free NULL pointer" when copying strokes for the first time
* Fix poll callback on "paste" operator, so that it is possible to paste
strokes when there are no editable strokes visible.
Commit 86abddc9, caused an error when the face-region included boundary edges.
Since removing the faces first, caused the edges along the boundaries to be removed.
Add support for deleting faces and internal edges, that keeps boundaries.
Now, when creating new keyframes in the graph editor by ctrl-clicking, only
the newly created keyframes will be selected. This is a little workflow tweak
to make it faster to work, as you no longer have to deselect all, and then
re-select the newly added keys in order to manipulate them.
The old behaviour (not modifying the selection status of the old keys) has been
kept, but is now available via Shift-Ctrl-Click.
Feature request from @Shhlife
This commit makes remove_table skip the freeing if the offset is
already set to invalid - or, if it wasn't, set it to invalid after freeing.
That's what the current code was already doing in the Manager classes,
this change allows them to just call remove without the additional code.
Also, two potential memory leaks where new tables were always allocated
without freeing the old ones are fixed.
Reviewers: sergey, dingto, brecht
Differential Revision: https://developer.blender.org/D1974
This is an experimental option that I found would have been useful to have,
when making my earlier anim test. The only contentious point is whether it's
better to use a separate boolean flag (perhaps bound to the labels) instead
of overloading this setting (and describing the usage in the tooltip).
In the nodes editor, Ctrl-LMB-drag is taken by "Cut", and it seems impossible
for any other keymap to override that. Instead, the default lasso select hotkey
there is Ctrl-Alt-LMB-drag instead.
To get Grease Pencil lasso select (in GP Editmode) to work in the Nodes Editor too,
this commit adds the Ctrl-Alt-LMB-drag binding here too. However, to make things easier,
this extra binding will be part of the Grease Pencil keymap everywhere. There doesn't
seem to be any conflicts with having this in place - until we find them, this should
be ok to have.
I'm still not happy with this layout as it is now, but it seems a bit less unbalanced
than what I'd been trying before. So, let's leave this as-is for now.
This commit adds some of the initial support for a properties region in the
Action Editor. There are currently no panels to display, as there is still
a lot of work required to port over the required internal architecture to
support the panels seen in the Graph Editor.
After some test, a new iteration parameter has been added in order to
apply repetitive smoothing to the stroke. By default 1 iteration is applied,
but can used any number between 1 and 3.
The repetition uses different levels of intensity from 100% of the defined smooth
factor for the first loop, 50% for the second and 25% for the third. We use in each
loop a smaller value in order to avoid deform too much the stroke.
Now, stroke-related things (thickness, volumetric, points) and fill-related things
(HQ fill) are in the relevant columns, instead of having some in each column.
There is still some instability in how the triangulations are happening,
where the triangle count of the resulting triangulation fluctuates resulting
in weird artifacts sometimes.
To reproduce, try drawing some U-shapes, and keep reloading the file.
While trying to track down why I still keep getting some random "extra" triangles
showing up when reloading files with fills, I noticed that there's an index mismatch
here that may cause problems in some rare cases:
There are "gps->totpoints" items in tmp_triangles, but there should only be
"gps->tot_triangles" triangles in the gps->triangles array. If for whatever reason
some of the triangles in tmp_triangles are invalid, this could have meant that
while a tmp_triangles candidate was skipped, a corresponding slot for a valid
triangle also got skipped.
Improve filling for concave shapes using a triangulation of the stroke.
The triangulation information is saved in an internal cache and only is
recalculated if the stroke changes.
The triangulation is not saved in .blend file.
Reviewers: aligorith
Maniphest Tasks: T47102
Differential Revision: https://developer.blender.org/D1705
`dynamicPaint_doMaterialTex` was called from inside an omp parallel section with brush->dm
itself, and not the local dm copy. Generating looptri data is not thread safe at all...
Swap the for-loops in the RenderBuffer reset code to follow the convention
of looping over y in the outer loop.
The improved cache performance won't really be noticable here, but it's nicer
if it follows the usual style.
Take curvature into account when calculating handle length.
Gives significantly better results for curve dissolve and 10-20% more efficient freehand drawing.
* When Baking wasn't used we got an error.
* On top of Volume Nodes (NODES_FEATURE_VOLUME), we now also check if we need volume sampling code,
so we can disable that as well and save some further compilation time.
If the CUDA Toolkit is installed and the user is on Linux,
adaptive, feature based CUDA runtime compile is now possible to enable via:
* Environment flag CYCLES_CUDA_ADAPTIVE_COMPILE or
* Debug menu (Debug value 256) in the Cycles UI.
This commit avoids draw of invisible parts of image in image editor,
making it faster to re-draw the image.
Especially handy when painting on a high-res image when zoomed-in.
This commit simplifies the code for the image arrays. Instead of having 2 arrays for float and byte textures,
we now use an array here. This simplifies the code (avoids code duplication), and makes it possible to easily extend it
with float1 and half-float types in the future.
Only tested with CPU yet, plus some cleanup / code de-duplication is still possible here.
Reviewers: #cycles, sergey
Reviewed By: #cycles, sergey
Subscribers: jesterking, sergey
Differential Revision: https://developer.blender.org/D1969
Originally the value was only needed when building the springs,
so a pointer to the input data was somewhat ok. However, recalculating
spring length dynamically requires keeping the actual value around.
The average length is used to adjust self-collision and stiffness
to match mesh density, and in either case considering the always zero
restlen sewing springs doesn't actually make much sense.
The struct_springs variable is also used for other things like total
spring count, so a new one needed for the true struct springs.
This comes out of considering a one-dimensional transition in weight
on a rectangular cloth grid. At the transition face loop, one side
of each rectangular face would be scaled by k1, and the opposite one
by k2, thus turning the rectangle into a trapezoid. Averaging would
be used to choose the scale factor for the remaining two sides.
If Geometrical Mean, i.e. sqrt(k1*k2) is used, it so happens that the
diagonals of the trapezoid also end up scaled by sqrt(k1*k2) compared
to the original rectangle. This means that the same scale factor is
correct for both structural and shear springs, which is not the case
with simple average.
Previously it was all over the place: without a vertex group it was
always enabled, and with it it depended on the Self Collision option
due to a likely editing mistake. Now it should be in sync with what
disables the fields in the UI.
The verts->shrink_factor field is changed to just cache the vertex
weight, with the shrink_min/shrink_max interpolation done later.
This is because cloth_apply_vgroup only runs if there are vertex
groups, and thus the factor may not update after property changes.
In addition, bending springs are now also affected by the shrink
factor to avoid visible distortion in object shape at high shrink.
From description, Use Coordinates evaluates the texture using
target coordinates in the local space of the force field object.
2D is supposed to ignore the Z coordinate. Thus one would assume
that if both are enabled, the force field effect would move with
the force field object, and Z would be 0.
However, instead first the 2D option projects points onto a plane
passing through the global zero and orthogonal to the local Z,
and only then the resulting point is transformed into local space.
Z is not locked at 0, so procedural textures like Spherical Blend
don't work as expected.
To fix this, apply local transform first, and then just clear Z if 2D.
Separate the creation of trees from EditMesh from the creation of trees from DerivedMesh.
This was meant to simplify the API, but didn't work out so well.
`bvhtree_from_mesh_*` actually is working as `bvhtree_from_derivedmesh_*`.
This is inconsistent with the trees created from EditMesh. Since for create them does not use the DerivedMesh.
In such cases the dm is being used only to cache the tree in the struct DerivedMesh. What is immediately released once
bvhtree is being used in functions that change(tag) the DM cleaning the cache.
- Use a filter function so users of SnapObjectContext can define how edit-mesh elements are handled.
- Remove em_evil.
- bvhtree of EditMesh is now really cached in the snap functions.
- Code becomes organized and easier to maintain.
This is an important patch for future improvements in snapping functions.
Using SSE2 intrinsics when available for this kind of conversions.
It's not totally accurate, but accurate enough for the purposes where
we're using direct colorspace conversion by-passing OCIO.
Partially based on code from Cycles, partially based on other online
articles:
https://stackoverflow.com/questions/6475373/optimizations-for-pow-with-const-non-integer-exponent
Makes projection painting on hi-res float textures smoother.
This commit also enables global SSE2 in Blender. It shouldn't
bring any regressions in supported hardware (we require SSE2 since
2.64 now), but should keep an eye on because compilers might have
some bugs with that (unlikely, but possible).
FSAA sample files in EXR format are no longer always updated (after some
changes between 2.73 and 2.74 releases), and the reported bug was caused
by old samples from a previous frame that were being merged by mistake.
The present revision addresses the documented issue by entirely skipping
the rendering of auto-generated scenes when there are no Freestyle strokes
to render, which suppresses sample merging of the render layers that were
not rendered.
Just avoid some unneeded initialization functions when the threaded
processor is simple enough to only depend on current chunk start
scanline and number of scanlines.
Caused by wrong alpha value in rB34b30f847429ac (used 0.01, should've been 0.1). Also made sure all editors initialize same default theme values for panels and cleaned up code a bit.
Works fine with old userpref.blend files.
Basically, issue is that outliner stores editbones for an edited armature, which are only valid during EditMode.
When more than one object use the same armature, selecting ebones from same armature but under non-edited object
would first select that object (and hence switch out of editmode), and then try to select editbone while editdata
no more exist.
Solution for now is to not change active object in this case - it's not totally ideal (not consistent),
but other solutions (like switching edited object without leaving editmode, or rebuilding (part of)
outliner tree in-between) are horribly hackish and most likely prone to epic failure anyway.
Shift-space was reserved for IME support, however IME will only
be enabled on certain languages. We can avoid any IME-related
exceptions form other languages without too much trouble.
There's one weak point around ui_ime_is_lang_supported() tho,
which is it might be doing string comparison a bit too much
often now, this we can avoid by handling those checks from BLT.
Not very efficient solution -- it'll update mapping node on init ntree exec and will
not work for viewport GLSL shading perhaps, but it's as good as it could be within
current dependency graph.
The issue here is that manual edit of values will cause cached matrix re-evaluation.
but using animation does not use rna update callbacks hence no matrix update was
happening.
- scripts that execute directly need to include their basedir in the sys.path
- modules which are in a directory without an __init__.py weren't importing.
__GCC_HAVE_SYNC_COMPARE_AND_SWAP_n are not defined on this architecture
for some reason, however those functions are available.
Adding a workaround for newly used __sync_fetch_and_and() function.
Own regression from rB019ce363b01bba0af, not totally sure why removing the heading
zero value in element_sum broke binary search distribution in random branch, but...
New code makes thing sooo much simpler, by simply getting rid of all zero-weight
items in element_sum (at the cost of adding a temporary extra index mapping array),
that way neither random nor 'even ordered' distribution code has to wory anymore
about skipping zero-weighted items!
In some cases, we can have nodegroups without attached nodetree, this case is checked
in top `node_group_has_output()` entry point, but not in recursive `node_group_has_output_dfs()`.
Issue appeared after rB42824f4403db3a35, presumably because check on node flag itself
was skipping that case, which is no more possible when using ID tags.
Reported by sebastian_k over IRC, thanks.
This commit mostly:
* removes some (very) old messages or hacks (like those for ubuntu 10.4...).
* Moves lib-specific dependencies installations to upper level
(simpler to manage, and those are small enough packages).
* Uses new dnf package manager for fedora (sigh, we now have tree different commands for fed/suse/rehl).
* fixes or enhances some options for commandline tools.
* Force anew building of oiio package, due to insane dependencies of this one in recent distro.
Part of D1933.
Now CD_SHAPEKEY_INDEX customdata is stored in edit-mode when hooks and vertex parents are used.
This also fixes a bug where undo would loose key-index data.
Move to structs for BM_mesh_bm_to/from_me to avoid passing many argument, which mostly aren't used.
The issue was caused by missing indirect dependencies from other scenes in the
current scene depsgraph, which was causing some threading conflicts.
Not sure what would be ideal solution here, for now use stupid but rather robust
approach to solve that. Maybe there's something better to do here tho.
Currently a lot of the nodes in the new dependency graph are empty placeholders
for organizational purposes. These nodes would, however, still be assigned a task
which gets scheduled and takes up some time for worker threads to pop from the
queue and run. This can be avoided by skipping these nodes during depsgraph
scheduling, and scheduling their childrent right away. Gives a few percent speedup
in BlenRig.
Some of the tagging functions would be called for every operation, and then
in turn tag their ID nodes with all their operations again. With extensive rigs
we get ID nodes with a lot (10,000+) operation nodes, which leads to millions
of unnecessary tagging calls.
Previously, adding a new vertex with Ctrl-LMB would move an existing
when there was an active spline but no active point.
This function is used as a fallback, extending an existing active point is already handled.
Request by Andy, should help him a lot doing weird and wonderful hair styles.
A bit experimental yet, details of behavior might be changed after some real
usage feedback.
Annoying bug caused by temp nature of looptri layer for CCGDM.
Fixed in a similar to CCG loops by using lock when allocating and
filling looptri arrays.
Real fix would be to make sure this array is allocated on object
evaluation using DAG's eval_flag, but that's more involved change
which we'll work on later.
Operator would call `WM_file_read()` directly whith G.main->name as filepath, which
gets freed whith main during new reading of file... Now use a local copy instead.
This is a follow-up fix for rBa10b2fe. The Action constraint uses a hackish
stub object and pose, which doesn't have a hash table for fast lookups.
This doesn't seem to be a big issue with the old depsgraph, but in the new
depsgraph it creates a large number of cache misses and significant slowdown,
possibly because of additional threading and less simple bone loops.
Normal Map node support for GLSL mode and the internal render (multiple tangents support).
The Normal Map node is a useful node which is present in the Cycles render.
It makes it possible to use normal mapping without additional material node in a node tree.
This patch implements Normal Map node for GLSL mode and the internal render.
Previously only the active UV layer was used to calculate tangents.
- break when object in hit-buffer.
- don't measure distance to object centers that can't be projected.
- take pixelsize into account for distance pixel distance limit.
The hash table is used to look up bone poses, particularly during constraint
evaluation. Without this the default BLI_findstring method on a plain ListBase
is used, which is really slow for extensive rigs.
Using 'shell-thickness' to offset UV's meant very sharp corners would offset far outside the image
causing project-paint to hang while collecting all pixels for each UV face.
Clamp the maximum offset to prevent this.
Couple of issues here:
- Was a bug in heap memory allocation when run out
of allowed stack memory.
- Debug MSVC was failing because it uses separate
allocator for some sort of internal proxy thing,
which seems to be unable to be using stack memory
because allocator is being created in non-persistent
stack location.
Part 2 of Fix for T48200. Without this, it's not possible to exit GPencil edit mode using tab key with pies enabled.
Menus are somehow similar to operators, which let the event pass through too after operator poll returns 0. So think we should handle them similar here.
Can't guarantee this is totally safe though ;)
This fixes a lot of things in NLA RNA update handling (which basically did not update anything previously).
There are more update issues with this editor though...
Issue was that before writing to disk grids are clipped against the
density field's tree to optimize for memory/disk space, which in the
case of simulations with no density field results in all grids having
empty trees.
For now avoid clipping against empty grids, but perhaps in the future it
can be interresting to have some UI parameters to let the user choose
the grid used for clipping.
This mainly touches extern libraries and few debug-only places in intern.
Some summary:
- External libraries are not strict at all about missing declarations,
so we can rather safely remove such warning together with other strict
flags.
- Bullet has some static functions which are not used.
Those were commented out.
- Carve now has some unused debug-only functions commented out as well.
While we're on the way of getting rid of Carve, it makes sense to make
things a bit cleaner for the time being.
- In LZMA we have some parts disabled which gives some set but unused
variables which is rather correct.
- Elbeem had quite some variables set and never used because their usage
is inside of debug-only code which is commented out.
Note about patching upstream libraries: surely one might say that we
have to make local patchset against this, but own experience says it
only gives extra work trying to merge such tweaks to a new upstream
version and usually it's just faster to re-apply such fixes again after
bundling new upstream library.
Mainly makes logging less verbose when doing progressive sampling in viewport.
Such kind of verbosity is not really possible to be filtered out with `grep`
so let's reshuffle few lines of code.
This introduces a snap-context that can be re-used for casting rays into the scene
(by operators such as walk-mode, ruler and transform code).
This can be used to cache data between calls too.
The script is updated for ArchLinux, since all dependencies are included in Arch's official repositories.
I've made a few changes, such as enabling OCIO and OSD without requiring locally installed lib-path, and a fix to ''get_package_version_ARCH()' so it ignores package epoch (as in the case of ffmpeg).
I intend to look at OpenVDB next.
Reviewers: mont29
Differential Revision: https://developer.blender.org/D1901
The 'precision' and 'dynamic' settings for binding are now always visible.
The settings that can not be edited after binding are disabled (not inactive).
I find it useful to see with what settings a mesh was bound, in case
the file is not mine or if I simply lost track of it.
Simple idea, use threads when dealing with "Copying Transformations to device"
scene update step. Only do it if there's enough objects in the scene.
Hopefully only brings less synchronization time and doesn't break anything.
From tests on my desktop this brings down transform update time from 58sec to
11sec on victor_cpu.blend scene from out benchmark.
This is an attempt to gracefully handle out-of-memory events
and stop rendering with an error message instead of a crash.
It uses bad_alloc exception, and usually i'm not really fond
of exceptions, but for such limited use for errors from which
we can't recover it should be fine.
Ideally we'll need to stop full Cycles Session, so viewport
render and persistent images frees all the memory, but that
we can support later, since it'll mainly related on telling
Blender what to do.
General rules are:
- Use as less exception handles as possible, try to find a
most geenric pace where to handle those.
For example, ccl::Session.
- Threads needs own handling, exception trap from one thread
will not catch exceptions from other threads.
That's why BVH build needs own thing.
Reviewers: brecht, juicyfruit, dingto, lukasstockner97
Differential Revision: https://developer.blender.org/D1898
This mimics behavior of default allocators in STL and allows all the routines
to catch out-of-memory exceptions and hopefully recover from that situation/
This commits implements threaded sorting of references when looking for
object spatial split. It mainly useful when doing initial binning, which
happens from main thread.
Gives nice speedup of BVH build for Bunny.blend: 36sec vs. 55sec for
the Rabbit mesh BVH build.
On more complex scenes the speedup is probably minimal, but still nice
to have more instant rendering for simplier scenes.
Some further tests with production scenes would be interesting.
Reviewers: juicyfruit, dingto, lukasstockner97, brecht
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D1928
While other borders are more like a toggle, it is an intrinsic behavior
of those operators. Render Border is intrinsicly split into two operators
and trying to expose it as a toggle will end up with rather confusing
situation when shortcut listed in the menu changes depending on the
context.
Solves following issues:
- Quite reasonable amount of paranoid warnings were solved by an upstream
- Upstream seems to have all fixes needed for FreeBSD and OSX already
- Brings all fixes and such from upstream
This commit adds a new `wm_files_link.c` which contains everything related to append/link code,
moved from `wm_operators.c` (rather small currently, but will expand quite a bit with future reload & asset works).
It also moves all load/save .bland files (and related userpref/startup stuff) from `wm_operators.c`
to `wm_files.c` (some helper funcs were already there).
This also makes `wm_operators.c` significantly lighter.
Cursor is now set to standard arrow-cursor when opening menu, and reset to previous one when closing it. Previously it just stayed as it was before, e.g. the edit mode cross-cursor stayed active even if a menu was opened.
Fixes T48192.
- Fix wrong current sample reported in the log
- Also includes fix for progressive refine log
- Explicitly print to the stdout that resumable render is enabled
- Print error message and abort when passing wrong values for the
resumable render. Never waste someone's compute power for wrong
render!
Fixes T48185: Cycles resumable num chunks breaks sample counter
Holes with flat surfaces could have their edges dissolved causing degenerate faces.
Now check that collapsing a vertices isn't creating self-overlapping faces.
This is to prevent situations such as when the camera gets very close to a mesh
and causes it to be tessellated into an excessive amount of micropolygons. In
REYES this is known as the eye-splits problem.
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D1922
This commit makes disabled/inactive color swatches look more obviously disabled
by halving their alpha values. It is necessary because the results of
ui_widget_color_disabled() (which is usually used when disabling widgets)
get overwritten by the color filling code here.
To avoid confusion, the checkerboard background (when the color genuinely has
alpha) will still only show in those cases. That is, when disabled, color swatches
won't show the checkerboard unless the color actually has an alpha component if it
wasn't disabled.
As soon as you started trying to smooth a stroke, the thickness of the stroke
would quickly drop right down to near zero, if "affect pressure" was enabled.
This step got accidentally missed when restoring the affect pressure functionality to
the stroke smoothing brush code, following cleanups from the stroke quality patch.
This makes it easier to control overall dicing rate without having to tweak
every object. The preview rate makes viewport editing more interactive. The
default preview rate of 8 is roughly 64 times faster for some operations.
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D1919
Instead of treating Fermi GPU limits as default,
and overriding them for other devices,
we now nicely set them for each platform.
* Due to setting values for all platforms,
we don't have to offset the slot id for OpenCL anymore,
as the image manager wont add float images for OpenCL now.
* Bugfix: TEX_NUM_FLOAT_IMAGES was always 5, even for CPU,
so the code in svm_image.h clamped float textures with alpha on CPU after the 5th slot.
Reviewers: #cycles, brecht
Reviewed By: #cycles, brecht
Subscribers: brecht
Differential Revision: https://developer.blender.org/D1925
* Reshuffled code for existing "View Frame" implementations, and removed leftover
comment from some of the the copy-and-paste used to build it.
* Added support for this operator in the NLA and Timeline
* Knock out "Python" FModifier entry - It really hasn't been coded yet!
* Add icon for "Match Indices" driver eyedropper mode. It should help provide
a bit more of a hint of what it does, but it also doesn't look quite as nice now.
This commit aims to streamline the driver variables layout a bit
* Each variable type now has an icon. (The loc diff and rot diff ones are placeholders,
which could deserve something better/dedicated if we continue to use this)
* Instead of taking up an entire row, the variable type dropdown now only shows
an icon, and is located before the variable name field.
Feedback wanted: Is this more/less confusing than it was?
Seems particular CUDA implementations has some precision issues,
which made integer coordinate (which was expected to always be
positive) to go negative.
Simple fix: release lock earlier.
Reduces spatial split build time from 96 to 53sec on the Bunny.blend
(using studio Intel for benchmark).
NOTE: Timing difference is not that spectacular when comparing numbers
with builds before memory optimization, but even then it's about 20%
faster build.
Simple solution is to get rid of this stupid `rna_TextureSlot_brush_update` func, and do what it was doing directly
in BR case of `rna_TextureSlot_update`. That way, all updates to brush texture will invalidate correctly the overlay.
Note that this double definition of texture pointer for brushes looks rather fishy, am really not convinced this helps
keeping things simple and clear at all (what happened here is that texture selector in image paint editor was using
'texture' member of brush itself, which was performing correct RNA update, while context-based brush texture selector
from buttons space was using 'texture' member of the TextureSlot struct, which had default texture update callback
which was not invalidating overlay).
- Access with Shift-LMB or from the 'Create' toolbar tab.
- Uses curve fitting for bezier curves, with error and corner angle options.
- Optional tablet pressure to curve radius mapping.
- Depth can use the cursor or optionally draw onto the surface,
for the entire stroke or using the stroke start.
- Stroke plane can optionally be perpendicular to, or aligned to the surface normal.
- Optional radius tapering and for start/end points.
- Supports operator redo and calling from Python.
This will be used for calculating bezier curves from freehand drawing,
may be used for other areas too.
Original code from GraphicsGems, 1990 (FitCurve.c),
with updates from OpenToonz, under 3 clause BSD license.
with own minor modifications for integration with Blender:
- support adding extra custom-data.
- improved handle clamping.
Similar to velocity, it was kind of supported by the mesh manager but
was missing a code in BlenderSession to get actual values.
In Cycles Heat is an attribute which goes from -1 to 1, where -1 is
the coldest ever temperature, 1 is the hottest ever one.
This was a feature request from a few years back (IIRC from ZanQdo?) to make it
easier to reuse one set of driver variables across several different drivers.
Dev Notes:
* Finally it's done! All that trouble for two little buttons.
* Grr... cmake... grrr!
* Use an RNA enum callback for NLA FModifier types too
* Cleaned up tooltips for these operators in Graph and NLA editors
* Fix update notifiers after adding FModifiers in the Graph Editor
This brings the UI here more in line with the Constraints and Modifiers UI's.
TODO:
* The tooltips/descriptions on the operators now need fixing, as we can no longer
use custom tooltips when defining these buttons.
* The operators need to be fixed to only operate on the active data
This change means that when going from having no FCurves active to having
an active FCurve, the properties panels will not get stuck with only showing
the "View" tab, which is not that useful a lot of the time.
This was only visible on systems with lots of threads and root of the issue
was that we've been pre-allocating too much memory for all the threads.
Now we only pre-allocate data for the main thread and rest of the threads
does allocation on-demand.
This brings down memory usage from 36Gig to 6.9Gig when building spatial
split for the Bunny.blend file on our Intel beast.
Originally regression was happened by the threaded spacial split builder
commit.
Couple of things:
- No need to use string streams to format the version string,
we can do it at compile time and don't bother with anything
at runtime.
- Function declaration was wring and would have caused linking
conflicts in cases when util_version.h was included from
multiple places.
We should have an utility function to get Cycles version so
applications which are linked to Cycles dynamically can query
the version, but that can't be done as an inlined function in
header and would need to be a function properly exported to a
global symbol table (aka, be implemented in a .cpp file).
Now Cycles has its own versioning, that is mainly interesting for external projects, which integrate the engine.
We start with version 1.7.0. Reasons for that:
* The engine is too mature for a 1.0 release.
* We assume that Cycles inside of Blender 2.61 was version 0.1. We count upwards in 0.1 steps, therefore Cycles inside of Blender 2.77 would be 1.7.
We use a common versioning scheme here, with 3 decimals for the major, minor and patch level.
At the moment cycles --version can be used to display the version, easy to parse for external projects. The info will be added to the UI later aswell.
This way we prevent cracks in the model due to discontinuous normals, by using
smooth normals for displacement instead of always getting flat normals after
linear subdivision.
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D1916
This is a regression since 4310128 which is cased by really bad logic:
there might be dependencies between dupli-objects, which means _if_ we
really want to ensure derived mesh on dupli creation we have to do it
before any matrix is overwritten.
I'm not sure if such derived mesh trick is really the only way to go
without major refactor, but seems simple fix for now will do it.
For some reason other parts of blender importer were assuming velocity
is supported, but actual loader was not aware of that.
Fixes T48064: Adding velocity attribute crashes render
The header which was put to the CMake project is not by any means used
during compilation, it is only used when manually updating CUDA wrangler
from CUDA headers.
This is only the UI change, there is no underlying code change in this commit.
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D1910
NOTE: this is only the first of many patches towards completing the subdivison
and displacement system in Cycles. These patches will be reviewed and committed
one by one over the coming weeks.
Reviewed By: brecht, sergey
Differential Revision: https://developer.blender.org/D1909
There are couple of reasons:
- Volume shader on hair does behave really weird anyway and it's
not something considered a bug really.
- Volume BVH traversal were only used by camera-in-volume check,
which doesn't really make sense to take hair into account since
it'll be rendered wrong anyway.
Such a removal makes both code easier to extend further (as in,
no need to worry about those traversal for hair bvh) and also
reduces stress on GPU compilers.
This prevents twisted quads from self-intersecting.
This change makes the duplicate surface match the first vertex in the face,
so the diagonal indices match in the face copy.
Previously the logic manager was used as a global variable for SCA_ILogicBrick::m_sCurrentLogicManager,
this request to always update it before run any python script and allow call function like
ConvertPythonTo[GameObject/Mesh]. The bug showed in T48071 is that as exepted the global
m_sCurrentLogicManager is not updated with the proper scene logic manager.
Instead of trying to fix it by updating the logic manager everywhere and wait next bug report to add
a similar line. The following patch propose a different way:
- Every logic brick now contain its logic manager to SCA_ILogicBrick::m_logicManager, this value is
set and get by SCA_ILogicBrick::[Set/Get]LogicManager, It's initialized from blender conversion and
scene merging.
- Function ConvertPythonTo[GameObject/mesh] now take as first argument the logic manager to find name
coresponding object or mesh. Only ConvertPythonToCamera doesn't do that because it uses the
KX_Scene::FindCamera function.
Reviewers: moguri
Differential Revision: https://developer.blender.org/D1913
When the Moon is full it was possible to have a dead-lock in task
scheduler's exit() method.
Similar problem was fixed in Blender's task scheduler 3 years ago
in bae2a2c.
Previous fix made another issue even more visible, leading to +1 particle on first vert
and none on last one. This commit should fix both original and new issues.
The patch contains an implementation of the wide lines and the line stipple that is necessary for OpenGL upgrade.
For the implementation I have chosen the geometry shader because it required minimum changes for the wrapper calls and such implementation is the best for the "basic shader" architecture.
There are few shortcomings that can be corrected in future. They all are related to the fact that the lines in one strip are not connected with each other. So the stipple pattern is not continuous on the common vertex of two lines. There is also no continuity of form (each line is an independent rectangular).
But the advantage is that even outdated glBegin, glVertex work! Though with the above restrictions.
Continuity of form and stipple can be implemented with additional attributes, and it will require more changes in calls.
At the moment, the patch replaces calls for some "gestures". It works satisfactorily for "cross" or "rectangular" and not so good for "lasso" and "circle" due to the above-mentioned shortcomings.
Don't forget to set USE_GLSL to true for testing.
Alexander Romanov (Blend4Web Team)
Reviewers: merwin, brecht
Reviewed By: merwin, brecht
Subscribers: aligorith, Evgeny_Rodygin, AlexKowel, yurikovelenov
Differential Revision: https://developer.blender.org/D1880
This patch depends on D1747, which upgrades the Recast version. It exposes the new Recast partitioning methods in the navmesh generation.
Reviewers: campbellbarton, moguri
Reviewed By: moguri
Projects: #bf_blender
Differential Revision: https://developer.blender.org/D1748
The version of Recast that Blender ships with is from 2009. This patch updates the Recast version to the latest version, 1.5.0. The Detour version remains untouched.
Reviewers: campbellbarton, moguri
Reviewed By: moguri
Projects: #bf_blender
Differential Revision: https://developer.blender.org/D1747
The addressed issue is a regression from Blender 2.75, after the internal
switch from double to single precision floating-point numbers in the
Freestyle code base. Face normal calculations require the higher
precision during the computations, even though the results can be stored
as single precision numbers.
Support for driver variables that don't resolve to numbers, eg:
objects, bones, curves... etc.
Without this, Python expressions to access this data needed to use an absolute path from `bpy.data`,
however this is inconvenient, breaks easily (based on naming) and wouldn't set the dependencies correctly.
This fixes reported issue, but there could be more of those hidden in Freestyle code,
it did not handle user refcount at all (and the fact it by-passes BKE in some places
for efficiency does not help here).
Note that this should be relatively harmless, since freestyle uses own Main during
render, so everything is cleaned up in the end in any case... But better try to
handle IDs correctly here too. :)
The title actually covers it all, This commit exploits all the work
being done in previous changes to make it possible to build spatial
splits in threads.
Works quite nicely, but has a downside of some extra memory usage.
In practice it doesn't seem to be a huge problem and that we can
always look into later if it becomes a real showstopper.
In practice it shows some nice speedup:
- BMW27 scene takes 3 now (used to be 4)
- Agent shot takes 5 sec (used to be 80)
Such non-linear speedup is most likely coming from much less amount
of heap re-allocations. A a downside, there's a bit of extra memory
used by BVH arrays. From the tests amount of extra memory is below
0.001% so far, so it's not that bad at all.
Reviewers: brecht, juicyfruit, dingto, lukasstockner97
Differential Revision: https://developer.blender.org/D1820
Policy here is a bit more complicated, if tree becomes too deep we're
forced to create a leaf node and size of that leaf wouldn't be so well
predicted, which means it's quite tricky to use single stack array for
that.
Made it more official feature that StackAllocator will fall-back to
heap when running out of stack memory.
It's still much better than always using heap allocator.
We had per-tree statistics already, but it's a bit tricky to see overall
time because trees could be building in parallel.
In fact, we can now print statistics for any TaskPool.
Currently they're staying at 1 (actual size over capacity), but we
will be changing it quite soon in order to avoid having too much
memory re-allocation happening at a BVH build time and will be
playing with different policies for that.
Note that we may want to review that some day, we have quite a bit of ID types here that are
'blend file' view only, might be worth splitting the check based on this.
In some files stack memory was overruning the pre-allocated stack.
Perhaps we should fall-back to a hep-allocated stack so release builds
don't crash in works case but just becoming slower.
That might consider a bit more objects to be considered deform modified,
but it covers common case of using taper object without require of doing
recursive checks.
In worst case it'll be just some extra synchronization time, no render
time difference will happen for false-positive because of extra checks
happening in Cycles.
The improved Hosek / Wilkie model was added during my GSoC 2013 and the default since then.
The older model was kinda kept for compatibility, but after more than 2 years it's time to remove it.
The Hosek / Wilkie model is more realistic anyway, and people who really want a day / night transition can mix the Sky Shader with another one (e.g. color) and fade between the two.
There are in fact some missing parts to it (Split BVH builder should
be creating bins from result of Object Split constructor).
Doable, but need to quickly fix issue for the studio here, easier to
revert for now.
Main use case of this ID will be to emulate TLS which otherwise
would require having some platform-specific implementations which
is not always really optimal.
See notes about the argument in util_task.h.
They were never running anyway since there's no checkout of
tests folder. We should either make tweaks to buildbot master
to make it to send cloning commands for tests folder and keep
tests working (currently some of them are broken).
Currently unused, but will be handy for an upcoming changes.
It'll also be nice to be able to do scoped_lock() for both
Mutex and Spin, but currently it's not really easy to do,
need some changes in typedefs and such, will happen as a
separate commit.
This has following advantages:
- Localizes all the run-time storage into a single structure,
which could easily be extended further.
- Storage could be created per-thread, so once builder is
threaded we wouldn't have any conflicts between threads.
- Global nature of the storage avoids memory re-allocation
on the runtime, keeping builder as fast as possible.
Currently it's just API changes, which don't affect user at all.
Uses new StackAllocator from util_stack_allocator. Some tweaks to the stack
storage size are possible, read notes in the code about this.
At this point we might want to rename allocator files to util_allocator_foo.c,
so the stay nicely grouped in the folder.
When non-random, particle distribution used a small start offset (to avoid
zero-weight faces), which is fine with "continuous" entities like faces, but not
for discrete ones like vertices - in that case it was generating some undesired
"jump" over a few verts in case step was small enough
(i.e. total number of verts/particles was big enough).
Point-cached particles (those using simulations) would not update at all outside of
first frame, due to PSYS_RECALC_RESET flag being ingnored in `system_step()`...
For some mysterious reasons, udate is still non-fully functional outside of startframe
(e.g. changing face distribution between random and jittered), but at least when choosing
'Vertices' you get particles from verts and not faces!
This feature is also known by the name Samples Offset, which allows
artists to render animation with given amount of samples N, but then
render more samples, starting from N and ending with M (where M > N)
and merge renders together as if they rendered exactly M samples.
Surely such effect could be achieved by changing Seed variable, but
that has possible issues with correlation artifacts and requiring to
manually deal with per render layer samples and such.
While we can't support all possible renderfarm-related features in
Cycles it's nice to support really commonly used stuff.
Here's a command how to run Blender with the new feature enabled:
blender -- --cycles-resumable-num-chunks 24 --cycles-resumable-current-chunk 2
This command will divide samples range in 24 parts and render
range #2 (chunk number is 1-based).
This feature might be changed a bit after we'll do some tests here
in the studio with it.
This reverts commit 2bd3acf7cd.
Commit is valid in theory - but in practice, it means nearly all edited hair systems
would need to be re-created from scratch... Not nice, so better revert and note in code
that particle distribution is ugly and does not respect 'use modifier stack' option in most cases.
As noted by Antonio and Daniel, performing smoothing on the pressure values
for new strokes doesn't work that great. From an artist POV, it is apparently
"very strange that line thickness shrink suddenly after drawing it".
* This includes the "manually create" modes, which correspond to the previous behaviour
for setting up drivers. This is necessary when the current screen layout is not
well suited to having multiple property editors open (e.g. small screen or heavily
subdivided screen).
* Only the modes relevant for the current property type (i.e. array vs single)
will be shown
* The "Add Driver" entries in the RMB context menu have now been replaced by a
submenu which will list all the available mapping types.
* NOTE: The code for the ANIM_OT_button_driver_add() operator is perhaps a bit hairy.
However, it currently allows us to have the desired behaviour. It can always get
cleaned up later though.
Caused by rBc24be7ec6e5.
Before rBc24be7ec6e5, wm_handlers_do always called handlers a second time with event value KM_PRESS in case of a double click. After it, this was only the case for non-LEFTMOUSE events.
Since ui_popup_handler (almost) always returned WM_UI_HANDLER_BREAK, the second handler iteration with KM_PRESS wouldn't run. This fix just makes sure we return WM_UI_HANDLER_CONTINUE for double click events instead (causing second iteration to run).
This never really worked that well, and often ended up being far too strong
to be of practical use. The new options do similar things, but with greater
control, so removing this old method now.
On second thought, these might be better names for these... I'm still not 100%
happy with these, but they will do for now.
(Best results currently seem to be with smooth 0.7, and subdivision steps 1 or 2)
Assuming that this subdivision + smoothing mostly only happens at the end of a stroke
anyway, enforcing this max-points limit in some cases was resulting in strokes that
weren't quite getting subdivided properly as the others.
Improve the quality of current grease pencil strokes adding a new dynamic smooth and subdivision. The level of smooth and subdivide can be adjusted using UI parameters. These options are disabled by default in order to keep the grease pencil stroke compatible with any existing add-on.
Both parameters are defined at layer level.
Reviewers: aligorith
Differential Revision: https://developer.blender.org/D1866
Unexpectedly found out what was going wrong here. If a file was saved with a filebrowser open, we searched for the channel region in the wrong list (see 'ListBase *regionbase = (sl == sa->spacedata.first) ? &sa->regionbase : &sl->regionbase;').
Minor annoyance is that I had to move the loookup to the 2.77.1 version patch now.
When linking a rotation property to a non-rotation property (going in either
direction - i.e. rot -> normal, normal -> rot), the driver expression will
now be set so that the new drivers behave as expected (i.e. you get the values
you see, instead of "weird" values that seem several orders of magnitude off).
This may not be that great for everyone (i.e. the rare users out there who
actually like to look at their rotations in radians), but they usually know
what they're doing anyway, so this will be easy to correct.
Although it isn't currently exposed, this allows for the old behaviour,
where an "empty" driver was added (without any target assigned yet).
For this reason, it's also referred to as the "Manual" mode.
There are also some attempts at improving the tooltips + names for the
other modes (again, not shown anywhere yet)
Looks like some distro still provide llvm-3.4-devel, while no more clang-3.4.
Since clang depends on llvm of same version, check clang only should ensure
us we also have matvhing llvm... *sigh*
Follow same logic in `psys_render_restore` as in `psys_render_set` - if hair and
display percentage is not 100%, we have to recompute particles...
With regular 'emitter' particles just hiding some is fine (though using random here
will not give a precise proportion...).
readfile.c would increment object usercount in three places, where it should not.
Remember kids: Objects are **only** refcounted by Scene's bases, and Object->proxy!
This commit brings some long requested improvements to the workflow for setting up
drivers, which should make it easier and faster to set up new drivers in a more
interactive fashion.
The new workflow is as follows:
1) Hover over the property (e.g. "Lamp Energy" or "Y Location") or properties ("Rotation")
you wish to add drivers to. We'll refer to this as the "destination"
2) Ctrl-D to active the new "Add Drivers" eyedropper
3) Click on the property you want to use as the source/target. The property under the
mouse will be used to drive the property you invoked Ctrl-D on.
For example, to drive the X, Y, and Z location of the Cube using the Y Location of the Lamp,
hover over any of the X/Y/Z location buttons, hit Ctrl-D, then click on the Y-Location
button of the Lamp object. Drivers will be added to the X, Y, and Z Location properties
of the Cube; each driver will have a single variable, which uses the Y-Location Transform
Channel of the Lamp.
Tips:
- Transform properties will automatically create "Transform Channel" driver variables.
Everything else will use "Single Property" ones
- Due to the way that Blender's UI Context works, you'll need two Properties Panel instances
open (and to have pinned one of the two to show the properties for the unselected
object). It's slightly clunky, but necessary for implementing a workflow like this,
as the UI cannot be manipulated while using eyedroppers to pick data.
- The eyedropper operator implemented here actually has three modes of operation.
1) The "1-N" (one to many) mode is the default used for Ctrl-D, and "Add Driver to All"
in the RMB Menu. This is the behaviour described above.
2) There's also a "1-1" (one to one) mode that is used for the "Add Single Driver" in the
RMB Menu.
3) Finally, there's the "N-N" mode (many to many), which isn't currently exposed.
The point of this is to allow mapping XYZ to XYZ elementwise (i.e. direct copying)
which is useful for things like locations, rotations, scaling, and colours.
Implementation Notes:
- The bulk of the driver adding logic is in editors/animation/drivers.c, where most of
the Driver UI operators and tools are defined
- The property eyedropper code is in interface_eyedropper.c along with all the other
eyedroppers (even though they don't share much actual code in common). However, this
turns out to be necessary, as we can't get access to many of the low-level buttons API's
otherwise.
Todo:
- It may be necessary to restore a way to access the old behaviour (i.e. "manual setup")
in case it is not practical to immediately pick a property.
- Other things to investigate here include extra hotkeys (e.g. Ctrl-Shift-D for Add Single?),
and to expose the N-N mode.
- Other things we could try include interactively applying scaling factors, picking
multiple targets (e.g. for location difference and rotation difference drivers),
and/or other ways of using these property picking methods.
Just using DKEY for this was resulting in too many drivers being accidentally created
by users drawing using Grease Pencil. This change also works better with the following
workflow changes to the driver setup process.
Disclaimer: this is tentative fix, seems to be working but you never know with particles.
If issue arise, we'll just revert this commit and hide 'use modifier stack' option
for grid distribution instead.
This commit also try to make choice of which dm to use during distribution more generic
(deduplication of code), again hopping not to break anything. :P
Steps to reproduce were:
* Open default video editing layout
* Drag down or click '+' icon in upper right corner of sequencer view
* A preview region appears, even though the view type is not set to sequencer-preview split mode
Another bug that has been there since ages :/ Reported by @Blendify via IRC, thanks!
This seems quite useful for the development, so you don't need to wait
all the kernels to be re-compiled when working on a new feature, which
speeds up re-iteration.
Marked as an advanced option, so if it doesn't work so well in practice
it's safe to revert anyway.
This commit makes it so casting subsurface rays will totally ignore all
the BVH nodes and primitives which do not belong to a current object,
making it much simpler traversal code and reduces number of intersection
tests.
Reviewers: brecht, juicyfruit, dingto, lukasstockner97
Differential Revision: https://developer.blender.org/D1823
Steps to reproduce were:
* Go to edit mode, select geometry
* Spacebar -> search for bevel OP
* Activate *using enter key*
* Immediately stops when releasing enter key
To fix we simply make sure the initial release event is ignored. For mouse events that wasn't an issue since activating an element is done on key release in menus. An alternative fix would be to do the same for enter key, but that's more risky. Also, checking event value is highly recommended anyway.
While it was indirectly disabled already, it's handy to use the config
as template and enable certain features. In case of Cycles enabling it
would also enable OSL which is not very expected.
This commit:
* Fixes bad handling of 'stop iteration' (by adding a status flag, so that we can actually
stop in helper functions too, and jumping to a finalize label instead of raw return, to
allow propper clean up).
* Adds optional recursion into 'ID tree' - callback can also decide to exclude current id_pp
from recursion. Note that this implies 'readonly', modifying IDs while recursing is not
something we want to support!
* Changes callback signature/expected behavior: return behavior is now handled through flags,
and 'parent' ID of id_pp is also passed (since it may not always be root id anymore).
Reviewers: sergey, campbellbarton
Differential Revision: https://developer.blender.org/D1869
This patch adds support for coloring point density textures based on several mesh vertex attributes.
* Vertex Color: Use a vertex color layer for coloring the point density texture
* Vertex Weight: Use a weights from a vertex group as intensity values. (for Blender Render engine the additional color band is used)
* Vertex Normals: Use object-space vertex normals as RGB values.
The vertex color source enum is stored separately from the particle color source, to avoid invalid values when switching.
Note that vertex colors are technically "corner colors" (MLoop), so each vertex can have as many colors as faces it is part of.
For the purpose of point density the mloop colors are simply averaged, which is physically plausible because corners can be viewed
as multiple points in the same location.
Now, when trying to insert a keyframe on a driven property (using IKEY, or with
autokeying enabled), the keyframes will get created on the Driver's F-Curve
(instead of creating a new FCurve that goes into the active action, but will never
do anything). Furthermore, the x-value of the new keyframe will be the current
result of the driver expression.
Why/Motivations:
This way, it becomes easier to create corrective drivers, as you can position all
the targets the driver depends on, then adjust the driver value until it does what
you need, and then you keyframe that value to bake it into the Driver F-Curve
(in effect, "training" the computer how to behave in that case).
Usage Notes:
* In practice, that particular workflow is still quite clunky to achieve, due to some
quirks of how the driver system and the UI widgets interact. Specifically, you'll
need to disable/mute the driver before trying to edit the setting (to prevent the
driver from immediately resetting the value - before even autokey fires!). However,
if you're using the Graph Editor to preview/monitor/manage the keying process, you'll
then want to re-enable the driver before changing the targets, so that you can see
how much of a change you'll want to be applying!
* The warning about editing driver values may need to be disabled or selectively
knocked out. I had it disabled while testing this functionality, but it's actually
harmless in its current state (if just a bit annoying).
When attempting to change a driver variable name to an "invalid" name,
an indicator will now be shown beside the offending variable name.
Clicking on this icon will show a popup which provides more information
about why the variable name cannot be used.
Reasons that it knows about are:
1) Starts with number
2) Has a dot
3) Has a space
4) Starts with or contains a special character
5) Starts with an underscore (Python does allow this, but it's bad practice,
and makes checking security of drivers harder)
6) Is a reserved Python keyword
While most scripted expressions are quite short (e.g. "var" or "frame + 1"),
those that are autogenerated by scripts (to take into account a wide range of factors)
may require a bit more headroom to operate in.
By doubling the maximum length here, we now make it possible to combine up to 38
3-letter variables in some linear combination (with all factors expressed to 4 dp).
That should be more than enough for anyone or anything for quite a while!
Unfortunately, this does increase memory consumption for drivers across the board.
White the config itself could be loaded this way, lookup tables can not. Additionally,
that's not really clear how to solve the issue with search path which is multi-byte
only in the API.
Reverting for further investigation.
This reverts commit ab4307aa08.
Thanks to D1080 by @rockets, I've now been able to easily implement the
ability to type multiple word snippets/partial words into the text filter
field (in the Animation Editors), and have it filter the channels which
contain just some of those parts (instead of having to match everything).
For example, the following search strings will now work:
* "loc rot" or "lo ro" will now filter all location and rotation FCurves
* "col loc" will filter all location and color FCurves
* "scale" will also work as before to filter all scale FCurves
But, the following will not work:
* "lc rt" will NOT filter all location and rotation, as the fuzzy search only
breaks down the search string based on whitespace placement
By default, this is not enabled when using name filtering (i.e. magnifying glass is checked,
and some filtering text is specified). Instead, you need to enable the "AZ" toggle beside
the name field. This fuzzy matching is not enabled by default as it could end up being
quite a bit slower on really heavy scenes. (There are probably some optimisation
opportunities, but that's only a future option if someone really needs it)
This is a mix of regression and old unsupported configuration.
Regression was caused by some checks added on Blender side which was
checking whether python function returned error or not. This made it
impossible to enable Cycles when running from a file path which can't
be encoded with MBCS codepage.
Non-regression issue was that it wasn't possible to use pre-compiled
CUDA kernels when running from a path with non-ascii multi-byte
characters.
This commit fixes regression and CUDA parts, but OSL still can't be
used from a non-ascii location because it uses non-widechar API to
work with file paths by the looks of it. Not sure we can solve this
just from our side by using some codepage trick (UTF-16?) since even
oslc fails to compile shader when there are non-ascii characters in
the path.
* As requested by venomgfx, rotations now only display to 2 decimal places (more than
enough for rotations in degrees)
* Also, made it so that the "factor" types now show 3 dp (previously only 2, which isn't
really enough when dealing with values from 0.0 - 1.0)
The radial control operator is commonly used for setting sculpt brush Size and Strength
settings. However, when using it, it can often be difficult to remember which hotkey
is responsible for which setting, meaning that you can often accidentally modify the
wrong setting without knowing it (or even what new value you've set it to).
This commit makes it so that the property name and the current value are always shown
on the header. Previously this info was only shown when the doing numeric input.
(It also fixes a bug where the numeric input print would hang around even after it
wasn't being used).
Previously it would average all points under the brush, giving an off feedback loop that often depended on the
direction of the brush stroke, giving not-very-useful smearing effect.
Now blend brush smooths between connected edges.
rB5b3af3dd made the poll function here slightly too laxist.
To be backported to 2.77 should we make an 'a' release.
Reviewed By: mont29
Differential Revision: https://developer.blender.org/D1861
This patch implements Mirror influence for environment textures. Approach matches the one from BI.
{F281871}
See the video https://youtu.be/BskgCv6dcIE
Example: {F281876}
Alexander (Blend4Web Team)
Reviewers: campbellbarton, merwin, brecht
Reviewed By: brecht
Subscribers: TwisterGE, blueprintrandom, youle, Evgeny_Rodygin, AlexKowel, yurikovelenov
Differential Revision: https://developer.blender.org/D1786
Using 32nd bit of integer with our int RNA prop is not a good idea, it does not
really support unsigned int even with PROP_UNSIGNED type...
Most likely this has been unveiled by recent work over property clamping in RNA.
Not critical, but should be backported should we make an 'a' release.
When all the names in Blender were updated to go from having 32 characters to 64,
the "pchan_name" field in the DriverTarget struct was missed. This meant that if you
had a bone with a suitably long name, it would have been impossible to use values from this
bone in a driver.
The only reason this probably hasn't been discovered yet is because the standard/old depsgraph
doesn't work that well for bones in the same armatures, so setups where this may have
occurred are not that common.
To prevent this problem from happening in future, I've added the standard "MAX_ID_NAME-2"
comment to the comments on each line likely to be impacted by such changes, making it
easier for whoever does a search and replace in future :)
While this isn't essential, accessing this setting required navigating to each scene and using render menu.
Expose in sequencer UI for more convenient access.
Something weird, seems some buffers _might_ be shared between codec context
and frame, which is quite weird. Could be a bug in FFmpeg or could be wrong
API usage somewhere else..
Was possible to do Alt-RMB on a strips handle - which only make the strip active (but didn't select).
This isn't really useful, so just select the strip and its handles in this case.
In fact, some areas in FFmpeg might require alignment up to 32, so now we
make temporary conversion in an aligned frame. We only do this if width is
not aligned to 32.
As suggested by Sergey, do not register those anymore, this way we keep undo step,
but user cannot 'redo' them (does not work, since cached DM in particle modifier data
is not yet re-created by depsgraph update after undo when operator is redone).
UI now has two buttons, one to (dic)connect current psys, the other to (dis)connect all.
Also fixed similar issue with Connect Hair op.
This has several benefits:
* User count remains coherent, regardless of the order in which you use 'user_one' & real refcounting users
(i.e. if you add to group, and then link in scene, or the reverse, you now always get same final user count).
* Avoids the need to check for potential 'user_one' extra user in count in several places in code (e.g. when
making IDs single users...).
* Users won't wonder why they cannot make 'single user' and ID even though its user count shows '2'!
* readfile.c now always uses code from BKE's library.c when modifying id->us. Which means we can consider
(asside from assignment during initialization) that id->us is read-only outside of library.c context.
Note that this commit reverts previous one (rB6b1d77a8052b) - please **do not** backport this one in 2.77.
Similar cause as in T47482, we used to have poor handling of 'user_one' cases of ID usage,
leading to inconsistent behavior depending on order of operations e.g.
Here, was object used by a group but not linked in any scene - once linked in scene,
their usercount would be 2, leading to 'making single copy', when it's actually not needed.
We now have better control here, so let's use it!
Note that other ID 'make single user' code will likely need similar fix (Images, etc.).
Safe to be backported to 2.77.
Problem was, during initialization of boids particles in `dynamics_step()`,
psys of target objects was not obtained with generic `psys_get_target_system()`
as later in code, which could lead to some uninitialized `psys->tree` usage...
Think it's safe enough for 2.77, though not a regression.
* Don't show name of keyframe type in timeline header. It's getting a bit too crowded
for that, so just show the keyframe icon. Hopefully that will be enough.
* Tweaked the tooltip and label to make it clearer what the setting is for
The new "default keyframe type" dropdown on the timeline header
(and also the "Keyframe Type" operator/properties in other places)
now has procedurally generated icons which reflect what that keyframe
type will look like when rendered in the Dope Sheet.
This was achieved using the ancient "VICON" (vector icon) stuff
that's lurking around in the dark parts of UI code. From memory,
the only other things that use (or used to use) this stuff included
some of the triangle icons for some dropdown buttons, or something
like that.
Notes:
* Theme colour changes are reflected immediately by these icons.
This is possible because they are all drawn procedurally
* These icons scale with the DPI setting. I manually guessed the size of
these icons. They can be adjusted further if needed.
* I've documented the steps for adding voodoo icons like this on the wiki
(http://wiki.blender.org/index.php/Dev:2.7/Source/Checklists/Vector_Icon)
* It's true that the rendering of these keyframes doesn't quite fit the rest
of the icons in the UI. However, since we're just leveraging the standard
keyframe drawing methods (to avoid discreptancies between the two), we'll
leave it as such for now. Maybe later we can consider blending in a bit of
the glossy keyframe icons in the Icon Sheet?
To make it easier for animators working in a multipass pose-to-pose workflow
when inserting breakdown keyframes and so forth, it is now possible to specify
the "type" of keyframe being created (i.e. the colour of the keyframe, when drawn
in the Dope Sheet).
Usage:
1) Choose the type of keyframe ("Keyframe", "Breakdown", "Extreme", etc.) from
the new dropdown located between the AutoKeying and KeyingSet widgets on the
timeline header.
2) Insert keyframes
3) Rejoyce that your newly created keyframes have now been coloured for you already
in the DopeSheet.
Todo:
* Look into a way of using the actual keyframe colours (from the theme) for the icons
of these types.
This patch adds a new `falloff_type` ('Inverse Coefficients') for Lamps in
Blender-Internal and GLSL.
The current falloff modes use a formula like this inverse-square one:
`I = E × (D^2 / (D^2 + Q × r^2))`
While such a formula is simple for 3D-artists to use, it's algebraically
cumbersome to work with. Game-designers authoring their own shaders
could benefit much more by having direct control of falloff-coefficients:
`I = E × (1.0 / (coefC + coefL × r + coefQ × r^2))`
In this mode, the `distance` parameter is unused (except for 'Sphere'
mode); instead relying on the designer to mathematically-model the
falloff-behavior.
The UI has been patched like so:
{F153843}
Reviewers: brecht, psy-fi
Reviewed By: psy-fi
Subscribers: brita_, antidote, campbellbarton, psy-fi
Differential Revision: https://developer.blender.org/D1194
We've got pixel-wide world-space derivatives which we can use in the
perspective camera sampling. This allows to get rid of two calls to
transform_direction() function.
In theory we can save two transform_perspective() calls if we'll also
save pre-calculated camera-space dx/dy.
Previously each call of this function was followed by a normaliztion, now it
is done in the function itself with an according note around the function.
Static schedule was responsible here...
Also, made a minor optimization in case adaptative (auto) subframes are enabled,
gives a few percent of speedup here.
At this point it's totally unclear why we're ignoring aperture and and rolling shutter
now for derivatives calculation but do not ignore direction change caused by stereo.
This is a bit tricky, find_package(Git) was only used to check if git is installed
and actual git command was assumed to be in the PATH (while ideally it should have
been GIT_COMMAND variable).
This commit makes re-compile output cleaner, especially when using Ninja.
Adding proper GIT_COMMAND to buildinfo.cmake is also possible via argument.
Reviewers: campbellbarton
Reviewed By: campbellbarton
Differential Revision: https://developer.blender.org/D1847
Was introduced by recent optimization. Not really sure derivatives are
intended to work like this, but better to stick to what Dalai had
originally for now.
Buffer params needs to know camera's border, otherwise it'll create full buffer.
There might be some issues with stereo camera still, but in worst case it'll
only update camera twice as far as i can tell. Not ideal, but better than no
border render at all.
This allows us to verify certificates of HTTPS connections, which is
mandatory for logins like on Blender ID.
Reviewers: campbellbarton
Differential Revision: https://developer.blender.org/D1845
This is a new option for panorama cameras to render
stereo that can be used in virtual reality devices
The option is available under the camera panel when Multi-View is enabled (Views option in the Render Layers panel)
Known limitations:
------------------
* Parallel convergence is not supported (you need to set a convergence distance really high to simulate this effect).
* Pivot was not supposed to affect the render but it does, this has to be looked at, but for now set it to CENTER
* Derivatives in perspective camera need to be pre-computed or we shuld get rid of kcam->dx/dy (Sergey words, I don't fully grasp the implication shere)
* This works in perspective mode and in panorama mode. However, for fully benefit from this effect in perspective mode you need to render a cube map. (there is an addon for this, developed separately, perhaps we could include it in master).
* We have no support for "neck distance" at the moment. This is supposed to help with objects at short distances.
* We have no support to rotate the "Up Axis" of the stereo plane. Meaning, we hardcode 0,0,1 as UP, and create the stereo pair related to that. (although we could take the camera local UP when rendering panoramas, this wouldn't work for perspective cameras.
* We have no support for interocular distance attenuation based on the proximity of the poles (which helps to reduce the pole rotation effect/artifact).
THIS NEEDS DOCS - both in 2.78 release log and the Blender manual.
Meanwhile you can read about it here: http://code.blender.org/2015/03/1451
This patch specifically dates from March 2015, as you can see in the code.blender.org post. Many thanks to all the reviewers, testers and minor sponsors who helped me maintain spherical-stereo for 1 year.
All that said, have fun with this. This feature was what got me started with Multi-View development (at the time what I was looking for was Fulldome stereo support, but the implementation is the same). In order to make this into Blender I had to make it aiming at a less-specic user-case Thus Multi-View started. (this was December 2012, during Siggraph Asia and a chat I had with Paul Bourke during the conference). I don't have the original patch anymore, but you can find a re-based version of it from March 2013, right before I start with the Multi-View project https://developer.blender.org/P332
Reviewers: sergey, dingto
Subscribers: #cycles
Differential Revision: https://developer.blender.org/D1223
This is a bit annoying, but FFmpeg can't deal with unaligned arrays in some
cases. There seems to be an easy workaround with using SWS_ACCURATE_RND flag
which should be harmless since we don't really scaling frames, but only
using sws_scale to change color space.
Additionally, this is what VLC is also using.
This assumed the OSX SDK version matched the OSX version, which isn't always true.
Also problematic for maintenance and would make building older Blender versions on OSX fail.
Passing in pre-defined OSX_SYSTEM is also supported,
if you have multiple and want to select one.
This patch enables libboost_iostreams and bumps the magic number to match.
Reviewers: dingto
Reviewed By: dingto
Subscribers: dingto
Projects: #bf_blender, #platform:_linux
Differential Revision: https://developer.blender.org/D1840
World space and view space normals were mixed up, we should only convert from
world to view space if a custom normal is connected, otherwise it is already in
view space.
Changing virtual pixelsize only had an affect when a second window was open. (Was also possible to trigger update by opening another window.)
Steps to reproduce:
* Change any editor into user preferences *in main window*
* Change virtual pixel size -> nothing happens
Only applied in a really few cases actually.
To reproduce:
* Open User Preferences *in own window*
* Search for node wrangler add-on (it's one of the few cases where this happens)
* Enable and open details
* Click on one of the menues in the add-on preferences
Actually this was reproducable in any window, user preference area just had to take up most/all of the width.
Note: I'm not totally sure if just disabling these lines is correct, but I didn't find any issues or any information why this was needed. So it seems to be redundant.
Once a dupli had a valid bbox, that bbox would be used for all following objects
without bbox, instead of skipping clipping check.
Issue unveiled by rB3fa0a1a5bc0ff2, but not related at all (in fact, bug was present before that commit).
Current startup .blend has old (percent?) values for particle brush strength.
Since rBe4e21480d6331903c90ab073746484498441e1ac, UI controls do not clamp automatically values anymore,
which means when you first enable comb (or any other brush) you get a 50 strength, waaaayyyy to powerful.
This commit fixes this in `BLO_update_defaults_startup_blend`, note that it does not fix custom users'
startup files, nothing to do here...
Handling `me` data here is not good idea anyway, we override it completly with data
from `tmp` (crash came from freeing already existing bb from me, while pointer still existed in tmp).
(rediscovered it while working on T47676...).
To be backported to 2.77.
Image filter was not set, but only if invoked from toolbar (image strip needs to be selected to see the button).
Caused by rB7fa72b8970, Wasn't aware there's another button for this for image strips.
Quite trivial idea -- just pass tread ID to the texture sampling function.
Implemented as a TLS to avoid passing huge amount of extra contexts around.
Should be working on all platforms, but compilation test is required.
Reviewers: juicyfruit, campbellbarton
Reviewed By: campbellbarton
Differential Revision: https://developer.blender.org/D1831
It'll be nice to eventually go C11/C++11 by default, but for until
then it's kinda handy to be able to build locally with C11 support,
Reviewers: mont29, campbellbarton
Reviewed By: mont29, campbellbarton
Differential Revision: https://developer.blender.org/D1752
Note that we could increase to 524286 without loosing subframe precision,
however this is a round number allowing for longer video export / animations.
See T46859 for details.
`m` unit when used after `cm`/`mm`/etc. ones would get ignored, and the alt version of miles
would be used instead.
The root of the issue is that, in `unit_find_str`, once we get a 'hit' for a unit, we check
it's actual unit (since 'm' would also hit on 'cm', 'mm', etc.). In case that hit is not a
valid unit one, we would just return NULL, breaking the cycle of checks over that unit, and
hence missing all later usages of it.
So now, in case we have an 'invalid unit hit', we immediately retry to find it within remaining string.
This was we don't have re-compo or viewport re-rendering happening when changing
nodes which are not connected to the output at all (for example when adding new
nodes or changing settings for unconnected nodes).
Only basic operations are covered for now. checks could be added to more tools
when needed.
Currently it's not fully optimal implementation, but seems to work fast enough.
Don't see reliable alternative to that -- keeping tag in the node wouldn't work
because of the node groups (which are probably already broken, but should be
easy to solve with current approach). So guess it's more matter of optimizing
path search from a node to output.
Before processing forward let's check whether it's indeed something we want
and whether the approach is indeed not fully bad.
Reviewers: campbellbarton, mont29
Subscribers: sebastian_k
Differential Revision: https://developer.blender.org/D1765
Pointers of faces were passed as face keys during parametrizer's face creation. Since those
addresses were different for every run, the layout of the faces ended up being different
in the internal hash, leading to inconsistent order of their evaluation during LSCM solving,
and slightly different UV maps.
Solved by simply using faces' indices as key instead, which ensures we always get same results
with exact same input data now.
Many thanks to Roman Nagornov (RomanN) for raising the issue, investigating it and finding
the solution! And thanks to Brecht for quick review too.
The issue was caused by some code accessing R from a functions which
are marked as safe for use from outside of render pipeline.
Now those functions are safe(er) for use.
This is an alternative to passing a copy callback which is some times inconvenient.
Instead the caller can write to the key - needed when the key is duplicated memory.
Allows for minor optimization in ghash/gset use.
Also add BLI_gset_ensure_p_ex
Code was using degrees as radians.
Still unclear why default cube will have 180 degrees angle, but new meshes 30,
but that's kinda separate topic which is to be addressed separately.
This is a subject for final 2.77 release.
See rB935e241fa6ea095493 for details of the issue, but first fix caused regression T47632.
So for now handling the issue in a localized way, this is not a real solution (since this could happen
in other cases), but will do for 2.77.
This commit is to be backported to 2.77.
This reverts commit 935e241fa6.
Issue will be fixed in a more localized way for now (not that nice, since this use-after-free can possibly happen
in other places too, but only safe solution for 2.77).
This commit is to be backported in 2.77.
Probably did not happen yet (since nobody uses addons translations...), but there was an
nice memleak during creation of translation ghash in case a same msgid/msgctx would be
added more than once.
Also, no need to allocate (and free) a temp key each time we lookup a msgid, we can use
given const strings directly here!
Needed to allow modal UI keymaps, but I'm sure we'll need this more often in future.
First item will be modal eyedropper keymap coming in a following commit.
Steps to reproduce fixed glitches were:
* Change any editor to be file browser from menu, Ctrl+O *from the file browser area*, Esc -> area reset to what it was before changing to file browser initially
* Ctrl+O from any area, F12, Esc -> returns to initial editor in full-screen (expected is file browser in full-screen)
Fixes T46229
Core of the fix is removing old area from spacedata list when going back to previous area (see ED_area_prevspace -> BKE_spacedata_remove). Also, when creating a new temp area we now don't exit old area anymore (needed so SpaceFile->op is kept, but it also makes sense in general)
Aaand finally removes some ugly hacks.
Tested quite a bit, so I think it's safe to apply (besides of remark below), just would like to get things double checked and confirmed. After all, this full-screen stuff finally starts to feel like it's working :P
Note, there's still a memory leak when quitting Blender with temp area open. Haven't found out how to solve yet, but it's not that important for review anyway.
Reviewers: campbellbarton, brecht
Reviewed By: brecht
Subscribers: plyczkowski, Blendify
Maniphest Tasks: T46229
Differential Revision: https://developer.blender.org/D1531
Those new functions invert the winding of polygons, effectively inverting their normals.
A helper was also added to allow swapping two items in customdata layers.
Being able to invert normals outside of BMesh area is very important in several places,
like IO scripts or customnormals modifiers...
Reviewers: campbellbarton
Differential Revision: https://developer.blender.org/D1814
creator.c was getting hard to follow.
- Split off argument and signal handling into own files.
- Move docstrings next to functions (to keep docs grouped with code).
Now using same UI as in object/armature properties, also save one line in 3DView panesl. ;)
Nothing crucial there, but nice & safe to backport to 2.77 imho.
ked->iterflags actually already had an enum defined, but at the time of the previous
fix, I'd forgotten that it was defined. As a result, "f1 needs NLA remap" got given
the same flag as "include handles when doing keyframe selection checks".
In response to user feedback, this commit brings back the ability to
limit motionpath clearing to only happening for those on selected
objects/bones.
By default, the "Clear" operator will clear from all objects/bones,
unless the Shift key is held.
Previously CMake would show many messages about missing libs before checking for a missing LIBDIR.
Now show the error immediately and exit.
Also allow for custom LIBDIR on OSX.
With continuous grab disabled, non-linear mapping for int buttons
wasn't working usefully with small mouse movements.
Now 2x pixels motion adjusts by at least 1 w/ int buttons.
OpenVDB is compiled with Blosc support, meaning we should pass Blosc to
a linker now.
Unfortunately, the Blosc has compiled-in crt and pthread library, which
still screws up linking.
Took me some time to figure out what was going on here... Was again that delayed button
callback stuff (`ui_apply_but_funcs_after()`), first calling button op, and then
its callback func.
Issue was that 'open file' op (through call to `WM_file_read()`) would clear
the splash screen (as more or less the entire 'dynamic' UI), but callback func of that splash
(`wm_block_splash_refreshmenu()`) would still try to access that freed menu's region.
So, root of the issue seems to be that setting context's wm/win/etc. would not clear
context's menu pointer (while clearing all other 'sub' pointers). I could not find
nor imagine any case where this behavior could be desired, so simply added nullification
of that pointer when setting context's wm/win/etc.
Note that crash was due to read-after-free, infuriating debug builds with asan,
but seems like release builds never actually crashed on it.
The issue was happening when having unconnected point density which
will cache data but will not free it because there's no actual call
to the actual sampling.
Now the idea is to make sure cache is zeroed on file load and undo
and then caching via RNA will free the data if any exists. This could
leave us with a single copy of cache in the node if it's not used,
but it's quite small amount of memory and it's not leaking.
This will respect the official build configuration where we should
have BLOSC enabled.
We can't really detect if OpenVDB was compiled with BLOSC or not,
so seems we can't really avoid this extra flag.
Mainly it's related on a bad practice in SDL to force-define __SSE__
and __SSE2__ flags which generates quite some warnings and causes too
much noise.
There are some other warnings fixed. Should be no functional changes.
NeXyon, please check the changes in audaspace :)
Displaying a button would clamp the value if the button was outside the range.
This could be OK in some cases,
however it's problematic with object dimensions which would re-scale objects on showing the panel.
Add `ui_but_update_edited` when its OK to modify the value.
Without this, grabbing with normal weight will continually select new normals
based on where you move the cursor,
causing the normal location to flicker in a way which isn't controllable in any useful way.
Also re-reported through IRC by Thomas Beck (@plasmasolutions), thanks.
Though it's not ideal in theory, we have quite poor handling of object datablock currently
from user PoV - before this commit, it was not easily possible to get fully rid of an object
anymore if you did not removed it from all its groups before deleting it.
So for now, restore 2.76 behavior (namely, unlink an object from avaerything in Blender
once it is no more used by any scene).
Better handling of all this is TODO for later (also related to much more heavy changes
done in id-remap branch regarding sanitizing our ID deletion process).
Avoid allocating the (tiny) array on the heap in the first place.
Reviewers: sergey, lukastoenne
Differential Revision: https://developer.blender.org/D1815
This re-enables the AA jittering, but with proper clamping so that u >= 0,
v >= 0 and u+v <= 1.
Differential Revision: https://developer.blender.org/D1254
Previous commit 9d18fd11 introduced the possibility to scale the modelview
matrix by setting a scale on the camera but this caused undesirable lighting
and dynamic side effects.
This commit reverts the scaling effect but keeps the axis inversion effect,
which was the purpose of the first commit anyway.
A negative/positive scale on the camera is applied respectively as -1/+1 scale
on the modelview matrix.
Did a full compile of debug build with C++11 enabled, it all passed compilation
apart from some deprecated type used in GE's Video Texture. Solved it inside of
ifdef block now.
In the future we should uncomment the MSVC part of it, it should all be safe and
correct (MSVC2013 does not define new C++ version but supports C++11). The reason
it is commented is to have absolutely no effect on the upcoming release.
Behavior is similar to python's set.pop(), it removes and returns a 'random' entry from the hash.
Notes:
* Popping will return items in same order as ghash/gset iterators (i.e. increasing
order in internal buckets-based storage), unless ghash/gset is modified in between.
* We are keeping a track of the latest bucket we popped out (through a 'state' parameter),
this allows for similar performances to iterators when iteratively popping a whole hash
(without it, we are roughly O(n!), with it we are roughly O(n)...).
Reviewers: campbellbarton
Differential Revision: https://developer.blender.org/D1808
This is an attempt to improve the User preferences panel for the 3DView. I made 2 changes:
- I reordered the sequence of properties by grouping them into more logical groups as it made sense to me. Please indicate where to rearrange the order if necessary.
- Then i added some changes in the code to get the groups better arranged visually. I am pretty sure that this can be done much better, more clever, more generic, whatever. This is just what i could figure out on my own so far.
Reviewers: aligorith, sergey, gaiaclary
Subscribers: sergey
Projects: #user_interface
Maniphest Tasks: T47295
Differential Revision: https://developer.blender.org/D1757
Those warnings are trigerred by stl classes in OCIO's public interface.
To quote MSDN: "C4251 can be ignored if you are deriving from a type in
the Standard C++ Library"
This is the only instance where those warnings hunts us, so for now we
can keep it all local in intern/opencolorio but this might be changed
in the future.
There's no need to mix ints and size_t here at all
because all the values fits into integer.
It's unlikely we'll be re-bundling Carve, so didn't
bother with the patchset.
The idea now is to have FFmpeg/OIIO headers listed after
the system ones. This is because FFmpeg/OIIO might define
some constants with the same name as the ones from math.h.
FFmpeg/OIIO has ifdef around defines, but math.h doesn't
check whether constants were already defined or not, which
causes some noisy warnings.
The problem is that node trees (such as the Material, Lamp, and Compositor node trees)
are stored as "nested node trees" on the affected datablocks. They are particularly
troublesome to deal with, as the are not easily identified, and also cannot be easily
mapped back to the ID's which actually own them. As a result, the usual automated
methods do not work when dealing with these!
Only Driver FCurves with named shapekeys (instead of shapekey indices) was
getting picked up by the UI code for testing whether a property had drivers
or not. So, while this version patching code worked when it was initially
written for the 2.4x -> 2.5 transition, some subsequent changes ended up
breaking this. As this stuff is not used often, the breakage wasn't noticed.
* Remove unnecessary big margins after triangle icons (esp. for keymap items they where waaay to big)
* Remove ugly margin after "Foo (Global)" entries
* Add little margin after expanded key maps
Actually, 3 issues here:
* Keymap items of "Foo (Global)" maps weren't indented enough
* Items after a opened "Foo (Global)" map were indented too much (col was overridden then).
Seems to be some compiler fault which leads to a wrong flag being used,
making it so wrong number of samples is used for the background.
This should in theory fix issue reported in T47213.
This patch adds a python method to get openGL bind code of material's texture according to the texture slot.
Example:
import bge
cont = bge.logic.getCurrentController()
own = cont.owner
bindId = own.meshes[0].materials[0].getTextureBindcode(0)
Test file: http://www.pasteall.org/blend/40679
This can be used to play with texture in openGL, for example, remove mipmap on the texture or play with all wrapping or filtering options.
And this can be used to learn openGL with Blender.
Reviewers: TwisterGE, kupoman, moguri, panzergame
Reviewed By: TwisterGE, kupoman, moguri, panzergame
Projects: #game_engine
Differential Revision: https://developer.blender.org/D1804
The issue here was actually somewhere else - the attached scene from the report used a light falloff node in a sunlamp (aka distant light).
However, since distant lamps set the ray length to FLT_MAX and the light falloff node squares this value, it overflows and produces a NaN
weight, which propagates and leads to a NaN intensity, which is then clamped to zero and produces the black pixels.
To fix that issue, the smoothing part of the light falloff is just ignored if the smoothing term isn't finite (which makes sense since
the term should converge to 1 as the distance increases).
The reason for the different results on CPUs and GPUs is not perfectly clear, but probably can be explained with different handling of
Inf/NaN edge cases.
Also, to notice issues like these faster in the future, kernel_asserts were added that evaluate as false as soon as a non-finite intensity is produced.
This way we are always sure that we are using latest ever configuration
without need to manually poke buildbot after doing changes in the building
environment.
Also uncomment code back, was a left-over from some debug.
This was a hard decision, because going newer CUDA toolkit makes
rendering up to 5% slower. But on another hand, it solves major
speed regressions (up to 30%) with branched path tracing on a
top level cards.
Neither of those regressions have a meaningful and sane workaround
from the code itself.
Toolkit 6.5 could still be used, but it's no longer recommended one.
It's not really handy to silence something unused hoping for it'll be
used in the future. We can end up with quite some silencing then.
Also made this flag which i find rather useless to NOT cause -Werror
in Cycles code.
When using multiple portals, scene areas behind one of the portals were rendered darker than they should.
The reason for that is a pretty stupid mistake: Since portals are only used at positions that aren't behind them,
only portals that are used should be accounted for in the PDF calculation. That was actually the case, but the final
divide incorrectly divided by the total amount of portals, not the amount of visible ones.
Another issue with areas behind portals was the PDF evaluation function.
The new evaluation code is shorter, simpler and fixes this issue.
Also, the threshold for the distance check was increased to avoid artifacts where portals touch a surface.
Regression from rB2dba2b3d71d9781bce45. Do not understand why MSVC needs this convoluted
allocation (looks like broken compiler crap?), but at least let's do it correctly!
Now a 'More' item is added to the pie when there are too many items. It opens a sub-pie that contains the remaining items.
Note that this only touches operator-enum pies (like the object mode pie is), it is not a complete support for pies with more than 8 items. For this further design and code work would be needed, but this is too urgent to wait for that.
This is a better fix for T46973, should definitely be applied for 2.77 release.
Patch D1800 by myself with some edits by @campbellbarton, thanks!
While it's not something we'll be using for the official release,
it's nice to support new libraries at least on "it compiles" level,
so it's not that many frustrated developers around.
Nexyon, please have a look into Audaspace changes :)
Supports both smoke/fire and point density textures now.
Reduces number of textures available for sm_20 and sm_21, but you have
to compromise somewhere on such a limited hardware.
Currently limited to linear interpolation only, and decoupled ray
marching is not supported yet. Think those could be considered just a
further improvement.
Some quick example:
https://developer.blender.org/F282934
Code is minimal and we can fully consider it a fix for missing
support of 3D textures with CUDA.
Reviewers: lukasstockner97, brecht, juicyfruit, dingto
Reviewed By: brecht, juicyfruit, dingto
Subscribers: mib2berlin
Differential Revision: https://developer.blender.org/D1806
BKE_main_id_tag_/BKE_main_id_flag_ were horrible naming now that we split those
into flags (for presistent one) and tags (for runtime ones).
Got rid of previous 'tag_' functions behavior (those who were dedicated shortcuts
to set/clear LIB_TAG_DOIT), so now '_tag_' functions affect tags, and '_flag_'
functions affect flags.
This made byte & float images behave differently, where other modifiers remain the same.
Also remove scene from the modifier (should have been passed as arg but no longer needed).
Based on an user feedback, previous implementation with providing
decoupled X and Y speeds didn't work in production at all: there
is no way to combine this speeds to an usable vector.
So now we're providing speed vector output instead, which provides
speed in an exactly the way Vector Blur node expects it to be:
first two components is a speed from the past, second two components
defines speed to the future.
Old behavior can be achieved by RGBA separating the speed output
and using first tow components.
Now this speed gives quite the same results as a speed pass, with
the only difference that track position speed uses "shutter" of
1 while pass uses shutter of 0.5 (and there's no way to affect on
that?).
The issue was caused by static vectors allocating some internal
data using rebound element allocator for them, which was causing
access to a non-initialized statistics objects and was failing a
lot when switching Blender to a fully guarded allocation.
Additionally, we were not able to free that internal memory before
Blender exits, which was causing false-positive memory leak prints.
Now we're not using GuardedAllocator for those proxy containers.
Ideally this should be done as a GuardedAllocator::rebind, but
it didn't work for vector<bool> because it seems some internal
parts are converting bool to char32_t, which either makes it so
we can't use GuardedAllocator for those vectors or the compiler
get's confused when we're trying explicitly allow GuardedAllocator
for rebind<char32_t>.
This with current approach we should be fine for the release.
Python name could include ABI-flags after the version,
since checking for all combinations of ABI flags can expand into many possibilities,
take the executable name from the build system.
Was caused by some safety things of making sure we've for NULL
terminator for the buffer when doing mbs<->wcs conversion, but
it turns out this simply confuses str::string and it can no
longer have proper .size(). Let's assume behavior of string
allocation is same all over the std, and we can avoid having
that extra null-terminator allocated.
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
C++ requires specific alignment of the allocations which was not an
issue when using GCC but uncovered issue when using Clang on OSX.
Perhaps some versions of Clang might show errors on other platforms
as well.
On undo, sculpting regular meshes would update _all_ GPU VBO's.
Avoiding the update gives noticeably faster undo.
This is also a fix/workaround for strange behavior with NVidia's driver (T47232),
Where locking and unlocking all buffers for updating
slows down redraw speed permanently after the first undo.
However the problem isn't avoided entirely since a single brush stroke might modify most of the mesh.
CLAMP was conflicting between Common.h and BLI_utildefines.h
Ideally we would use macro from BLI, but it's a bit involved change
to make it working with C++, will keep it for later.
We don't have vectors re-allocation happening multiple times from inside
a loop anymore, so we can safely switch to a memory guarded allocator for
vectors and keep track on the memory usage at various stages of rendering.
Additionally, when building from inside Blender repository, Cycles will
use Blender's guarded allocator, so actual memory usage will be displayed
in the Space Info header.
There are couple of tricky aspects of the patch:
- TaskScheduler::exit() now explicitly frees memory used by `threads`.
This is needed because `threads` is a static member which destructor
isn't getting called on Blender's exit which caused memory leak print
to happen.
This shouldn't give any measurable speed issues, reallocation of that
vector is only one of fewzillion other allocations happening during
synchronization.
- Use regular guarded malloc (not aligned one). No idea why it was
made to be aligned in the first place. Perhaps some corner case tests
or so. Vector was never expected to be aligned anyway. Let's see if
we'll have actual bugs with this.
Reviewers: dingto, lukasstockner97, juicyfruit, brecht
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D1774
Basically the idea is to make code robust against extending
enum options in the future by falling back to a known safe
default setting when RNA is set to something unknown.
While this approach solves the issues similar to T47377,
but it wouldn't really help when/if any of the RNA values
gets ever deprecated and removed. There'll be no simple
solution to that apart from defining explicit mapping from
RNA value to Cycles one.
Another part which isn't so great actually is that we now
have to have some enum guards and give some explicit values
to the enum items, but we can live with that perhaps.
Reviewers: dingto, juicyfruit, lukasstockner97, brecht
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D1785
Since Cmd + A works elsewhere, it should work during font objects editing as well.
There is still a mysterious issue with Cmd + Z not working for UNDO the edited
text in OSX (probably GHOST related).
When pasting text, the style (bold, material, ...) is maintained, if it was originally copied from Blender.
This fixes the issue of missing copy/paste options for font objects
(they were present back in Blender 2.49)
Reviewers: Severin, campbellbarton, brecht
Now take into account `dupli_offset` of groups.
Also fixed another issue related to objects renderability, previous code was giving
bad results due to added temp objects to generate previews (camera, lamp...).
Was only trying to install $BOOST_VERSION or higher boost - being fairly recent
this always failed. We can live with older versions too, so now using
*BOOST_VERSION_MIN as with other distros.
This happens when the properties panel is pinned to the material tab.
Patch by Ralf Hölzemer (aka cheleb), thanks!
Differential Revision: https://developer.blender.org/D1776
Really annoying bug, the code was not forward compatible at all and
resulted in crash. And it is really good to keep at least one release
forward compatibility so possible regressions could be verified easily.
The idea now is to use new property name for the pixel filter type,
but keep old property around for a couple of releases, so we have at
least some forward compatibility.
Don't like this situation at all, but seems it's least of the evil
we can choose.
Thanks Brecht for the review!
Some dynamic enums, which do not need a valid context pointer, have their 'itemf'
callback always called. This is annoying for introspection tools (like the ones generating
translations, or API documentation), because it means they never have access to all possible
options (enum items).
So now, there is also an `enum_items_static` accessor to get only statically-defined
enum items.
Note: only i18n tools take advantage of this currently, others are still to be updated.
Reviewers: campbellbarton, sergey
Differential Revision: https://developer.blender.org/D1782
Accidentally left off null check when cleaning up the code there in
34993bf97d (GPencil_Editing_Stage3 branch)
Thanks to Sami Pelkonen (pltsi) who reported this bug, along with
some nice and detailed backtraces which made it easy to locate
the problem :)
Basically the default U.glalphaclip was 1.8367099231598242e-40 in some
computers (e.g., OSX 10.11, AMD Radeon HD 6750M) this value would need for bad
gl clipping.
The new default is 0.0. To test this before/after the fix, just run:
$ blender -b --factory-startup --python-expr "import bpy; print(bpy.context.user_preferences.system.gl_clip_alpha)"
For some reason, using 'dead keys' to compose some accentuated latin chars will generate
some 'unknown' key events, and in this case, direct `event.keymodifier` was correctly
cleared, but not its 'backup' version in `win->eventstate`, so all further events would
get an invalid modifier until some real one would be pressed again...
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.
* Added a new API function to test if a GPencil layer is visible or not
* Replaced all editability checks with this new "super check"
* Replaced all magic number thresholds for opacity visiblity with a single define
This may have resulted in situations where the order of GP keyframes was
incorrect (leading to some frames not being able to be found), or in some
redraw problems when trying to delete GP keyframes (that I was getting earlier,
but can't seem to reproduce now)
TODO: We now need to hook up a proper api to do the GP key sorting
In the other Dopesheet modes, the "Speaker" icon was used to refer to
"animation playback muting", while for GP layers, this was being incorrectly
abused for "layer visibility in viewport".
This commit fixes that by making the following changes:
* A new "eye" icon toggle is added for controlling GP Layer visibility
* The "speaker" icon toggle now controls "Lock Frame to Current" functionality,
which functions more like the "animation playback muting" that is generally
expected.
When animating to the camera, and working in "Continuous Drawing Mode",
it be necessary to return to the active camera viewport, after briefly pivoting
the view using MMB (to check on perspective or something like that). However,
before this fix, you would have had to exit drawing mode to do this.
Sculpt settings are often quite handy to be able to easily access while drawing,
so make it easier to access these when drawing without having to first expand
the panel.
(On the other hand, the "Edit Strokes" panel contains tools already found in
various other places - menus,etc. - so no need to expand it by default)
Corrects mirror syncing for invert, levels & smooth.
Note that the code changed to process mirroring even if both verts are selected,
since group flipping can mean that is still meaningful.
Since rBbbc7dc169dc365889bad3f3aed7b868efb432710 bool-valued RNA functions are expected to
return only 0 or 1. For flag tests as in these functions the returned int value needs to be
explicitly converted to bool.
Not sure why I made the logic so complicated before, this change should solve
the deadlock when the render thread tries to acquire the main thread lock while
the main thread is waiting for threads to finish.
The bake system had a 200ms sleep for each frame substep, to give the UI time
to redraw. I don't think there is a good reason to have this, with fair thread
scheduling this will give UI thread 2x more time at best, and the UI doesn't
need to be that responsive during bake.
Now pass_filter is modified to have exactly the flags for the light components
that need to be baked, based on the shader type. This simplifies the logic.
There were two issues:
1. Memory leak: std:;erase does not call delete on the
pointer (which is actually a good idea),
2. After MIS was disabled in viewport render there was
no way to bring MIS back.
Now instead of removing light from the scene data we
kind of tagging it for an ignore. Possible cleanup
would be to add Light::is_enabled and use that instead
of passing weird and wonderful function arguments.
The title says it all actually, the idea is to make Cycles
only requiring Boost via 3rd party dependencies like OIIO
and OSL.
So now there are only few places which still uses Boost:
- Foreach, function bindings and threading primitives.
Those we can easily get rid with C++11 bump (which seems
inevitable sooner or later if we'll want ot use newer
LLVM for OSL),
- Networking devices
There's no quick solution for those currently, but there
are some patches around which improves serialization.
Reviewers: juicyfruit, mont29, campbellbarton, brecht, dingto
Reviewed By: brecht, dingto
Differential Revision: https://developer.blender.org/D1764
This is an initial move to have unittests to at least cover
utility functions, which then could be extended further to
test such areas as shader optimization and such.
Currently only based on initial "infrastructure" layout and
writing tests needed to test the no-boost patch.
Note: This patch starts to use "<dir>/<header>.h" notation
for the include statements which i just got used to do in
other projects. Something what would be cool to use globally
in the code eventually.
Reviewers: dingto, juicyfruit, lukasstockner97, brecht
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D1770
We've upgraded to Boost-1.60 and MSVC2013 since the workaround
was originally committed. After checks with current compiler and
libraries the original bug is no longer happening.
This will make string comparison much faster in Windows, solving
synchronization bottlenecks of fewzillion objects.
Thanks Martin Felke (aka scorpion81) for the tests!
Unfortunately this doesn't make all tests compilable due
to all sort of weird and wonderful bad levels includes
on Windows (G referenced from bf_blenlib) but at least
allows to selectively build tests for now.
This change the following values:
- World settings:
- Use MIS: On
- MIS Samples: 1
- MIS Resolution: 1024
Enabling World MIS per default won't make simple backgrounds (flat background color) slower,
see previous commit. This gets disabled internally if World MIS is not actually needed.
When World MIS is enabled by the user, we now check if we actually need it.
In case of a simple node setup (no procedurals, no HDRs..) we auto disable MIS internally to save render time.
This change is important for upcoming default changes.
Same causes as when snapping to faces, so same solution: in case of ortho view,
offset start of ray to be just slightly outside of the target's bbox, to avoid
too much far away start point that generate floating point computation instability
in BVH raycasting.
Note that this lead to some refactoring, to avoid duplicating too much of code.
For now, edge snapping seems to behave OK (uses different logic), so not touched.
Based on patch by Germano Cavalcante (@mano-wii), thanks!
When in TweakMode on NLA strips that had an offset, it was not possible to select
those keyframes in the Summary Channel in the Dope Sheet.
The main gist of it is that the current code is from before the summary track was
introduced, and so could assume that ANIM_nla_mapping_get() would work for all channels
present. Thus, simply converting the clicked frame to nla-mapped time once would be
enough. However, for summary channels, nla-mapping_get() doesn't do anything, since
we can potentially include keyframes from several different objects!
Please remember to change those each time you update building code and there is no version change.
Otherwise poeple re-running the script won't get the updated builds.
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.
Since we're now on 1.7 we can skip building the OSL testsuite. It was giving various compilation errors on different machines and we don't need them anyway...so just disabling them in the script.
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.
Issue was caused by update RNA callbacks freeing the dependency
graph, which is only needed to tag depsgraph for rebuild.
Solved by using a flag for the depsgraph which indicated that it
is to be rebuilt.
Only happened when rendering second time with OSL enabled.
Was caused by fix for T47201. Hopefully now all the issues
are solved (meaning, we don't keep files opened and don't
crash either).
D1751, remove this library since its quite a specific - only supports an older version of this codec.
Also ffmpeg has added support for recent versions of the codec.
Regression from rB12c71508c2d7.
Now, we systematically first try keycode from `XLookupKeysym()`, and only fall back to
the one from `XLookupString()` if it failed to convert to a valid gkey.
The issue was caused by fix for 31017 which resulted in some
missing intersecitons recorded which screwed inner/outer checks.
This is an old bug, so didn't bother with forcing re-distribution
to happen on file open to avoid possible other regressions.
The issue was introduced by a fix for T44713 which only made GLSL
consistent with Cycles.
Now we do have conditional averaging or proper luma weighting based
on whether we're new old old shading system. Not totally ideal but
should work for until we re-design viewport possibly breaking how
Blender Internal does implicit conversion.
Original change in d631252 was not quite correct because it
was missing release since the acquire() will return NULL.
Hopefully now all the cases are covered.
Such configuration used to cause quite confusing situation when
stamp will use actual scene's statistics but metadata from strip
will be used for the saved file (basically, causing different
information stamped and saved as metadata).
Don't think it was desired behavior and it's something what
artists here in the studio wants to be fixed.
This way it's possible to have some color-correction modifications on
top of the render result and yet still have proper metadata stored.
Usecase: Access per-frame render-time of the movie frames from the
final export.
In some cases transform modes would use the custom-data pointer,
other times the transform conversion functions would.
However with some combinations (bone mirror + bend for eg),
both conversion & transform mode would use this pointer causing a crash.
Fix this by having 2 custom-data pointers:
one for the mode, another for the data-type.
This also simplifies time-slide which was conditionally mixing mode/type data in the one array.
Those two were the only one rotating around some world axis (turntable mode),
so better make their behavior consistent with other rotating numpad keys.
This breaks the turntable-rotation with numpad keys, though.
That was the main issue (in both T46455 and T46690), solved by 'flattening' those chains (v1 -> v2 ->v3 etc.)
before calling `CDDM_merge_verts()`.
Also added note to `CDDM_merge_verts()` that it does not support chained mapping, along with
a basic assert that should catch most of those cases in future.
The logic of 'following mapping' was also rather broken, making a special case here when using
object-controlled offset is very weak. Further more, blindly following mapping in this case
was far from ideal, this could end to merging vertices rather far from each other.
To address this issue, we now always follow mapping, but only as long as 'final' vertex remains
close enough from mapped one.
Finally, the search of 'closest' vertex to merge with was also quite bad, would just pick the first
one matching distance limit, instead of using the actual closest one - could lead to rather ugly
geometry deformations in case one would use not-so-small merge threashold!
This way we avoid passing structures which could be up to
few hundred bytes by value to the utility functions.
Ideally we'll also have to add `const` qualifier in majority
of the calls, but C++ RNA does not allow us to do that because
it does not know if some function modifies contents or not.
The idea is to switch from allocating separate buffers for shader data's
structure of arrays to allocating one huge memory block and do some index
trickery to make it accessed as SOA.
This saves quite reasonable amount of lines of code in device_opencl and
also makes it possible to get rid of special declaration of ShaderData
structure.
As a side effect it also makes it easier to experiment with SOA vs. AOS
for split kernel.
Works fine here on NVidia GTX580, Intel CPU amd AMD Fiji cards.
Reviewers: #cycles, brecht, juicyfruit, dingto
Differential Revision: https://developer.blender.org/D1593
Seems i was the only one who was really up to using it and
i do have gcc-5 finally backported and installed here so
such a fine-tune flags are no longer needed.
It was mainly unfinished code for volume in a split kernel which
should be done differently anyway to avoid such a code copy-paste.
The code didn't really work, so likely nobody will cry.
OpenVDB is like a 30-years old lib when it comes to building it - not even a configure.sh script!
This makes it impossible to auto-build (install_deps is not a new CMake!).
So for now, just storing some template code for it, and expected version/deps data
in --show-deps option.
Previously it was only possible to replace all existing modifiers with the
new list, which isn't so great for grading.
Ideally we should also have some sort of merge policy here, but that's for
later.
Use KernelGlobals to access all the global arrays for the intermediate
storage instead of passing all this storage things explicitly.
Tested here with Intel OpenCL, NVIDIA GTX580 and AMD Fiji, didn't see
any artifacts, so guess it's all good.
Reviewers: juicyfruit, dingto, lukasstockner97
Differential Revision: https://developer.blender.org/D1736
The idea is to use known number of frames in the stream when possible
instead of trying to deduct it from a stream duration and time base,
which could potentially give some rounding errors.
This, i.e., solves quicktime encoded files from animators to open
just fine in sequencer.
Another report done by the main Nieve guy (aka venomgfx)
Why this is not working in original code and works int this one remains mystery
(see comments for details).
Note that we still do not support at all non-latin keymaps for our shortcuts,
this would be nice to add someday, but that's a TODO, not a bug.
Reviewers: sergey, campbellbarton
Reviewed By: campbellbarton
Differential Revision: https://developer.blender.org/D1746
The issue was caused by some functions in compositor returning
garbage values. Partially the issue was caused by vc12_xp toolset
we're using, but even with regular vc120 toolset nodes like bokeh
image did not work correct.
This is a bit weird solution, which could indicate some sort
of compiler bug, but is also actually makes sense because we do
use SSE intrinsics in the compositor now. Maybe it all gets
interfered in some way.
In any case, if someone wants to find a real solution for the
issue please go ahead.
This shouldn't have affect on supported platform because we
already demand CPU to have SSE2 support.
One draw call for all screen area outlines instead of one per line.
(Two for retina displays)
One draw call for area split line instead of two.
En/Disable GL_BLEND once for area join arrow instead of twice.
Fixed this with three changes:
* filter_glob is now 255 char max (63 could be a bit limited in some rare cases).
* IO templates now explicitely define max len of that property (such that scripters are aware of the limit).
* ED_fileselect_set_params() is now safe regarding too long strings from a 'filter_glob' op property.
Debug print here is not that useful to common user, and keyframe deletion
does report warning, so do the same for clear op.
Reported by venomgfx over IRC, thanks.
It was not so much better from performance point of view to have
a special case here and it only made us having two entry points
to the same things, which was rather annoying.
Should no be functional changes for artists.
Should be the last bit of sculpt/paint code, now this is fully using BLI_task.
Note that PBVH normals update is now about 20% quicker than with OMP code
(from 27ms to 21ms with a big stroke over a 500k vertices monkey), even though
a missing thread-protection was added... atomic primitives ftw!
For the records, with the missing `#pragma omp critical` section added,
previous code was like four times slower (above 100ms).
This variant behave exactly as TIMEIT_START etc., but it also sums up all times in
a static var and prints out average execution time - very useful when dealing
with small/quick pieces of code that get executed often, to get some meaningful results.
Needed by incomming changes in pbvh.c.
Note that we make it much simpler than for other primitives in this file - think
we could revise its content to make it simpler one day...
Issue was that the domain matrix was not initialized properly on the
first frame (in smokeModifier_init), which caused OpenVDB to throw an
exception for trying to create a VDB grid with non-affine transform.
This patch supports "Image or Movie" and "Environment map" types of world texture for the viewport.
It supports:
- "View", "AngMap" and "Equirectangular" types of mapping.
- Different types of texture blending (according to BI world render).
- Same color blending as when it lacked textures (but render via glsl).
{F207734}
{F207735}
Example: {F275180}
Original author: @valentin_b4w
Regards,
Alexander (Blend4Web Team).
Reviewers: sergey, valentin_b4w, brecht, merwin
Reviewed By: merwin
Subscribers: campbellbarton, merwin, blueprintrandom, youle, a.romanov, yurikovelenov, AlexKowel, Evgeny_Rodygin
Projects: #rendering, #opengl_gfx, #bf_blender:_next
Differential Revision: https://developer.blender.org/D1414
We used to require the 3Dconnexion driver installed at build-time so
this was turned off.
I reworked the code last year to remove the driver dependency so
there’s no reason not to enable this.
Not so useful now that we use BLI_task! Not sure why those were ever added actually,
readng carefully that code only modified data here is the PBVHNode, which is only
used/affected by one thread at a time ever. And shared read data (PBVH itself) is
not modified during brush execution itself, so it's safe to use it threaded too.
Only concerns poly normals computing, have usual 10% speedup of affected code for OMP -> BLI_task switching.
Also parallelized the 'weighted accum' part (used when computing both polys and vertices normals,
when using modifiers e.g.), which gives nice 325% speedup (from 66ms to 20ms for a 500k poly monkey
with simple deform modifier e.g.). ;)
The issue was caused by different AABB used by Cycles and texture sampler.
Instead of trying to keep this two functions in sync we now do have an
utility call in the point density node to query the AABB.
I tried to carefully preserve all patches since the last upgrade.
Improves T47195, cloth collision detection bug.
Differential Revision: https://developer.blender.org/D1739
Basically, the 'fake undo' restoring orig coordinates in this case cannot work with dyntopo,
since it assumes nothing was added/removed.
For now, just prevent this 'restoration' when dyntopo is used, this is no ideal solution
from user PoV - but it's better than plain ugly crash!
Complete solution seems much more involved and outside of scope of bug handling, added a TODO note:
http://wiki.blender.org/index.php/Dev:Source/Development/Todo/Tools#Sculpting
Title actually tells it all, it is rather simple function which totally makes
sense to be inlined.
This gives up to 5% of speedup when updating scopes for a large image.
Reviewers: campbellbarton
Differential Revision: https://developer.blender.org/D1310
This reverts commit 3dbc123061, "BGE:
allow setting velocity to zero in a motion actuator" as it caused more
issues than it solved. Zeroing linear or angular velocity with logic
bricks is discussed further in https://developer.blender.org/D1545
Doing something like:
bpy.data.objects["Cube.001"].data.shape_keys.animation_data.drivers.find('eval_time')
wouldn't work, because the drivers.find(datapath, index) method wasn't implemented
yet. Previously, it was only implemented for FCurves in actions.
When using an empty render result (after re-opening the file, i.e.)
some tools (like masking) could have used wrong image resolution
because of not being aware of special cases supported by the image
space to display the render result.
This should fix selecting mask points when mask is opened op top
if an empty render result.
This fixes some of the issues noted by venomgfx that were caused by
Se25ba162c0b62b19cf367f0f29e29d0c0960978d
Specifically, this commit fixes:
* Timeline: Keyframe lines
* Graph Editor: Curves and Handles
This is handy for cases when mask is created on top of the edit and
used for tasks like color grading and other enhancement.
That was the main purpose of the masks which was totally broken in
6786ef6. Now it's possible to have masks created as both a part of
input movie roto process (which then better be re-mapped to the strip
timing) and as a grading tool (which should be using scene timing
for the animation).
Thanks artists from the Nieve for screaming about such a broken case.
There is no reason to do separate image buffer release when there's
found buffer but with empty rects because of the following reasons:
- All the acquire() calls are followed with corresponding release()
calls, regardless of whether image buffer was empty, missing or
whatever.
- It was done wrong -- since lock was passed as NULL, it'll only
de-reference the image buffer itself, this causes following:
* Wrong user counter since there'll subsequent release() call with
the proper lock passed to it.
* Global locks are to be released prior to the spin locks,
and such an extra release violated this rule.
Compositor tried to be too much smart and avoid unneeded re-calculations of
the distortion model, but the way it was implemented is by falling back to
the nearest interpolation first.
We can't really cheat here, better to just look into faster models estimation.
While it's not really working reliably to drive something on runtime
it's handy to have such possibility to tune particle system using
proxies in the scene files.
Just another quick thing for the Nieve project.
Since weight_other is equal to weight_accum_prev[i_other], the original
lines actually are no-op. The visible effect is that when smoothing just
two vertices with weights 1 and 0, the expand value has no effect until
it reaches exactly 1. This change makes it gradual.
The Vector Transform node is a useful node which is present in the Cycles renderer.
{F144283}
This patch implements the Vector Transform node for GLSL mode and the internal renderer.
Example: {F273060}
Alexander (Blend4Web Team)
Reviewers: brecht, campbellbarton, sergey
Reviewed By: campbellbarton, sergey
Subscribers: psy-fi, duarteframos, RobM, lightbwk, sergey, AlexKowel, valentin_b4w, Evgeny_Rodygin, yurikovelenov
Projects: #bf_blender:_next
Differential Revision: https://developer.blender.org/D909
simulations.
This commits implements OpenVDB as an extra cache format in the Point
Cache system for smoke simulations. Compilation with the library is
turned off by default for now, and shall be enabled when the library is
present.
A documentation of its doings is available here: http://
wiki.blender.org/index.php/User:Kevindietrich/OpenVDBSmokeExport.
A guide to compile OpenVDB can be found here (Linux): http://
wiki.blender.org/index.php?title=Dev:Doc/Building_Blender/Linux/
Dependencies_From_Source#OpenVDB
Reviewers: sergey, lukastoenne, brecht, campbellbarton
Reviewed By: brecht, campbellbarton
Subscribers: galenb, Blendify, robocyte, Lapineige, bliblubli,
jtheninja, lukasstockner97, dingto, brecht
Differential Revision: https://developer.blender.org/D1721
Only re-enable blending if the function that disables it was actually
called.
Still not ideal flipping this on & off repeatedly, but now there are
fewer flips.
Each LINES draw call is now responsible for its own line width. No need
to set it back to its 1.0 default after every draw.
This eliminates half our calls to glLineWidth , similar to last week’s
work on glPointSize.
Calculate the clipped min/max factor along the segment,
only applying to the coordinates at the end (will give better precision too).
Also make split input/output args.
This changes the following defaults:
- Render settings:
* Samples: 100
* Preview Samples: 50
* Filter: Blackmann-Harris
* Tile Order: Hilbert Spiral
- Lamp settings:
* Use MIS: On
- Material settings:
* Volume Sampling: Multiple Importance
Old files are not affected, I tested the versioning code back and forth.
More changes are to come (World, BVH...) but that needs a bit more work.
Skipping computing of shadow pass when diffuse color is pitch black is fine... unless
you actually need/want that shadow pass!
The 'noisy' issue with picture texture remains a bit mysterious to me currently. :/
The previous threshold used to prevent the Graph Editor from imploding if
presented with a flat (or nearly flat, accounting for floating point precision)
curve was too coarse, meaning that in some cases, the "View All" tool would end
up behaving weirdly.
This brings the dopesheet more in line with the NLA and Graph Editors, where
similar initial ranges were also used. The benefit is that it should save
animators a small amount of time getting the dopesheet timeline into the
right zoom level before starting work.
UI_panel_category_draw_all was setting PolygonMode to LINES before
drawing LINES.
stitch_draw was setting PolygonMode to its default FILL value — any
function that deviates from the default should’ve changed it back to
FILL.
This commit adds a "Select Grouped" operator. Although it is set up to
allow more types of "grouping" in future, it current only supports
a single mode (i.e. "Same Layer"). As a result, it does not pop up
any menus/submenus in all the usual places.
Wireframe drawing doesn’t use the following GL state, so don’t update
these:
- ShadeModel
- NormalPointer
- ColorPointer
Normal & color data are still uploaded as part of the interleaved VBO,
but those attributes are disabled for wireframe.
I've been using this fix in another branch locally, so it seems to work fine.
The other #ifdef checks should be checked on too, as __MINGW32__ and __MINGW64__
do NOT seem to be defined when compiling that file
As reported on the Blender Institute Podcast 009. See my comment on the cloud blog
for further details.
When used a second (or third, etc.) time, the breakdowner's (Shift-E) percentage value
would initially be the last-used value (e.g. 33% or 75%), before suddenly jumping
to another value as soon as the mouse moves. The cause of this behaviour was that it
was initially reusing the value from the previous time the operator was run, but then
as soon as the mouse moved, it would snap to the percentage implied by the mouse position.
(Note: The mapping from mouse position to percentage is "absolute" - i.e. the percentage
is based on how far across the 3D view the mouse is, instead of being some kind of
relative offset thing).
To make things a bit less jumpy, I've changed the behaviour so that the mouse position
always gets used immediately, instead of having it jump suddenly only when making
some mouse movement.
Apply X and Y blur as separate step, this reduces number of accumulations
required and makes effect more realtime.
Another quick thing for the Nieve project.
Fix T47213.
There was actually no real bug here, just clarify now in the UI that Mesh, World and Lamp samples only have an effect if we sample all lights (direct or indirect).
The simplest way of handling mirroring in multi-paint is creating a
uniform symmetric selection and relying on existing symmetric weights
to direct changes to the appropriate vertex groups. This already works
if mirror bones are selected manually, and can be made easier to use
by doing it implicitly.
Use the collective weight when using blurring.
This brings blur in sync with the rest of multi-paint modifications.
Also, blur has no business calling `defvert_verify_index` anyway when
it simply wants to read weights.
Similarly, the displayed weight distribution should be used as the
target of brush application instead of the invisible active group.
Multi-paint is split into a new function because all the calls to
defvert_verify_index at the top of the old code should never be
done in multi-paint mode.
Since the coloring uses sum or average of the weights of all selected
groups, the weight pick tool should also use that instead of reading
the weight of the single active group that you can't see.
This condition can actually happen quite often if weight painting for
a rig that uses separate control bones, so the color shouldn't be so
bright that it's hard to look at for a significant amount of time.
The threaded code is twice quicker now (from an average of 20ms/frame to 10ms/frame
while baking 10000 particles here e.g.)!
Think this is mostly due to usage of 'dynamic' scheduler in OMP code though,
from my experience so far this tends to have dramatic effects over performances,
static scheduler is usually much much more efficient.
Problem is actually similar in both engines - in some cases, we changed
'natural' quad splitting order to alternative one, without properkly 'notifying'
UV/VCol/other tessface data about it.
So code would use a 'wrong' triangle of UVs etc.
Fix for Cycles was committed by sergey as rBa6eae7339190d1.
The issue was discontinuity in logic when importing vertices from blender
and then importing data layers regardless of how we split the face. Quite
interesting we didn't notice this issue before.
Thanks Bastien for the investigation, based on D1742 but redid it to make
patch a bit more clear to follow.
LoopTri changes in 2.76 calculated all tangents as triangles,
this gave different results though in most cases it was hard to notice.
Though no bugs were reported we should keep our tangents compatible with other users of mikktspace.
Specifically, when only one bone is selected and it's not really active.
(With multiple bones on the other hand that behavior is forced on,
since multi-paint can't modify purely zero weight verts and that's important.)
There is no function pointers in OpenCL specification. For as long
as we want to support this platform we should follow the specifications.
While the code is not totally optimal now, it should not be that huge
of performance issue on CPU since it does jump tables just nicely, so
it's not that much extra computation here.
Now image will be opened for while render session is active, this is
needed to keep image cache working correctly. But stopping render
should now release all files descriptors.
D1713 from @angavrilov (with own edits)
The way current weight paint code works is that instead of making normalization lock aware, a separate `enforce_locks` function is called to do a different kind of normalization, hoping that by the time real normalize is called, there is nothing for it to do. The problem is that:
- That different normalization is based on adding the same amount to all unlocked groups, whereas true normalize uses multiplication to preserve ratio between them. The multiplicative approach should match the way weights operate better.
- If `enforce_locks` fails to achieve perfect normalization, true normalize will change locked groups.
Try to fix this by replacing `enforce_locks` with true lock-aware normalization support.
Supporting locked groups in normalize means that freezing the active group can make full normalization impossible if too much weight was added or removed, so it may be necessary to do two normalize passes. This is similar to how enforce_locks operates.
Also, now there is no need to go through the multi-paint branch just because of the locks. In the actual multi-paint case, the same normalize code can be used via a temporary flag array that represents the union of selected and locked groups.
User-visible effect should be:
- Auto-Normalize doesn't change behavior vertex to vertex depending on whether it has any locked groups.
- It never changes locked groups; if you lock some groups and start painting with seriously non-normalized weights, it's assumed you intended that.
- It never adds vertices to new groups, since the computer can't do that intelligently anyway - it was especially broken in case of mirroring.
- It always acts to preserve the ratio between groups it changes, instead of (sometimes, see point 1) adding or subtracting the same amount.
Checking for 'Closest' here isn't needed since
TransSnap.snapTarget callback is already ensuring the selected target is the closest.
Also don't reuse the pre-calculated distance,
since its only valid to do this when there are no snap points
and this isn't a significant gain to avoid the extra calculation - run once per update.
This patch adds a new API which allow us to access light shadow settings from python. The new API can be used to write custom GLSL materials with shadows.
Reviewers: brecht, kupoman, agoose77, panzergame, campbellbarton, moguri, hg1
Reviewed By: agoose77, panzergame, campbellbarton, moguri, hg1
Projects: #game_engine
Differential Revision: https://developer.blender.org/D1690
This reverts commit b5bd39b518.
This is problematic because it means minor differences in system headers can cause errors
(glew in this case errored for me both with system and Blender's bundled version).
Developers can enable -Werror=* locally for warnings that aren't reliable across different systems.
This is how the old code used to initialize it. Current value was
failing big time when baking all caches (always set to MAXFRAME), and it
was added right before commit to quiet a warning... (I know, I know...)
This is the second time at least that kind of stuff slips in during one of my refactor,
a mere warning here is really not enough (too easy to miss it)!
Issue was with very thin domains along one or two axes, these could lead to simulation
with only one cell width - and smoke code assumes we have at least 4 cells in each direction.
So now, we clamp resolution to a minimum of 4 in smoke_set_domain_from_derivedmesh().
Note: in extreme cases like this report, this will generate very un-cubic cells,
check it still works OK in 3DView is needed here.
Thanks to @genscher and @kevindietrich for help on this issue. :)
Based on usages so far:
- Split callback worker func in two, 'basic' and 'extended' versions. The former goes back
to the simplest verion, while the later keeps the 'userdata_chunk', and gets the thread_id too.
- Add use_threading to simple BLI_task_parallel_range(), turns out we need this pretty much systematically,
and allows to get rid of most usages of BLI_task_parallel_range_ex().
- Now BLI_task_parallel_range() expects 'basic' version of callback, while BLI_task_parallel_range_ex()
expectes 'extended' version of the callback.
All in all, this should make common usage of BLI_task_parallel_range simpler (less verbose), and add
access to advanced callback to thread id, which is mandatory in some (future) cases.
Our RenderSettings struct may have a bit too much levels, makes it hard to track
all 'pointer' data during copy...
Fixed several issues here, but not sure I found all existing ones. :/
Displacement shader was not updating motion vertex positions.
Current solution is not totally correct because it applies same offset
for all time steps. Ideally we'll need to evaluate displacement shader
for every time offset separately, but currently we don't have subframe
image access.
For the time being will consider this a TODO.
I put all usage of GL_POINTS under the microscope. Fixed problems &
optimized a couple of spots.
- reduce calls to glPointSize by about 50%
- draw selected & unselected vertices together for UV editor & EditMesh
- draw initial gpencil stroke point the proper size
- a few other smaller fixes
New policy: each GL_POINTS draw call needs to set its desired point
size. This eliminates half our calls to glPointSize (setting it back to
its 1.0 default after every draw).
Compiling OSL scripts with errors in them would cause Blender to crash since the OSL version
bump to 1.6.9 instead of printing the error to the console as it did before.
With version 1.6.2, OSL added a pointer to an OpenImageIO ErrorHandler as an argument to the
OSLCompiler constructor. However, since it defaults to the NULL pointer, Blender still compiled
fine after the OSL version bump.
It turns out, though, that this pointer is used without further checks inside the OSL code, which
makes it crash when it tries to report an error unless a valid ErrorHandler pointer is specified.
Therefore, this commit simply passes a pointer to the static default handler that OIIO offers,
which prints the error to the console just like OSL did before.
Using this feature for a more advanced error handling and displaying from the Blender side would
be possible and seems reasonable, but for now it's not really relevant for fixing this bug.
The combined pass is built with the contributions the user finds fit.
It is useful for lightmap baking, as well as non-view dependent effects
baking.
The manual will be updated once we get closer to the 2.77 release.
Meanwhile the new page can be found here:
http://dalaifelinto.com/blender-manual/render/cycles/baking.html
Reviewers: sergey, brecht
Differential Revision: https://developer.blender.org/D1674
- Only do this if PBVH is not used for display, this should correspond
to whats' happening in sculpt.c now.
- No need to do such synchronization for solid drawing, it supports
optimal display from PBVH.
In fact, such synchronization is only needed in the following case:
Drawing callback does not support PBVH draw and sculpt session is
configured to use PBVH for display and related operations.
Previously it was simplier logic which only checked whether the mouse
is inside of some area around sliding zone, which was resulting in
wrong sliding zone used when zoomed out.
924c626 broke loading on systems with different endian.
We could support reading values from SDNA, it isn't really worth the extra hassle.
Long support is still removed, just keep enum values the same.
This commit removes the experimental CUDA kernel, making SSS and CMJ
regular features.
Several improvements have been made in the past few
weeks (thanks Sergey!) which make SSS render several times faster (2-3x
compared to 2.76b) on the GPU, and the increased VRAM usage has also been
fixed. Therefore the experimental kernel is no longer needed.
Differential Revision: https://developer.blender.org/D1726
Manual has been updated: too:
https://www.blender.org/manual/render/cycles/features.html
It is handy when doing some roto work and it's required to blur some
mask or overaly before alpha-overing it on top of the footage.
Quite straightforward option with the only limitation that variable
size blur is not supported.
Reviewers: campbellbarton
Subscribers: hype, sebastian_k
Differential Revision: https://developer.blender.org/D1663
This patch aims at providing multiple compression modes for TIFF output,
particularly uncompressed mode.
At this moment have None, Deflate, LZW and Pack Bits modes been integrated,
mimicking The GIMP export modes (except JPEG mode, which returned encoding
errors).
More modes could be added if needed.
Default remains Deflate.
Reviewers: campbellbarton, mont29, sergey
Differential Revision: https://developer.blender.org/D1709
The goal is to make Experimental kernel closer in performance to the
official kernel, avoiding spills and such.
There should not be big impact on official kernel, own tests showed
few percent performance drop on laptop's GPU. CPU was always the
same speed on AVX, AVX2 and SSE4.1 CPUs i've been testing here.
This seems to be the last essential step before we can get rid of
Experimental kernel and enable SSS officially on GPU without causing
some major performance issues.
Surely some more tweaks are possibly required, but that we can do
for until cows go home anyway.
Should be no functional changes at all, just speeds up re-compilation
when some features needs to be disabled for development purposes.
For example, when running lots of Valgrind it's handy to disable any
GPU devices because otherwise you'll be wasting quite some time in
the driver while enumerating devices.
Reviewers: dingto, lukasstockner97, brecht, juicyfruit
Differential Revision: https://developer.blender.org/D1730
Previously several areas were calling TEST_SHARED_PTR_SUPPORT and
TEST_UNORDERED_MAP_SUPPORT which isn't that bad on it's own but
was causing some quite verbose output with same information line
printed multiple times. additionally, what's more worse, define flags
for Ceres were duplicated in main CMakeLists and Ceres's CMakeLists.
Now we've got a single place where checks for those classes are
happening and other areas are simply checking for variables set by
those check macros, keeping CMake output clean and nice.
The main purpose of such linking is to make Blender compatible with
NVidia's debuggers and profilers which are doing some LD_PRELOAD
magic to intercept some function calls. Such magic conflicts with
our CUDA wrangler magic and causes segmentation faults.
The option is disabled by default, so there's no affect on any of
artists.
In order to make Blender linked directly against CUDA library use
the WITH_CUDA_DYNLOAD CMake option (it's marked as advanced).
Reported as part of T47123
Checks for locked groups assumed auto-normalize
and would distribute wight between groups even with auto-normalize disabled.
Regression in 4d33c37c9
Only copy normal arrays from sculpt to the DerivedMesh when the mesh is deformed.
Constructive modifiers calculate their own normals.
How to check?
1) Set Rendered mode for 3d View
2) Set "Blend Sky" and "Real Sky" flags for world.
Rotate view you will see right color blending from Horizon to Zenith color.
3) Add environment texture
Rotate view and you will see incorrect texture mapping that is not correspond to color blending that was observed in previous step.
So I conclude that this is a bug.
This issue must be solved before the patch D1414 because BI behavior is the viewport target.
Example: {F273051}
Alexander (Blend4Web Team)
Reviewers: sergey
Reviewed By: sergey
Subscribers: Evgeny_Rodygin, AlexKowel, yurikovelenov
Differential Revision: https://developer.blender.org/D1729
This panel is only visible when debug_value is set to 256 and has no
affect in other cases. However, if debug value is not set to this
value, environment variables will be used to control which features
are enabled, so there's no visible changes to anyone in fact.
There are some changes needed to prevent devices re-enumeration on
every Cycles session create.
Reviewers: juicyfruit, lukasstockner97, dingto, brecht
Reviewed By: lukasstockner97, dingto
Differential Revision: https://developer.blender.org/D1720
We cannot use FLT_MAX as initi distance for raycast...
Renamed TRANSFORM_DIST_MAX_RAY to BVH_RAYCAST_DIST_MAX, moved it into BLI_kdopbvh,
and use in RNA raycast callbacks (and all other places using that API).
In fact, code would behave strangely here with any negative value, applying a 'wrapped positive clamping'
(comes from original 2.5 commit feature, rBfafbd9d71b95776d1c7583476de74fccefab7f10)...
This commit is conservative - it keeps same behavior for all particle properties affected by textures,
except for size and length, where we apply a real [0, 1] clamping.
Easy to change in future in case new odd cases popup.
Actually, it was the whole 'cloth' simulation of hairs that was broken,
since grid data could not be initialized correctly.
Stupid mistake from looptri conversion, rBc582e186d90291a19a4e404111c492f1fd2c41a4
Icon buttons with text labels now use the same color as labels.
This was causing problems with theming since there was no way to set color
for all labels without also changing button text color.
When the mirrored vertex group was created on the first stroke,
the mirrored weight failed to normalize.
aused by the valid vgroup map being created before adding the mirror vertex group.
When using D+RMB using a tablet (e.g. holding down the side button of the stylus
while hovering it over the surface of the tablet) to erase, the tablet would
report zero-pressure. This causes problems when using the new pressure-sensitive
Grease Pencil eraser, causing it to have no effect.
Screenshot ignored errors, some render code printed 'Saved' without checking for failure.
note: errno is now cleared from IMB_saveiff so all callers don't need to.
mode index instead of mode name can be used when opening the card:
VideoDecklink("1/2vuy",0)
The list of video mode is specific to each board and this method is
therefore not generic. Only use it for debugging purposes.
VideoDecklink will not throw if dvp.dll is not found when opening a Decklink
card on Windows with a nVideo Quadro GPU; instead it will fallback on
standard OpenGL transfer. It is preferable to use dvp.dll when possible as it
provides faster image transfer to Quadro GPU.
Audio Input is now enable by default with 48KHz sampling, 16bit samples and
2 channels. This is normally not necessary because audio is not used in
VideoTexture; it is just in case it is necessary on certain Decklink card.
Active color wasn't copied over if it differs from what the user plugged
in in the UI. Also use a darker color for the default color so smoke
doesn't doen't appear too bright.
Reported in IRC by mib2berlin.
Beside the obvious ARB -> GLSL change, the texture slicing algorithm had
to be rewritten.
Although this new algorithm has the same behaviour as the old one (view
aligned slicing), it works with an arbitrary number of slices (which
could eventually be set by the user), which means we can preallocate the
buffer. The previous algorithm would slice from the begining to the end
of the volume's bbox, and draw the slices as it generates them.
Also support for ARB program was removed.
Patch by myself, with some minor fixes by Brecht.
Reviewers: brecht, #opengl_gfx
Differential Revision: https://developer.blender.org/D1694
Although the code made it impossible to use time_start_ uninitialized, at least GCC did
still produce multiple warnings about it.
Since time_dt() is an extremely cheap operation and functionality does not change in any way when
removing the check in the constructor, this commit removes the check and therefore the warning.
This patch adds the "Hilbert Spiral", a custom-designed continuous space-filling curve, as a tile order for rendering in Cycles.
It essentially works by dividing the tiles into tile blocks which are processed in a spiral outwards from the center. Inside each
block, the tiles are processed in a regular Hilbert curve pattern. By rotating that pattern according to the spiral direction,
a continuous curve is obtained, which helps with cache coherency and therefore rendering speed.
The curve is a compromise between the faster-rendering Bottom-to-Top etc. orders and the Center order, which is a bit slower,
but starts with the more important areas. The Hilbert Spiral also starts in the center (unless huge tiles are used) and is still
marginally slower than Bottom-to-Top, but noticeably faster than Center.
Reviewers: sergey, #cycles, dingto
Reviewed By: #cycles, dingto
Subscribers: iscream, gregzaal, sergey, mib2berlin
Differential Revision: https://developer.blender.org/D1166
In fact, it was not working with BI either - 'UV from dupli' would always take active UVLayer,
not render_active one.
Fixed now for both Cycles and BI, and for both particles and 'simple' dupli_face.
Code 'localizing' the node of an ID (a material, here) was kind of a joke,
setting id.lib to NULL is not enough, by far! Now calling ntreeMakeLocal().
And ntreeMakeLocal was also missing indirect->extern switch for its directly used IDs,
which explains why they were lost after a save/reload.
Note that here again, a big part of this 'localizing' code should be made generic
using library_query stuff, but this is for id-remap, not current master...
This change is for a few reasons:
- it works with color, and (therefore) will need to be color managed, at
some point. This will be much easier to do if the code is closer to the
actual color management code (in Blender's core, so to speak).
- it has nothing to do with the actual fire simulation, as it is just
used to create a lookup table
- it can be reused for other purposes (i.e. in Blender internal
renderer, if people are interrested in a blackbody node à la Cycles)
- cleanup: some functions (`contrain_rgb`, `xyz_to_rgb`) already exist
in BLI
Reviewers: brecht
Reviewed By: brecht
Subscribers: brecht
Differential Revision: https://developer.blender.org/D1719
As in the title. In the smoke version, there was also an extra
'for_render' parameter that wasn't used, and wasn't used by the callers
either, so it was removed altogether.
Reviewers: brecht
Reviewed By: brecht
Subscribers: brecht
Differential Revision: https://developer.blender.org/D1718
Also added a DEBUG_TIME macro in the related files to comment time funcs out.
Reviewers: brecht
Reviewed By: brecht
Subscribers: brecht
Differential Revision: https://developer.blender.org/D1717
- Fixed a bug that the sound when changed in the properties panel was not updated.
- Added the option to make a sound mono while adding a sound strip.
- Added the option to make a sound mono in the sequencer properties panel.
Related bug report: T47140
There were 2 issues:
- toolbars were set initialized in user-defaults
so their scroll & zoom level were set.
- initializing new 2d views included the scroll width,
which scaled the new views zoom level, especially when dragging out.
In response to feedback from plasmasolutions that having this option "always on"
caused some examples of more destructive errors, this commit introduces an attempt
at seeking a better balance between doing what people expect when casually importing
clips, and not clobbering existing timelines.
Specifically:
1) When there's nothing in the scene, the option will default to being enabled.
This accounts for users doing this for the first time with no strips
2) When there are already sequence strips, the option will default to being disabled.
In this case, changing the framerate could be dangerous. Hopefully this new check
is sufficient.
3) There's always still the option for users to control what exactly happens. However,
there have been reports that this cannot always be seen/set?
Issue is with rounding up of length reported by audaspace for audio part - when it matches nearly exactly
the actual video length, using ceil() would make it one frame longer. Now apply a small (0.0001 frame)
negative offset to prevent this effect.
When importing new movie strips into the sequencer, the scene's frame rate
will now be adjusted to match that of the movie being loaded by default.
To get the old behaviour (e.g. if importing a clip into an existing project),
disable the "Use Movie Framerate" option in the file browser when selecting
the strip.
This change is designed to solve the common problem that users trying to import
video clips will forget to adjust the frame rate before importing the clip, thus
causing the sound and video strips to be out of sync (as the sound strip ends
up longer/shorter than the video stream).
If anyone finds OS X UI drawing glitches with different graphics cards please
report them and I'll add an exception specifically for Intel, but in theory this
should work fine for all graphics cards.
Everything I've tested works fine without these hacks. Variety of Mac, Linux, Windows, Intel, nVidia, AMD.
If these workarounds are for old unsupported systems let's clean house.
Reviewers: #opengl_gfx
Differential Revision: https://developer.blender.org/D1707
Similar to object mode 'Select All by Type',
This menu is for various options to select based on some rules.
Moving into a submenu removes clutter, and means we can add other selection types.
This uses selection history to select the next vert/edge/face based on surrounding topology.
Select previous just removes the last selected element.
Uses key-bindings: Ctrl-Shift +/-
This isn't really ideal fix, but currently i can't tell how proxies should work
with the new dependency graph. This backward links are crazy, non-threadsafe
and mindblowing. Need to do some smarter/easier system.
While previous code was already compiling with OSL 1.6 it was using some symbols
which were considered deprecated in upstream.
This commit adds some ifdefs, but soon we'll get rid of all them rather soon
with the upcoming OIIO/OSL update.
Patch from be28706 made it so integrator will use last shader's transparent
shadow flag, which is wrong since last shader might not have transparent
shadow while shaders prior to it might have one.
CYCLES_OPENCL_TEST was removed, there was an insonsistency between
opencl_kernel_use_split() and opencl_get_usable_devices().
From now on, to test non whitelisted devices please use either
CYCLES_OPENCL_MEGA_KERNEL_TEST or CYCLES_OPENCL_SPLIT_KERNEL_TEST.
This commit changes the way how we pass bounce information to the Light
Path node. Instead of manualy copying the bounces into ShaderData, we now
directly pass PathState. This reduces the arguments that we need to pass
around and also makes it easier to extend the feature.
This commit also exposes the Transmission Bounce Depth to the Light Path
node. It works similar to the Transparent Depth Output: Replace a
Transmission lightpath after X bounces with another shader, e.g a Diffuse
one. This can be used to avoid black surfaces, due to low amount of max
bounces.
Reviewed by Sergey and Brecht, thanks for some hlp with this.
I tested compilation and usage on CPU (SVM and OSL), CUDA, OpenCL Split
and Mega kernel. Hopefully this covers all devices. :)
This mainly adds bpy.data.user_map() method, which goes over the whole Main database
to build a mapping (dict) {ID: {users_of_that_ID}}.
Very handy to check and debug ID usages, but could also be really valuable for py addons
creating temporary scenes, or some exporters, etc.
Note: current code in master's libquery misses some IDs (and reports some it should not,
like nodetrees), this is fixed in id-remap but still needs serious review before going to master.
This basically means that current bpy.data.user_map() **will not** report a complete and exhaustive
state of dependencies between IDs. Should work OK in most cases though.
Original work/idea comes from id-remap branch, was heavily reworked by @campbellbarton
and myself for master.
Reviewers: campbellbarton, sergey
Differential Revision: https://developer.blender.org/D1678
The issue was caused by wrong scaling applied on top of the image sampling.
It seems there's no reason to apply alpha clipping when using EWA filtering,
this filtering currently works in the way that it apples all needed clipping
and scaling to a clipped kernel already.
Note that this has little impact on current master - yet it allows to fix the 'Image Editor' bug
(open image in editor, use same image in texture, delete image from texture, image us is 0 and
red in image editor...) and probably a few other similar cases.
But that change is mandatory to get a proper handling of ID deletion/reamapping/reloading/etc. as done in id-remap branch.
Instead of just adding a user if none already present, new code use two new ID tags to get a three-states status:
- Normal: nothing changes.
- Needs extra user: we know that ID needs an extra user, so we take of never going down to 0 in 'real' usercount handling.
- Has extra user: we do have increased that ID usercount to get our needed extrauser.
Reviewers: sergey, campbellbarton
Differential Revision: https://developer.blender.org/D1696
This can happen quite often in forloops, and would be annoying to have to check for this
in caller code! So now, just return without doing anything in this case.
Looptri are not mesh data, they are allocated and built for PBVH only, and totally 'local' to it,
so duplicating like we do for verts & co leads to leaking memory.
The remaining bits in editors' sculpt.c, which is now free from OMP.
All in all, changes in this file allow from below 2% to above %20 speedup per stroke, depending
on brush complexity, size, and definition of the mesh below it...
Note that this commit also unifies all 'threaded data' into a single struct, a few more
floats/pointers do not hurt here, and it makes code a bit simpler.
The idea is to split them into two separate targets and have dedicated include
directories list for each of them in order to avoid some annoying include header
modifications in comparison with upstream.
Reviewers: campbellbarton, juicyfruit
Differential Revision: https://developer.blender.org/D1706
It also fixes another issue (crash) related to symmetric editing.
Quite involved, we (try to!) fix complete broken logic of parts of particle code, which would use poly index
as tessface one (or vice-versa). Issue most probably goes back to BMesh integration time...
This patch mostly fixes particle editing mode:
- Adding/removing particles when using generative modifiers (like subsurf) should now work.
- Adding/removing particles with a non-tessellated mesh (i.e. one having ngons) should also mostly work.
- X-axis-mirror-editing particles over ngons does not really work, not sure why currently.
- All this in both 'modes' (with or without using modifier stack for particles).
Tech side:
- Store a deformed-only DM in particle modifier data.
- Rename existing DM to make it clear it's a final one.
- Use deformed-only DM's tessface2poly mapping to 'solve' poly/tessface mismatches.
- Make (part of) mirror-editing code able to use a DM instead of raw mesh, so that we can mirror based on final DM
when editing particles using modifier stack (mandatory, since there is no way currently to find orig tessface
from an final DM tessface index).
Note that this patch is not really nice and clean (current particles are beyond hope on this side anyway),
it's more like some urgency bandage. Whole crap needs complete rewrite anyway,
BMesh's polygons make it really hard to work with current system (and looptri would not help much here).
Also, did not test everything possibly affected by those changes, so it needs some users' testing & validation too.
Reviewers: psy-fi
Subscribers: dfelinto, eyecandy
Maniphest Tasks: T47038
Differential Revision: https://developer.blender.org/D1685
While SCons building system was serving us really good for ages it's no longer
having much attention by the developers and started to become quite a difficult
task to maintain.
What's even worse -- there started to be quite serious divergence between SCons
and CMake which was only accumulating over the releases now. The fact that none
of the active developers are really using SCons and that our main studio is also
using CMake spotting bugs in the SCons builds became quite a difficult task and
we aren't always spotting them in time.
Meanwhile CMake became really mature building system which is available on every
platform we support and arguably it's also easier and more robust to use.
This commit includes:
- Removal of actual SCons building system
- Removal of SCons git submodule
- Removal of documentation which is stored in the sources and covers SCons
- Tweaks to the buildbot master to stop using SCons submodule
(this change requires deploying to the server)
- Tweaks to the install dependencies script to skip installing or mentioning
SCons building system
- Tweaks to various helper scripts to avoid mention of SCons folders/files
as well
Reviewers: mont29, dingto, dfelinto, lukastoenne, lukasstockner97, brecht, Severin, merwin, aligorith, psy-fi, campbellbarton, juicyfruit
Reviewed By: campbellbarton, juicyfruit
Differential Revision: https://developer.blender.org/D1680
nVidia Linux driver reports GL_CONTEXT_PROFILE_MASK = 0, which is a bug.
In that case check for the ARB_compatibility extension.
Non-buggy drivers will continue to use
GL_CONTEXT_COMPATIBILITY_PROFILE_BIT.
Thx to Dr Hackerman for reporting.
This time, all tools' code itself.
Not much to say, except that we can also get rid of that OMP caching pre-process ugly stuff
for multires smoothing.
Together with previous commit, we have about 5% average speedup on stroke execution
(though this vary a lot, up to 30% speedup in rare cases, and in even rarer cases some
odd massive slowdowns...).
Tech note: we may want to add 'guided'-similar feature to our BLI_task threaded loop,
I suspect this could explain random massive slowdowns of new code (very rare, but annoying...).
The issue was caused by OSL using TLS which is required to be freed before the
Cycles session is freed. This is quite tricky to do in Cycles because different
render session are sharing the same task scheduler, so when one session is being
freed TLS might need to be active still.
In order to solve this, we are now doing JIT optimization ahead of the time
which ensures either TLS of JIT is freed before the render on multi-core system
or freed on OSLRenderSession destroy on single-core system.
This might increase synchronization time due to JIT of unused function, but
that we can solve later with some smart idea,
Logic here is weird. Generally, textured drawing overrides
material color unless material uses object color instead?
It doesn't make sense, material color is a material color
whatever the circumstance.
Repeating: idea as always is to push all those options out of per-
polygon callbacks and make decisions such as color enable in the higher
level functions.
It would be good to get rid of this entirely, ideally decision
about mcols can be taken at material level and not done per face. More
work needs to be done for that to work though.
Only used for color for now, but we need this for any kind of buffer
updates actually.
This should get rid of some allocation/deallocation, making
vertex painting a bit faster.
This commit overrides the user's choice of tile order in the case of viewport rendering and always uses bottom-to-top instead.
This was already done until the TileManager redesign, but since it removed the distinction between viewport and regular rendering
in the manager, the viewport was now also using the selected order. Since this requires sorting of the generated tiles,
it slows down rendering a bit. With the forced bottom-to-top order, this sorting step can now be avoided again.
Since the tile order is invisible anyways for viewport rendering, this commit won't have any impact on users (apart from a slight speedup).
This commit adds "Bands Saw" and "Rings Saw" to the options for the Wave texture node in Cycles, behaving similar to the Saw option in BI textures.
Requested by @cekuhnen on BA.
Reviewers: dingto, sergey
Subscribers: cekuhnen
Differential Revision: https://developer.blender.org/D1699
This gives usual nice boost in parallelized sections themselves, on global sculpting stroke
it's a bit hard to say, seems like we have an average 5% speedup, but it varies a lot...
Note that this area is only parallelized when you have a big brush over dense geometry, otherwise
there's no change at all.
As a side note, this commit also factorizes code here, three times nearly the same code, tst...
This is an attempt to emulate real CMOS cameras which reads sensor by scanlines
and hence different scanlines are sampled at a different moment in time, which
causes so called rolling shutter effect. This effect will, for example, make
vertical straight lines being curved when doing horizontal camera pan.
This is controlled by the Shutter Type option in the Motion Blur panel.
Additionally, since scanline sampling is not instantaneous it's possible to have
motion blur on top of rolling shutter.
This is controlled by the Rolling Shutter Time slider which controls balance
between pure rolling shutter effect and pure motion blur effect.
Reviewers: brecht, juicyfruit, dingto, keir
Differential Revision: https://developer.blender.org/D1624
Vector mapping node was doing some weird mapping of both original and mapped
coordinates. Mapping of original coordinates was caused by the clamping nature
of the LUT generated from the node. Mapping of the mapped value again was quite
totally obscure -- one needed to constantly keep in mind that actual value will
be scaled up and moved down.
This commit makes it so values in the vector curve mapping are always absolute.
In fact, it is now behaving quite the same as RGB curve mapping node and the
code could be de-duplicated. Keeping the code duplicated for a bit so it's more
clear what exact parts of the node changed.
Reviewers: brecht
Subscribers: bassamk
Differential Revision: https://developer.blender.org/D1672
eg:
blender -b --python-exit-code 1 --python script.py --render-anim
This causes blender not to continue parsing command line arguments and exit if the script raises an exception.
This gives about 2x speedup (3.2sec vs. 11.9sec with 32716 handled nodes) when
updating shader for the shader tree.
Reviewers: brecht, juicyfruit, dingto, lukasstockner97
Differential Revision: https://developer.blender.org/D1700
From recent experience, turns out we often do want to use something else than basic
range of parallelized forloop as control parameter over threads usage, so now BLI func
only takes a boolean, and caller defines best check for its own case.
This makes it possible to move some parts of evaluation from host to the device
and hopefully reduce memory usage by avoid having full RGBA buffer on the host.
Reviewers: juicyfruit, lukasstockner97, brecht
Reviewed By: lukasstockner97, brecht
Differential Revision: https://developer.blender.org/D1702
Main goal is to make kernel signatures editing easier and less prone to the
errors caused by missing function signature update or so.
This will also make it easier to add new CPU architectures.
Reviewers: juicyfruit, dingto, lukasstockner97, brecht
Reviewed By: dingto, lukasstockner97, brecht
Differential Revision: https://developer.blender.org/D1703
The idea is to have separate sets per node name in order to speed up the
comparison process. This will use a bit more memory and slow down simple
shaders, but this extra memory is not so much huge and time penalty is
not really measurable (at least from initial tests).
This saves orders of magnitude seconds when de-duplicating 17K nodes and
overall process now takes 0.01sec on my laptop,
This time, with have over 300% speedup!
But no, this is not due to switch to BLI_task (which 'only' gives usal 15% speedup),
but to enhancement of the algorithm, flatten loop over covariance matrix items now allows
to compute (usually) all items in parallel, instead of having at most 3 or 4 working threads
(with unbalanced load even)...
Gives the usual 10%-30% speedup on affected functions themselves (BLI_bvhtree_overlap() and
BLI_bvhtree_balance()), and about 2% speedup to overall cloth sim e.g. (measured from
main Cloth modifier func).
Also rearranged code here to not issue a draw call (explicit flush) per
face and not set shader per face either when stippled drawing is mixed
with regular drawing. Not good at all for performance.
Use Summary structure to collect all summary related on the shader compilation
process which then could be either simply reported to the log or be passed to
some user interface or so.
This is type of the summary / report which is most flexible and useful and
something we could use for other parts like shader optimization.
It's quite small chance we'll be supporting suitesparse for Blender due to all
the complexity of 3rd party libraries, so removing implementation files which
are only needed when suitesparse is enabled.
The idea of this commit is to merge nodes which has identical settings
and matching inputs into a single node in order to minimize number of
SVM instructions.
This is quite simple bottom-top graph traversal and the trickiest part
is how to compare node settings without too much trouble which seems to
be solved is quite clean way.
Still possibilities for further improvements:
- Support comparison of BSDF nodes
- Support comparison of volume nodes
- Support comparison of curve mapping/ramp nodes
Reviewers: brecht, juicyfruit, dingto
Differential Revision: https://developer.blender.org/D1673
This snippet creates a white balance modifier for the video sequence editor. It is useful for everyone who likes to set a new white point in the video source (easily via the eyedropper). Just select a point in the source file where you know that it should be white. The algorithm will then shift the colors towards your new white point.
See attached the image for a quick demo.
{F270576}
Reviewers: psy-fi
Reviewed By: psy-fi
Subscribers: Blendify
Projects: #bf_blender
Differential Revision: https://developer.blender.org/D1698
Gives a global speedup of about 5% in smoke simulation (as usual, parallelized chunks themselves
are about 15-25% quicker with BLI_task than with OMP), using a simple setup with two generators
(one from mesh, one from particles), an obstacle and a windfield.
When called with very small range, `BLI_task_parallel_range_ex()` would generate a zero `chunk_size`,
leading to some infinite looping in `parallel_range_func` due to `parallel_range_next_iter_get` returning
true without actually increasing the counter!
So now, we ensure `chunk_size` and `num_tasks` are always at least 1 (and avoid generating too much tasks too).
Issue was that dispatchEvent might call removeWindowEvents/
removeTypeEvents which will delete the event before we can do so.
To address this, handled events are now put in a separate list.
Reported by psy-fi and reviewed by brecht in IRC.
* Don't copy name before entering new_id function. new_id does that for
us already.
* Take a main argument to make the function possible to use with
different databases
* Append BKE_ to rename_id
renamed to Rotate to match the tool name.
It is not actually compulsory that the two must match since users can
change the name of a brush arbitrarily but at least try to have
consistent naming in our defaults.
The events are allocated on the heap, then pushed on a stack. Before
being processed, they are popped from the stack, and deleted after
processing is done. When the manager is destroyed (e.g. application
closing), any remaining event in the stack is detroyed.
Issue is that when the "application closing" event is processed, it is
never freed, because the manager gets destroyed before the call to
`delete` is made and the event is not on the stack anymore.
Now events are left on the stack while they are processed, and only
popped and deleted after processing is done.
As a slight bonus refactor: use void as return type for dispatch events
functions, as no caller is checking the return value, and it is not
clear what it means (suggested by the reviewer).
Reviewers: brecht
Differential Revision: https://developer.blender.org/D1695
Historically blender had an audio sample rate of 44.1 kHz as default which is mostly popular because it's the sample rate of audio CDs. Audaspace kept using this default from the pre 2.5 era. It was about time to change to 48 kHz, which is a more widespread standard nowadays, especially in video. It is the recommended sampling rate of the Audio Engineering Society.
Further reading: https://en.wikipedia.org/wiki/44,100_Hz#Status
This is purely internal sanitizing/cleanup, no change in behavior is expected at all.
This change was also needed because we were getting short on ID flags, and
future enhancement of 'user_one' ID behavior requires two new ones.
id->flag remains for persistent data (fakeuser only, so far!), this also allows us
100% backward & forward compatibility.
New id->tag is used for most flags. Though written in .blend files, its content
is cleared at read time.
Note that .blend file version was bumped, so that we can clear runtimeflags from
old .blends, important in case we add new persistent flags in future.
Also, behavior of tags (either status ones, or whether they need to be cleared before/after use)
has been added as comments to their declaration.
Reviewers: sergey, campbellbarton
Differential Revision: https://developer.blender.org/D1683
Allows to quickly select alternate elems in a path (matching checker-deselect options).
- adds redo support to MESH_OT_shortest_path_pick, allowing for other options.
- de-duplicates code between 2x path select operators.
- expose 'Topology Distance' property for path picking.
- remove unused 'extend' property.
cmake_qtcreator_project.py now takes a '--build-dir' argument.
Since introduction of argparse, accessing last argv from project_info is no longer working.
Now require a call to project_info.init before use.
The is intended to replace the deprecated glPolygonStipple() calls with a shader
based alternative, once we switch over to GLSL shaders.
Reviewers: brecht
Differential Revision: https://developer.blender.org/D1688
Usual 10%-15% speedup. Note that here OMP was rather erratic, with typical
timing about 10% slower than BLI_task, but having some 'burnouts' (~10% of times)
were it would be over ten times slower than usual... BLI_task looks much more stable.
Changes to remove doubles face creation,
Recent change to remove doubles broke when the new faces already existed (rare occurrence),
however theres no point to return an existing double face.
Now check if the face exists before creating it.
Other changes:
- avoid 2x hash lookups on all mapped verts.
- fill in the vert array instead of calculating from edges.
- remove inefficient search of entire edge-array before adding to it.
(flag verts to ensure they're not used multiple times).
- move logic for transfusing edge-flags to edge creation.
- When rendering in the Viewport, next_tile is sometimes called after a reset has been performed, but before
new tiles were generated. In that case, the tile list would be invalid, causing Blender to crash randomly.
- When generating new tiles, the TileManager would not clear the tile lists before re-generating them, leading
to some tiles being skipped during viewport rendering.
- When popping the next tile from a tile list, a reference to the just-deleted object would be returned, now the
object is copied before deleting it.
This way socket type conversions (such as color to float, or float to vector) do not stop the folding process.
Example: http://www.pasteall.org/pic/show.php?id=96803 (selected nodes are folded).
This commit modifies the TileManager to sort render tiles once after tiling the image,
instead of searching the next tile every time a new tile is acquired by a device.
This makes acquiring a tile run in constant time, therefore the render time is linear
w.r.t. the amount of tiles, instead of the quadratic dependency before.
Furthermore, each (logical) device now has its own Tile list, which makes acquiring
a tile for a specific device easier.
Also, some code in the TileManager was deduplicated.
Reviewers: dingto, sergey
Differential Revision: https://developer.blender.org/D1684
Delimiting on seams was only ever intended for face selection,
Previously this option didn't work for vertices,
now it's fixed the defaults aren't right for vertex/edge select-linked.
Add a workaround that bypasses operator-defaults - since this is such a rare case.
Reasons:
- Only parallelized piece of code gives little local speedup (code block only about 25% quicker even on 1M polys cube).
- No gain nor loss using new BLI_task system.
- At 10% of build, parallelized piece of code is only about 5% of total func runtime (run-time explodes as build proportion increases).
See no point in adding (in utmost best optimistic case, in real use-case, when depsgraph will likely already fire several evaluations in parallel,
speedup would be even smaller) 1% speedup here at the cost of threading complexity...
Note that since later code uses hashes, I don't think it's easy to thread it, so think we can leave with it for now.
- use 'Intersect' prefix for boolean and regular intersection.
hints that they use the same basic logic with different modes.
- 'split by edges' while correct - isn't very descriptive.
This was causing some random black bakes. It was introduced in of of the commits replacing calloc by malloc. Things should be better now.
This error was only noticeable when baking the selected objects (not when baking from selected to active).
- Remove deprecated/unused builders
- Remove unused SCons OSX slave configuration
- Remove SCons slave logic, it is not giving error about unknown building
system used for the slave.
Previous implementation was using GL_RED_INTEGER texture format for
all non-RGBA pixel format. This isn't supported by intel GPU (although
the GL_texture_rg extension is declared supported).
Floating RGBA texture are now used for the following pixel formats:
8BitYUV: Cb->B, Y0->G, Cr->R, Y1->A
10BitYUV: no fixed mapping between CbCrY and RGB, a shader is required
8BitARGB: direct mapping
8BitBGRA: direct mapping
10BitRGBXLE: direct mapping (A is undefined)
Other pixel formats are mapped to GL_RED_INTEGER (GL_R32UI internal
format, usampler2D must be used in the shader).
Note: the 10BitYUV, 10BitRGBXLE and 8BitARGB mapping only works on little-
endian host CPU. For big endian CPU, other formats must be used (not yet
implemented).
The texture MIN/MAX_FILTER is now set to NEAREST. Previously it was set to
LINEAR, which was detremental on multibytes pixel format.
The sample shader in the documentation for the above formats will be
changed to reflect the new mapping.
When we have an active button in modal state, completely bypass the whole 'pie' handling part of the menu.
Note that behavior is probably still not ideal here (e.g. would be nice to avoid enter/escape to quit
completely the pie menu in that case - but on the other hand Pies were not really designed to handle
that kind of modal stuff either, so think having minimal support for it is enough for now.
This is actually intended behavior to return NULL when the socket is not
found. It's used in certain BSDF nodes to query whether some inputs exists
or not.
Perhaps we can be more explicit here and have dedicated logic to query
socket existance and keep assert in place.
In any case, even if we lost assert() for the constant fold now it's
still somewhat better than duplicated code. Perhaps.
To ensure all items of a pie are always at the same position, invisible dummy buttons were added for unavailable items. This caused mainly two issues: Command line printed warning because of the > 8 elements, and some modes weren't visible in some cases ('Object Mode' entry was missing in stroke edit mode).
To solve this nicely, we had to support > 8 items per pie. Will look into that this week, but for now, drawing dummy buttons is disabled.
From a user POV this has two ugly consequences: 1. While this temporary workaround is used, *some* pie items are positioned differently than before, 2. The 'Edit Strokes' mode entry might change its position depending on the amount of available modes.
For now, we just change the labels of these toggles so that it makes sense what
results we should expect from each. A better solution may be to create separate
"direction" props for the different brush types here, so that we can also define
more appropriate tooltips too to go with the labels.
velocity is measured in pixels per frame. It is basically a coordinate
difference of track coordinate at current frame and previous one (no future
prediction happens).
It's not really most intuitive place for such a things, but historically the
node was called this way..
Track velocity could be used to face effects like motion blur bu piping it
to the vector blur node.
Reviewers: campbellbarton
Reviewed By: campbellbarton
Subscribers: hype, sebastian_k
Differential Revision: https://developer.blender.org/D1591
Changing path of a sound strip (select strip->C->'Path/Files') opened a file browser without filter for sound files, so sound files weren't visible.
Also, for movie/image files, now only movie **or** image files are visible in the file browser by default (instead of both).
Reported by @venomgfx, thanks!
* Added the ability to add a keyframe at the current cursor location in the
Graph Editor. This is useful for precisely defining the shape of driver F-Curves.
* Fixed a bug where the wrong cursor-x time was being used in Drivers mode
(i.e. it was still using time, and not just any time-value, but the NLA-mapped time!)
- show extensions one per line in order for easier human scanning
- show DrawElements vertex & index limits
- assume GL version >= 2 for GLSL limit queries
This option helps users protect themselves from accidentally changing the cursor
location (and not being aware of this until it has already caused problems)
when drawing using Grease Pencil (or with other tools where this is equally likely).
It seems to occur most frequently when using a tablet.
Currently, this only affects the use of the mouse to set the cursor, as this is
where most accidental invocations occur.
(I'm aware that this change may turn out to be quite contentious. Fortunately, it
should be simple to just revert this commit in that case :)
Use float in moto instead of double for MT_Scalar.
This switch allow future optimization like SSE.
Additionally, it changes the OpenGL calls to float versions as they are
very bad with doubles.
Reviewers: campbellbarton, moguri, lordloki
Reviewed By: lordloki
Subscribers: brecht, lordloki
Differential Revision: https://developer.blender.org/D1610
This patch improves clock management in BGE, to be able to accelerate /
slow the time, and also to finely synchronize clock with external
engines. Several new python functions have been added and existence ones
have been improved for that purpose. Now we have:
- getClockTime(): Get the current BGE render time, in seconds. The BGE
render time is the simulation time corresponding to the next scene that
will be rendered.
- getFrameTime(): Get the current BGE frame time, in seconds. The BGE
frame time is the simulation time corresponding to the current call of
the logic system. Generally speaking, it is what the user is interested
in.
- getRealTime(): Get the number of real (system-clock) seconds elapsed
since the beginning of the simulation.
- getTimeScale(): Get the time multiplier between real-time and
simulation time. The default value is 1.0. A value greater than 1.0
means that the simulation is going faster than real-time, a value lower
than 1.0 means that the simulation is going slower than real-time.
- setTimeScale(time_scale): Set the time multiplier between real-time
and simulation time. A value greater than 1.0 means that the simulation
is going faster than real-time, a value lower than 1.0 means that the
simulation is going slower than real-time. Note that a too large value
may lead to some physics instabilities.
- getUseExternalClock(): Get if the BGE use the inner BGE clock, or rely
or on an external clock. The default is to use the inner BGE clock.
- setUseExternalClock(use_external_clock): Set if the BGE use the inner
BGE clock, or rely or on an external clock. If the user selects the use
of an external clock, he should call regularly the setClockTime method.
- setClockTime(new_time): Set the next value of the simulation clock. It
is preferable to use this method from a custom main function in python,
as calling it in the logic block can easily lead to a blocked system (if
the time does not advance enough to run at least the next logic step).
Rationale are described more precisely in the thread
http://lists.blender.org/pipermail/bf-gamedev/2013-November/000165.html.
See also T37640
Reviewers: sybren, panzergame, #game_engine, lordloki, moguri
Reviewed By: sybren, panzergame, #game_engine, lordloki, moguri
Subscribers: moguri, hg1, sybren, panzergame, dfelinto, lordloki
Projects: #game_engine
Maniphest Tasks: T37640
Differential Revision: https://developer.blender.org/D728
Binding code was re-building its own DM for the cage, now it uses given one instead.
I cannot see really any good reason not to use 'visual' modified cage for binding process,
using base mesh instead was breaking any 'advanced' binding as described in the report.
C++ does not allow the assignment of a void pointer to a typed pointer without
explicit casting. Since we use a generic macro in bmesh for iterators we only
ever get a void* back and cannot cast it to the target type. However, casting
the target to a void* as well solves that issue.
This tweak is #ifdef'd to be used in C++ code only.
This uses a bmesh-intersection, BLI_kdtree and watertight intersections to perform boolean operations.
For now keep both BMesh and Carve booleans usable at once for testing & bug reports,
however we plan to phase out Carve by next release.
Works much the same as intersect operator,
expose as a new operator since for users its quite different.
Access from face menu.
Internally, this adds boolean args to BM_mesh_intersect function.
Performance is about the same or slightly better for typical IK chains.
In extreme cases with many bones and multiple targets, of which some are
unreachable, I've seen 2x speedups.
Those stupid ones only have one version of llvm (obviously not 3.4 one ;) ), so we have to build again
LLVM3.4 in those cases. Thing is,
* I did not update LLVM magic number when fixed a stupid typo breaking OSL building (the terminfo thing),
so many people were still using previously-built LLVM.
* Even worse, options passed to OSL to specify own LLVM from /opt/lib were wrong (not sure when this got
out of sync...).
Thanks to mib2berlin and slikdigit for the report & testings!
FBOs are a GL 3.0 feature but enjoy nearly universal support via
extensions.
The newer ARB extension brings these features to GL 2.1 without needing
an ARB suffix.
The older EXT extensions *do* use a suffix. Since we don’t know which
is used until runtime, I added the suffix to all functions & enums.
Also updated the check to look for the FBO feature set instead of the
specific EXT extension.
Maybe this is pedantic but I read it’s best to explicitly set the
desired component size.
Also append “_ARB” to float texture formats since those need an
extension in GL 2.1.
This class did nothing but print out extensions if they were found.
Instead, the code from bge.logic.PrintGLInfo() is now printed as the
Rasterizer is initialized. This gives better information, and it removes
some GL code from KX_PythonInit.cpp (the PrintGLInfo method now calls
the Rasterizer to print the information).
Differential Revision: https://developer.blender.org/D438
The only use we had for RAS_StorageIM was to render derived meshes using
Blender's mesh drawing. This is now handled as a special case in
RAS_OpenGLRasterizer instead of in RAS_StorageIM.
We are now left with RAS_StorageVA and RAS_StorageVBO. At the moment
vertex arrays are still the default since our vertex array with display
lists implementation is still much faster than our VBO code in a lot of
cases. As we improve our VBO code, we can drop vertex arrays since
Blender's minimum OpenGL version is being bumped up to 2.1, which
supports VBOs.
Use new GPU_legacy_support() function.
Determine GLSL version once instead of per shader.
For Texture Buffers, allow ARB or EXT version of the extension. Either
one will do.
Is current context compatible with legacy GL (version 2.1)?
My earlier approach -- checking for GLEW_ARB_compatibility -- was not
enough.
This should always return true if we set our GL context up properly. It
will return false when we switch to core profile.
In practice this gives us a context that is *compatible* with GL 2.1. On
my machine it gives a GL 3.3 or 4.3 compatibility profile context,
depending on graphics card installed.
Also fixed enum for core profile (not used yet).
Also added option for GL 3.2 compatibility profile. This will be useful
during Blender 2.8 development, until we are able to use the core
profile. On my machine this gives exactly a GL 3.2 compatibility profile
context, not 3.3 or 4.
My previous edit to this check was too lax.
OSD's shader for the Transform Feedback evaluator declares itself
#version 410 so disable the feature if user's GL < 4.1.
This way, connecting Value or RGB node to e.g. a Math node will still allow folding.
Note: The same should be done for the ConvertNode, but I leave that for another day.
Would be true in most cases (and in particular with own generated geometry),
but in case one would be using original geometry this could have crashed badly.
Note that I tried to parallelize the loops porting result of the simulation to the
DM data itself, but that ended up being 20% slower than non-threaded code!
Compared to previous revision, this gives 20% speedup on the whole modifier evaluation!
Wondering a bit how improvement can be so impressive here, would have expected very
small increases given how simple is the code here... Maybe it's the fact we get rid
of many additional OMP threads (tests are done with ten Ocean mod evaluated in parallel)?
Calling ImageRender.refresh() without passing a buffer object will
now perform the render (if not already started earlier with
ImageRender.render()) and wait for the render to complete.
This is done internally with a GLSync object to synchronize with the GPU.
This is only useful when doing offscreen render on texture target:
when refresh() returns, you are guaranteed that the texture is ready for use.
fbo = bge.render.offScreenCreate(width,height,samples,bge.render.RAS_OFS_RENDER_TEXTURE)
tex = fbo.color
ir = bge.texture.ImageRender(scene,camera,fbo)
ir.refresh()
--> tex is now ready for use
Previously RGB Curves node will clamp input to 0..1 which is rather useless
when one wants to use HDR image textures and do bit of correction on them.
Now kernel code supports extrapolation of baked LUT based on first/last two
table points and performs linear extrapolation.
The only tricky part is to guess the range to bake the LUT for. Currently
it's using simple approach -- minmax of the input curves. While this behaves
ok for the simple cases it's easy to trick the system up causing incorrect
results.
Not sure we can solve those issues in a general case and since the new code
is giving more expected results it's not that bad actually. In the worst
case artist migh always create explicit point to make sure LUT is created
for the needed HDR range.
Reviewers: brecht, juicyfruit
Subscribers: sebastian_k
Differential Revision: https://developer.blender.org/D1658
There was a bug with logic.setRender(False): the shadow buffers
were not computed, which was visible in the bge.texture.ImageRender()
custom renders.
This is now fixed: the shadow buffers will be computed at most once per
frame by the first ImageRender() or by the default render.
`bmo_create_grid_exec` was not tagging created vertices with `MARK_VERT`, which seems
mandatory to get them selected? This sounds a bit hacky/odd to me, but that's what
all other primitive funcs do...
This must have happened months ago, but as I did not `make clean` any build folder since then,
so only noted that today.
Issue is same as dirty patch we have to apply to ODL sources before building it in install_deps.sh - for
some mysterious reason, it has become impossible to compoile .osl files into .oso ones without
giving explicit output file name (otherwise it just produces `.oso` file - utterly stupid and useless).
We could probably fix that in own OSL source, but think being explicit here does not hurt anyway, so...
Let's go the easy way.
Adds default-generated UVs to mesh primitives (cone, cylinder, icosphere, uvsphere, cube, circle, grid)
when they are added to the scene, since some of them can be pretty awkward to unwrap manually.
Original patch: Liam Mitchell (CommanderCorianderSalamander).
Main review work: Campbell Barton (campbellbarton).
Finalization, fixes and cleanup: Bastien Montagne (mont29).
Reviewers: mont29, #mesh_modeling, campbellbarton
Reviewed By: mont29, campbellbarton
Subscribers: lkruel, campbellbarton, michaelknubben, kevindietrich
Maniphest Tasks: T37879
Differential Revision: https://developer.blender.org/D481
This is totally matching the way how buildbot was naming the directory.
Currently there's a bit of code duplication, but it'll be eliminated once
we'll get rid of SCons ;)
Nodes have a feature for moving existing links to unoccupied sockets when connecting
to an already used input. This is based on the standard legacy socket types (value/float,
vector, color/rgba) and works reasonably well for shader, compositor and texture nodes.
For new pynode systems, however, the hardcoded nature of that feature has major drawbacks:
* It does not take different type systems into account, leading to meaningless connections
when sockets are swapped and making the feature useless or outright debilitating.
* Advanced socket behaviors would be possible with a registerable callback, e.g. creating
extensible input lists that move existing connections down to make room for a new link.
Now any handling of new links is done via the 'insert_links' callback, which can also be
registered through the RNA API. For the legacy shader/compo/tex nodes the behavior is the
same, using a C callback.
Note on the 'use_swap' flag: this has been removed because it was meaningless anyway:
It was disabled only for the insert-node-on-link feature, which works only for
completely unconnected nodes anyway, so there would be nothing to swap in the first place.
No idea why they works on windows, but points are:
- libs/tests folder is NOT checked out on any platforms
- render_povray addon does mkdir on a path which might not have parents
There might be some other issues, but those are kinda stoppers for us.
Looks like FILE_OT_hidedot was actually the only one to remove here.
Also, added more common parentdir/previousdir/nextdir shortcuts, and R for reload.
Was lost since the original patch sent to Cambo, didn't notice the failure before
because had it applied locally.
Without this flag certain configuration wouldn't fins libgomp.
It is expected to be in the build folder for the cmake.
Ideally it should be build/<builder> or install/<builder> but that's a bit more
involved change. Will look into it later.
Yet another Space/NoSpace mismatch when searching keymaps in WM_keymap_guess_opname().
Spent again some time trying to chase those, was the only one I found...
This is so called "seems to work in dry tests" commit which is aimed to switch
linux release environment to CMake.
Some notes:
- There's no special handle of libstdc++, but it wasn't really static for quite
some time in SCons configuration and nobody really complained.
- It was quite tricky to get OpenMP linked statically with just using some
configuration so we went ahead and added a special option to CMake now which is
only exist on Linux and advertised as shouldn't be used.
- Packing is happening manually in slave_pack.py. This is because we have to add
some really special files to the archive (mesa libraries for example) which we
can't really handle from CMake/CPack in a nice generic way.
Don't think it's bad approach, at least crappynness is localized and it's not
_that_ crappy anyway.
- Windows buildbot should keep working, but needs doublechecing. It's just a
build folder changed, but you never know what it might imply.
- Some further tweaks are likely needed to ensure all builders are working.
Thanks Campbell for assistance in this patch!
This reduces stress on the the stack memory which could be really handy
on certain operation systems which applies strict limits on the stack.
Reviewers: brecht, juicyfruit, dingto
Reviewed By: brecht, juicyfruit, dingto
Differential Revision: https://developer.blender.org/D1656
- It wasn't possible to know when Object.ray_cast was successful
in the case when it hit a face with no original index.
- Take (ray_start, ray_direction) vectors.
- Take an optional distance argument.
- Scene/Object.ray_cast now use matching arguments & return values.
See D1650 (own patch)
This was getting very hard to follow,
- mixing input/output args.
- mixing arg order between functions.
- arg names (mode, snap_mode) rename to (snap_to, snap_select)
Normally we don't allow adding new theme options if we can avoid it, but this is a legit exception since all other strips are themeable.
Default color for text strip is now yellow-ish. Not nice but there are also not many other colors left.
Caminandes team request. In current master it's not possible to do this in a total
clean way, so we are simply setting all user counts of given lib's datablocks to zero
(similar to rna_ID_user_clear()'s doing).
This is a bit crappy because it still lets datablocks floating around (with invalid
user count values), and requires a save & reload cycle to be finalized. But for now
it should be good enough. Propper implementation will be added to id-remap.
Was the case of several Mesh operators actually (and probably others, but cannot check
everything). Added `RNA_def_property_float_distance` helper, avoids having to
set PROP_DISTANCE subtype explicitly each time...
Looping twice on same ID pointer may not be an issue in master currently, but with work done in id-remap
branch this should be avoided as much as possible, so for now assuming we do not need this here.
Note: if we really need this and have to add it back, then please at least use IDWALK_USER, and not
IDWALK_NOP flag!
Previously using RAS_OFS_RENDER_TEXTURE was not producing usable
textures (although the render was ok); this was due to missing
mipmap levels in the texture.
This is now fixed: the mipmap levels are explicitly created at the end of
the render as it appears that FBO do not generate them automatically.
Place non-editor related keymaps together, place animation editor keymaps together, move 'Header' keymap (which had exactly 1 entry) into 'Screen' keymap, move SCREEN_OT_header (for showing/hiding header) into 'Header' keymap.
Again, this shouldn't break any key configs, but key configs that contain SCREEN_OT_header will add an entry for this under 'Screen' again. Just a very minor glitch.
Apparently this is the result of some sloppiness during 2.5 project and since then it confused people who were trying to understand the area-region relation (myself included).
Sorry if this causes merge conflicts for anyone, but at some point we really had to do it :/
C++ structures might be rather big and passing them by value is not really
optimal. There should be no functional changes apart from better memory
access pattern when synchronizing huge scenes to Cycles.
There were multiple issues which are solved now:
- It was possible that ray wouldn't be bounced off the BSSRDF, for example
when PDF or shader eval is zero. In this case PathState might have been
left in pre-bounced state which would have been gave incorrect shading
results.
This is solved by having separate PathState for each of the hits.
- Path radiance summing wasn't happening correct as well, indirect rays
were using wrong path radiance in the case when there were more than
one hit recorded.
This is now using a bit trickier state machine which calculates path
radiance for just SSS (both direct and indirect) and then sums it back
to the final radiance.
- Previous commit wasn't totally correct either and was an induced bug
due to wrong path state left from the "un-happened" ray bounce.
There should be no special case happening here, BSSRDFs will be replaced
with diffuse ones due to PATH_RAY_DIFFUSE_ANCESTOR flag.
- Merged back codebases for "delayed" and "immediate" indirect SSS ray
tracing, hopefully making it easier to maintain the codebase.
Sure this changes brings memory usage back by about 4-5%, but overall
it's still about 2x memory reduction for the experimental kernel here.
Thanks Brecht for the review!
This is actually how it was intended to work, just didn't notice it wasn't
really happening in the main ray loop.
Solves some memory issues reported in T46880.
Restore fixed function lighting code for now and control use of GLSL shader
with a variable, make light types more clear, reduce state changes, some other
minor tweaks.
Issue was with datablocks which names would include '/', new filebrowser filelisting code
would cleanup the entire filepath, hence giving invalid filename in this case.
That 'path separator in ID names' bit us already in lib/datatype/datablock separating func,
this is really stupid to allow that in something handled as a filepath imho, but well...
Note: would have break the same under *nix with '\' char.
There are some issues to be solved with the recent optimization we did for
the indirect rays for the SSS. Those issues will take a bit of a time to
be fully solved still and we need to unlock Caminandes team now, so let's
revert some changes back.
CUDA will still use delayed indirect rays since it's an experimental
feature.
For the details about what's to be done still please refer to T46880.
This wasn't really a complete fix and only worked if there was a single scatter
event recorded only. Proper fix requires some more thoughts to make it correct
without memory use increase.
This reverts commit bf9e88bfbe.
The issue was caused by a conflict between std::error_code linked into
Blender which is compiled without C++11 support and POCL's expectations
of std::error_code which was actually linked against C++11's STL.
Radiance sum and reset was happening in different order after 26f1c51.
This is a quick fix to unlock Caminandes team, perhaps we can avoid having
separate variable to detect when radiance is to be sum.
Keep index using the outer scope for GHASH iter macros,
while its often nice, in some cases to declare in the for loop,
it means you cant use as a counter after the loop exits, and in some cases signed/unsigned may matter.
API changes should really be split off in their own commits too.
GPU_buffer no longer has a fallback to client vertex arrays, so remove
comments about it.
Changed a few internal structs/function interfaces to use bool where
appropriate.
Use for-loop scope and flexible declaration placement. PBVH does the
same thing but needs ~150 fewer lines to do it!
The change to BLI_ghashIterator_init is admittedly hackish but makes
GHASH_ITER_INDEX nicer to use.
In gpu lib:
- GPU_glsl_support() always returns true
- internal cleanup & comments
Outside gpu lib:
- remove check from various code, remove the “else” path
- sprinkled a few C99-isms
We can remove GPU_glsl_support() when BGE stops calling it.
It uses edge extrapolation code from Gimp's antialias plugin now,
which has advantage of giving symmetrical results, which makes it
possible to add two antialiased ID masks and have a constant 1.0
all over the frame. But has difference from the old implementation
because it uses 3x3 matrix only, which doesn't give so much smooth
looking edges. Perhaps it's not so bad, since if edges are really
need to be smooth one might use Blur node.
Another advantage is that the node is now nicely threaded.
Reviewers: campbellbarton
Reviewed By: campbellbarton
Subscribers: ania
Differential Revision: https://developer.blender.org/D1617
Previously render nodes will be always created with just a VECTOR socket
type and then those sockets will try to be set as all point, vector and
normal to work around lack of such a subtype distinguishing in blender.
This change makes it so subtype is being queried from OSL itself and
proper subtupe is being used for socket.
It's still not in use for the official builds because it requires changes
applied recently on the 1.7 branch of OSL:
https://github.com/imageworks/OpenShadingLanguage/commit/f70e58f
This solves artists confusion reported in T46117.
Reviewers: #cycles, juicyfruit
Reviewed By: #cycles, juicyfruit
Subscribers: juicyfruit
Differential Revision: https://developer.blender.org/D1627
Graph::disconnect() actually modifies links, needs to create a copy to iterate
if disconnect happens form inside the loop.
Question tho whether we can control this somehow..
Reported by BzztPloink in IRC, thanks!
This makes it possible to use scenes as a kind of
multi-user meta-strip (with their own time).
Currently this supports rendering & drawing nested strips,
but no convenient way to tab-enter into a scene strip.
When a user tries to load a non-existing blend file from the CLI, the path
is set and Blender acts as if it is loaded. This allows the user to create
a new file by typing 'blender filename.blend' and then saving. This
behaviour is common in other tooling, such as vim and Mypaint.
Blender's current behaviour (print an error message and open the default
scene) doesn't make much sense. It ignores the filename passed on the CLI,
whereas with this patch that filename is actually remembered, and used when
saving.
Reviewers: campbellbarton, sergey, mont29
The issue was caused by possible use of object->derivedFinal from the render
thread, The patch tries to eliminate (or at least minimize, huh) amount of
access to the derivedFinal of a source object. It's still possible that in
the case of particle source derived mesh will be still unsafely used, but
with the patch applied we can easily change runtime part of the code and
cache derived mesh on the preparation stage.
Some ideas for the future:
- Check whether cache() was called on the point density node when calling
calc().
- Cache derivedMesh in the runtime part of point density node to avoid
possible remained thread conflicts.
- NULL the runtime part of the node on .blend load
Reviewers: campbellbarton, plasmasolutions
Reviewed By: plasmasolutions
Differential Revision: https://developer.blender.org/D1614
Seems set_intersection() requires passing explicit comparator if non-default
one is used for the sets. A bit weird, but can't really find another explanation
here about whats' going on here.
This mimics OpenMP's 'firstprivate' feature. It is sometimes handy to have some persistent local data during a whole chunk.
Reviewers: sergey
Reviewed By: sergey
Subscribers: campbellbarton
Differential Revision: https://developer.blender.org/D1635
The issue was caused by not really optimal graph traversal for gathering nodes
dependencies which could have exponential complexity with a long tree branches
connected with multiple connections between them.
Now we optimize the depth traversal and perform early output if the node was
already traversed.
Please note that this adds some limitations to the use of SVM compiler's
find_dependencies() in the cases when skip_node is not NULL and one wants to
perform dependencies find sequentially with the same set. This doesn't happen
in the code, but one should be aware of this.
The issue was than nodes dependencies were stored as set<ShaderNode*> which
is actually a so called "strict weak ordered", meaning order of nodes in
the set is strictly defined, but based on the ShaderNode pointer. This means
that between different render invokations order of original nodes could be
different due to different pointers allocated for ShaderNode.
This commit makes it so dependencies and maps used for ShaderNodes are based
on the node->id which has much more predictable order. It's still possible
to trick the system by doing some crazy edits during viewport rendfer and
cause difference between viewport and final render stacks.
Reviewers: brecht
Reviewed By: brecht
Subscribers: LazyDodo
Differential Revision: https://developer.blender.org/D1630
This gives much lower stack usage on GPU and reduces kernel memory size to
around 448MB on GTX560Ti (comparing to 652MB with previous commit and 946MB
with official release). There's also a barely measurable speedup of around
5%, but this is to be confirmed still.
At this stage we're using only ~3% for the experimental kernel and SSS
rendering seems to be faster by 40% and after some further testing we might
consider making SSS and CMJ official features and remove experimental
precompiled kernels.
The idea is to delay shooting indirect rays for the SSS sampling and
trace them after the main integration loop was finished.
This reduces GPU stack usage even further and brings it down to around
652MB (comparing to 722MB before the change and 946MB with previous
stable release).
This also solves the speed regression happened in the previous commit
and now simple SSS scene (SSS suzanne on the floor) renders in 0:50
(comparing to 1:16 with previous commit and 1:03 with official release).
This commit introduces a SSS-oriented intersection structure which is replacing
old logic of having separate arrays for just intersections and shader data and
encapsulates all the data needed for SSS evaluation.
This giver a huge stack memory saving on GPU. In own experiments it gave 25%
memory usage reduction on GTX560Ti (722MB vs. 946MB).
Unfortunately, this gave some performance loss of 20% which only happens on GPU.
This is perhaps due to different memory access pattern. Will be solved in the
future, hopefully.
Famous saying: won in memory - lost in time (which is also valid in other way
around).
Input array length is implicitly set at link time, based on the geometry
shader's layout. Specifying the wrong value here is an error; specifying
no value is the same as getting it right. (inspired by a recent codegen
change)
Fixes this bug:
1. add `logic.setRender(False)`
2. run game, exit game
3. remove the `logic.setRender` line
4. run game ... the render is still off
Now the render flag will be on by default on game start.
Note: during the game, the render flag stays unchanged across
game restart/load. It's only on initial game start that it is
initialized on.
the paint cursor drawing when a stroke is underway; this information is
already available from the stroke itself.
Should improve performance when "show brush" is enabled (which is
always, for sane people). Finding an intersection is not always cheap,
especially on heavy meshes.
Cudos to the dyntopo test thread for the report and the sculpt love :)
yet another bug introduced by recent shadowing changes -- q and r CCG arrays
were overwritten by the temporary evaluation because the code was changed to
use global pointers instead of the local ones.
Naming of the variables could be changed to be a bit more accurate.
This is sort of extension of existing Use Environment option which now allows to
disable AO on the render layer basis.
Useful in cases like disabling AO for the background because it might make it
too flat and so.
Reviewers: juicyfruit, dingto, brecht
Reviewed By: brecht
Subscribers: eyecandy, venomgfx
Differential Revision: https://developer.blender.org/D1633
This patch allows the game engine to keep running while performing things like PNG compression and disk I/O.
As an example, my crowd simulation rasterizer saves a screenshot for every frame. This now takes up 13 msec per frame, which was 31 msec before this patch. Effectively, it allows the simulation to save every frame and still run at 60 FPS.
Reviewers: lordloki, moguri, panzergame
Reviewed By: moguri, panzergame
Projects: #game_engine
Differential Revision: https://developer.blender.org/D1507
This commit adds '--build-foo' options to force the script to build relevant libraries
instead of trying to use packages from the distribution.
In addition, it also now offers (with those '--build-foo' options) the possibility
to build libraries on distributions that are not fully supported.
This is limited, but should still help people once they have installed themselves
the basics of dependencies - boost, llvm, osl/osd etc. are not libraries that are
really easy to build.
DISCLAIMER: I did not take the 20 (or more) hours needed to test all combinations
over all distributions, and given the size of the changes, bad sneaky typos are quite
probable, so please report if you get some errors!
GLSL 130, 140, 150 with extensions as needed.
Similar logic to my recent gpu_extensions changes.
Partially fixes T46706. Matcaps now work with OpenSubdiv, as do basic
materials. Anything with UV coordinates is still broken.
Formatting of generated GLSL code:
- attribute/varying for version 120
- in/out for version 130+
- minor cosmetic stuff
Tested working on Windows 10, GL 4.3.
Fix GLSL version & geometry shader support query to consider core vs
compatibility.
All shaders need to be compatible with each other, and for now that
means GLSL 120. For drivers that support compatibility profiles, choose
the highest available (up to 150). If only core profile is supported,
max out at GLSL 130.
Several changes. Tested working on Windows 10 GL 4.3 and MacOS 10.11 GL
2.1.
- document extensions used in this file
- some simple ARB/EXT suffix deletion
- stop checking for pre-2.1 features — they’re available!
- convert old ARB shader API to the one adopted in GL 2.0
- remove checks for old (pre-R600) ATI cards
- choose GLSL version at runtime, between 1.2 and 1.5
- prefer GLSL 1.5 for geometry shaders, fall back to
EXT_geometry_shader4 if needed
Differential Revision: https://developer.blender.org/D1632
Performance is roughly the same because it's using the same COLAMD ordering
and supernodal LU factorization algorithms. Solve results also appear to be
identical.
The main new feature is mixed variable declarations and code, which can help
reduce uninitialized variables or accidental variable reuse.
Due to incomplete C99 support in VS 2013, variable length arrays are not
supported, BLI_array_alloca must still be used. The header <tgmath.h> is also
not supported.
Differential Revision: https://developer.blender.org/D1631
GPUBuffer rendering is now done using vertex buffers.
Vertex arrays are completely removed from GL 3.2 core profile, so we'll
have to do this change at some point anyway.
This commit, though big, is not modifying blender in any way. Use should
be exactly as if the vetex buffer option is constantly on.
This patch contains the following changes:
- the vertices and edges arrays would be assigned default values, and
then reassigned new ones right away. It appears that those arrays were
once global and then made local
(rB06a2ee4afed4237398b69ddf253e29a730b2f9f0), so it makes sense now to
initialize them with the right values.
- the flame spectrum texture was created whether it was needed or not,
so now it's only created if there's flame to be drawn, also split the
code in a separate function.
- reduce the number of parameters to the main draw function, as most of
them are member of SmokeDomainSettings.
- some other minor cleanups: fold multiple operations into one to get
rid of one local variable, mark variables as `const` when necessary,
unecessary gl draw calls, reorder the code a bit...
Reviewers: campbellbarton, psy-fi
Differential Revision: https://developer.blender.org/D1368
Unfortunately there's no easy way to show a messagebox here, so just
print a warning on fstderr and exit. If we don't call exit() here we get
crashes on other blender systems (python, opensubdiv) and it can get
tricky to track the initialization state here, so just using exit()
should do the trick for now.
simple stuff!
- remove ARB suffix from core functions & enums
- remove checks for core features (VBOs, generic attribs)
- keep checks for non-core features (draw elements base vertex)
This gives few percent extra memory saving for the CUDA kernel when
using regular path tracing.
Still more like an experiment, but will be handy in the future.
Summary: By calculating the Camera-to-Screen-Matrix first, one inversion can be saved in the Camera sync.
It won't really improve speed and/or precision, it's mainly a small cleanup.
Reviewers: sergey, dingto
Subscribers:
fbo = bge.render.offScreenCreate(w,h,samples,target)
target = bge.render.RAS_OFS_RENDER_BUFFER: send render to RenderBuffers (compatible with MSAA on Intel GPU)
target = bge.render.RAS_OFS_RENDER_TEXTURE: send render to Texture (not compatible with MSAA on Intel GPU)
The default (RAS_OFS_RENDER_BUFFER) is preferable in all cases expect
if you need to access directly the texture for some applications.
The texture bind code is accessible with the new attribute color (fbo.color).
The attribute value is 0 if target is RAS_OFS_RENDER_BUFFER.
Gives few percent of memory improvement for regular feature set kernel
and could give significant memory improvement for Experimental kernel.
It could also give some degree of performance improvement, but this I
didn't really measure reliably yet.
Code is ifdef-ed for now, since it's only working on Linux and requires
CUDA toolkit to be installed (other platform only use precompiled
kernels).
This is just an experiment for now and a base for the proper feature
support in the future (with runtime compilation using CUDA 7?).
This just replaces internal argument `experimental` with `requested_features`
making it possible to access particular requested settings when building
kernels.
Technically this was more of a feature request, but now the Alt-I operator will
only remove keyframes related to selected bones in Pose Mode. In Object Mode,
it will continue to operate on all keyframes of the object.
This change makes this operator more meaningful when animating in the 3D view.
Face paths including triangle-fans would often be avoided because
the face center and edge would 'zig-zag'.
Resolve using wighted face center and picking an edge-point between face centers.
Basically we can not use sharp closure as a substitude when filter glossy is
used. This is because we can not blur sharp reflection/refraction.
This is quite quick and not really clean implementation. Not really happy
with manual handling of original settings, but this is as good as we can do
in the quick patch. It's a good acknowledgment and we now can re-consider
some aspects of graph simplification to make such cases more natively
supported.
P.S. This failure would have been shown by our regression tests, so please,
bother a bit to run Cycles's test sweep before doing such optimizations.
Issues was again the ugly hack of swapping last generated tri with original face
we use in BMesh triangulate code - here it could lead in some rare case to have
invalid face pointer in doubles list.
In case two neighbor buttons are very small, their total width (or height) can remain below
max authorized delta, and hence wrong side could be detected as 'common side' for the pair.
This is now fixed by checking both opposite sides at once.
Also, we expect buttons to have some width and height to be considered alignable now!
Took me two days to sort that out, grrrr!
Existing collapse functions were strict regarding the number of verts sharing an edge.
BM_edge_collapse allows any edge to be passed in without first having to manipulate geometry.
This commit adds the Blackman-Harris windows function as a pixel filter to Cycles. On some cases, such as wireframes or high-frequency textures,
Blackman-Harris can give subtle but noticable improvements over the Gaussian window.
Also, the gaussian window was truncated too early, which degraded quality a bit, therefore the evaluation region is now three times as wide.
To avoid artifacts caused by the wider curve, the filter table size is increased to 1024.
Reviewers: #cycles
Differential Revision: https://developer.blender.org/D1453
We fallback to Sharp closures for Glossy, Glass and Refraction nodes now, in case the Roughness input is disconnected and 0 (< 1e-4f to be exact).
This way we gain a few percentages of performance, in case the user did not manually set the closure type to "Sharp" in the UI.
Sharp will probably be removed from the UI as a followup, not needed anymore with this internal optimization.
Original idea by Lukas Stockner(Differential Revision: https://developer.blender.org/D1439), code implementation by myself.
This performs a range search on the kdtree, running a callback instead of allocating an array.
Allows the caller to perform extra checks in the case of overlap,
avoids redundant array allocations, since caller can handle matches.
Another fix for a slightly different case from the original report, to give better orientation.
Vertex normals influence on the orientation is now scaled by the alignment with the cut direction.
Also use face normal instead of loop normal which can be unreliable when the corners edges are co-linear.
Aside from some minor cleanup, this commit:
* Fixes checking twice for multiple usage of same vert by a same poly.
* Fixes handling of ME_VERT_TMP_TAG vert flag by that check (there was no guaranty
that flag was cleared for a poly's vertices before we start checking).
The idea is to avoid having roundtrip from byte to float and back to byte buffer
and use render result's byte buffer to store result of sequencer rendering.
This actually matches to what regular render pipeline is doing and this gives
around 2-3 times speedup of sequencer export on a simple scenes.
This is something what IMB_colormanagement_imbuf_for_write() is intended to do.
Gives around 40% speedup in test wile with OpenGL render and about 20% when
rendering simple cube scene animation.
Fixed extension check.
This feature requires ARB_texture_buffer_object which was subsumed into
OpenGL 3.0. Intel driver on Windows doesn't claim to support this
extension, but GL version is > 3 so it actually does work.
This fixes two issues:
* Normals were not being recalculated correctly when not using optimized
drawing for CDDerivedMesh (Multires actually handles that correctly).
* Loop normals (autosmooth option) were also not being calculated. Doing
this calculation is not desirable, since it can't be done correctly
without a severe performance hit. This is easy to test by doing a
dependency flush on the mesh after each scuplt stroke step. Instead they
are now disabled during sculpting.
We have currently a gooseberry file (scenes/01_island/01_meet_franck/01_01_01_A/01_01_01_A.anim.blend)
that links against two -pre repo libs, which are hence not available for common mortals,
and generate warnings and placeholders during load step.
Issue is, among those missing (directly) linked datablocks, we have two shapekeys!
This should never happen nor be possible at all. I tried understanding how this could happen,
with no luck at all, best bet would be some wild/bad call to `id_us_plus()` over those skeys
at some point...
Anyway, this commit:
- Handles a bit better those 'cases that should never happen' at load time.
- Adds several checks in ID handling code (and save/load code) to try to detect where/when
a non-linkable datablock becomes LIB_EXTERN (i.e. directly linked).
Better support of bicubic sampling of ID mask output.
The idea is to generate ID mask into a temporary buffer which is then being
interpolated using current sampling method.
This works fine for upscaling or rotating the ID mask but does not work for
scaling down. This is much-much bigger problem of the compositor design and
can't really be solved currently. Same will happen with other nodes like
blur for example.
Reviewers: campbellbarton
Subscribers: ania
Differential Revision: https://developer.blender.org/D1612
The issue is caused by transform tool temporary affecting on the pose
constraints, which actually changes the way how pose is to be evaluated.
This isn't ideal patch from the performance point of view, but only
limited to the new depsgraph, so we've got some time to work on partial
graph updates.
The issue was in fact a feedback loop between render database init which needs
to know which render layers to initialize and scene update for newframe which
needs to know which layers to update.
Worked around by updating scene animation before render database initialization
which isn't really ideal, but doesn't really have much better alternative.
The issue was caused by the image save operator ignoring Z-Buf if the input
buffer due to the way how IMB_dupImBuf() worked. There's no reason to to
ignore z-buf on imbuf duplication, it only asks for issues in the future.
Note that in master this is mere cleanup, which shall not change much things
(aside from FAKE_USER now being taken into account here too), and mostly
makes things more consistent, but in id-remap branch this becomes mandatory
due to better handling of this ugly 'ensure_real' stuff re user count.
The motion actuator goes out of its way to prevent setting zero velocities,
which should actually be supported. This patch just works around it as a
first test. We should investigate whether the flags
`m_bitLocalFlag.ZeroLinearVelocity` and `m_bitLocalFlag.ZeroAngularVelocity`
are actually needed/desired at all.
One of the issues that's already visible with this simple change, is
that objects aren't actually frozen but still move a little bit; see
test with {F241908}.
Reviewers: lordloki, hg1, moguri, panzergame
Reviewed By: lordloki, panzergame
Differential Revision: https://developer.blender.org/D1545
Looks like UI code can produce widgets with same left/top positions (in very narrow space case).
Not nice, but we do not really care, UI becomes unusable way before we reach that point anyway.
There was a bug in OpenSubdiv library which is now fixed by updating library in
the build environments.
Still requires testing, but now being at the beginning of release cycle is
should be safe to simply enable option and let everyone to test :)
RNA could define strings as dynamically sized,
but the interface ignored this and clamped them to UI_MAX_DRAW_STR.
Fixes T46734, also fixes possible pasting non-utf8 text into utf8 buttons.
We have callbacks for that, they also do some checks and help ensure things are done
correctly. Only place where this is assumed not true is blenloader (since here we
may affect refcount of library IDs as well...).
USER_ONE was only ensuring id->us was non-zero, while USER_REAL ensures
it is non-zero **and** >1 in case fake_user flag is set (which at least
ensures us unsetting fake_user won't leave id->us in invalid zero state).
CMake 3.0 ignores the suffixed COMPILE_DEFINITIONS_<CONFIG> variants if policy
CMP0043 is NEW, preferring instead to consume the unsuffixed COMPILE_DEFINITONS
property and evaluate generator expressions in it.
Setting the policy to OLD causes CMake 3.0 to not ignore those properties.
Use the suffixed properties for compatibility only for CMake versions older
than 3.0 using an explicit version check. Use the non-suffixed property with
generator expressions otherwise.
These definitions should mostly not be needed at all - CMake already sets
NDEBUG in suffixed CMAKE_CXX_FLAGS_<CONFIG> variables. The _DEBUG macro is
non-standard, but is used in several places in blender (usually in an OR
combination with the DEBUG macro). Additionally, blender overwrites (instead
of appending to) the CMAKE_CXX_FLAGS_<CONFIG> variables on WIN32.
Rather than try to fundamentally change how debug-macros are handled in
blender, change the buildsystem to keep the same behavior without requiring the
CMP0043 policy set to OLD.
As we can see in (original) read code, ob pointer in groupobject is actually a 'USER_ONE' case.
This was not done in 'add object to group' code, probably because we can assume objects always
have at least one user in that case? Made it explicit now. Also fixed foreach_ID_link looper.
In general we have waaayyyyyy too much 'own handling' of ID->us count in code currently,
will clean up that...
(if they're on different properties).
This patch basically gets rid of (ui-related) hack where we was adding specific
'fake' enums for datatypes needing additional options.
That was done because of broken UI code - to summarize, 'align' code did not support
correctly layout mixing vertical and horizontal sub-layouts, in particular if
it was 'column-major'.
A complete rewrite of this align code has been done, so now we can use a more
sane UI code.
This new code fixes a tons of issues with previous one, which basically was epic-failing
in many non-basic cases (especially mixed columns and rows with column-dominant layout).
It basically no more relies over order of buttons declaration in the uiBlock, instead it
finds and stores spatial neighbors and uses that data to compute needed stitching.
See code comments for details.
New code seems to be roughly ten times slower than old one (for complex grouped layouts),
that is, about a few microsecconds per alignment group - this remains reasonable.
Also, ui-align code becomming rather big in itself, it was separated in
own new `interface_align.c` file.
Reviewers: campbellbarton, severin
Reviewed By: campbellbarton, severin
Differential Revision: https://developer.blender.org/D1573
This commit makes drivers on node groups to with when using new dependency graph.
Still TODO: Need a relation between drivers and tree evaluation perhaps, so we
guarantee proper order of all operations.
Darkens the colors for limit and mist indicators of non-active cameras. This makes it easier to see which indicators belong to the active camera and which don't.
Useful for layouts with multiple cameras.
Requested by the Caminandes team.
RASOffScreen was previously using GPUOffScreen object from the gpu library,
which in turn uses textures as render destinations, also when multisample is
requested. This is a problem for Intel integrated as they generally don't
support multisample textures.
ImageRender doesn't need textures as render destination: Renderbuffers are
sufficient because it only read pixels from the render with glReadPixels and
glCopyTexSubImage, both of which work with Renderbuffers.
Therefore, RASOffScreen is reimplemented with Renderbuffers and multisample
offscreen renders are now supported on Intel GMA. The gpu library is no
longer used. There is no API change.
Custom-data on newly created face data was often rotated.
Now the API doesn't copy data from adjacent loops when creating faces.
Most functions were already overwriting this anyway.
Since such decisions are better made at a higher level, now it's the responsibility of the caller.
This is really similar to what GPU module was already doing. There are number of
possible improvements still:
- Re-use allocated VAOs when requesting new ones instead of going to the trouble
of freeing VAO and then re-creating it again.
- Move VAO handling to GPU module.
Fixes T46589: OpenSubdiv crash with drivers
There is no real elegant solution here, ideally a modifier shall never *modify*
a source DM, but that would imply much better ways to ensure required data
is available in that source DM, which we do not have currently.
So instead, let's use brute force solution for now and always create a local copy
of our source DM, that we can modify to our heart content!
The bug header is wrong, the file contains the high pitched sound, but the bug that existed was that animation rendering did not use the high quality resampler, while audio mixdown does.
Blender uses the low quality resampler to be as little CPU consuming as possible.
Otherwise user may have a hard time understanding why their modifier is no more working
after they e.g. deleted some vertices... including the root one. ;)
KX_IPOActuator is not used since 2.4.
This patch removes:
- KX_IPOActuator.h/cpp;
- Python doc;
- Write of IPO actuator in write.c;
- Allocation of IPOactuator in sca.c;
- Conversion in KX_ConvertActuators.cpp;
- Initialization of the python proxy in KX_PythonInitTypes.cpp;
- Other minor remove in logic_windows.c and in KX_PythonInit.cpp.
Reviewers: sybren, campbellbarton
Reviewed By: campbellbarton
Subscribers: moguri
Differential Revision: https://developer.blender.org/D1603
Crash was happening due to OpenGL free from the non-main thread. Ideally we need
to implement somewhat delayed unused buffer free for OpenSubdiv similar to what we
are doing in GPU module.
We needed the 'background' feature here, since we never wait on those preview-generation tasks.
Note that it also simplifies the code, and as usual testing is needed here on all possible platforms...
With current code, in single-threaded context, a pool of task may never be executed
until one calls BLI_task_pool_work_and_wait() on it, this is not acceptable for
asynchronous tasks where you never want to actually lock the main thread.
This commits adds an extra thread in single-threaded case, and a new 'type' of pool,
such that one can create real background pools of tasks. See code for details.
Review: D1565
In previous code, worker would exit in case it gets awoken from a condition_wait() and
task queue is empty. However, there may be spurious wake up (either due to pthread itself,
or to some race condition between workers) that would lead to wrongly exiting a worker before
we actually exit the whole scheduler. See code for more details.
3rd optional argument to offScreenCreate() to specify the number of
multi-sample. If the GPU does not support MSAA on FBO, this is ignored.
No other change in the API.
Metas are scanning all scenes duplis,
which can go into particle systems without an initialized derived-mesh.
For now just do NULL check, its not correct but real fix is not fitting well with current design.
Brush.toggle_brush was allowed to be an invalid pointer,
it worked for the one operator that used it - but in general bad practice,
requiring a lookup on every access.
Ensure the pointer is kept valid now.
Fix for T46494 wasn't working properly when de-selecting faces,
adjacent faces would remain selected but have unselected edges.
Logic here is admittedly rather fragile since it relies on both
selection functions and flushing afterwards.
Certain shapes could trick the inside/outside test.
An edge between 2 planar faces could be selected for detecting face-flipping (which failed).
While this could be prevented by skipping those edges,
use a method which searches for the outer most face-loop, then check it faces the center.
Works totally similar to backdrop in the compositor.
Requested by Sean Kennedy, but could be useful for lots for VFX guys.
Reviewers: campbellbarton
Reviewed By: campbellbarton
Subscribers: sebastian_k, hype
Differential Revision: https://developer.blender.org/D1590
Grabbing now doesn't 'jump' when shift is released (matching rotation modes).
This simplifies most logic for transform input,
where mouse input callbacks can choose to use the 'virtual' cursor,
which accounts for precision when shift is held.
Each time we setted the mass the own object gravity was divided by its
old mass (i.e you could convert you car in a flying DeLorean with a
simple mass car modification).
A note will be included in release notes due to retro compability
issues.
The start position must be initialized at the first call of KX_IpoSGController::Update when m_ipo_start_initialized is to false, not when also the frame time is not 0.
- Add spaces around operators.
- Replace !(a == b) by a != b
- Add "f" at end of float value.
- Remove extras lines in fonctions declaration.
- Remove indentation to align attributes definition.
To make consistent with KX_GameObject.replaceMesh, we don't allow this behavior but print an error message for the replace mesh actuator.
e.g : Warning: object "Cube" from ReplaceMesh actuator "Edit Object" uses a mesh not owned by an object in scene "scene1"
Reviewers: youle.
Previously it was possible that same component will be tagged for update
again and again, making update flushing really slow. Now we'll store flag
whether component was fully tagged.
This is still temporary solution because ideally we should just support
partial updates, but that's for the future.
Gives around 10% speedup on file from jpbouza.
This behavior caused a double free.
Before when we duplicated an instance of a group the new instance keep the pointer of the group but was not added in the group instance list (normal).
And during the object deletion we tried to remove the object in the instance list but anyways if it failed decrement the reference count.
Set the group and the instance list to NULL in ProcessReplica avoid these kind of problems.
'Distance from Object' color/alpha/thickness modifiers without a target
object were raising a run-time exception although it is not considered an
error condition.
The reported crash was confirmed as a segmentation fault in std::sort().
The cause of the crash was traced down to a binary comparison function
that was not satisfying the so-called strict weak ordering requirements of
the C++ standard sorting function. Specifically, the comparison operator
has to return false when two objects are equivalent (i.e., comp(a, a) must
be false), but that requirement was not met.
Since the binary comparison operator in question could be a user-defined
Python function, here a safety measure is implemented in the C++ layer to
make sure the aforementioned requirement is always satisfied.
Bug introduced in 583fa7d1e, KX_GameObject.setActionFrame can make BL_Action::m_starttime negative. But in BL_Action::Update m_starttime is set to the current time if it's negative.
To fix it we use a boolean BL_Action::m_initializedTime to know if we should initialize the time in BL_Action::Update, it's more stable than comparing times.
Tested with bug task T45945 and T32054, with an extra patch about to fix suspend resume scene issues with actions : D1569
Currently disable all of them, in practice i think way to go should be:
- Disable Experimental kernels on 32 bit, build up to sm_35
- Later we can drop all 32bit kernels, but try to keep at least one release
with some of the kernels (they'll cover 99% of users anyway)
Before doing any changes we should surely communicate such a changes before
we apply them.
Shutter curve now can be controlled using curve mapping widget in the motion
blur panel in Render buttons. Only mapping from 0..1 by x axis are allowed,
Y values will be normalized to fill in 0..1 space as well automatically.
Y values of 0 means fully closed shutter, Y values of 1 means fully opened
shutter.
Default mapping is set to old behavior when shutter opens and closes instantly.
This shutter mapping curve could easily be used by any other render engine by
accessing scene.render.motion_blur_shutter_curve.
Reviewers: #cycles, brecht, juicyfruit, campbellbarton
Differential Revision: https://developer.blender.org/D1585
Previously shutter was instantly opening, staying opened for the shutter time
period of time and then instantly closing. This isn't quite how real cameras
are working, where shutter is opening with some curve. Now it is possible to
define user curve for how much shutter is opened across the sampling period
of time.
This could be used for example to make motion blur trails softer.
Previously curve mapping was always setting to only a single slope which then
was symmetrizied by a tools (such as brush or compositing).
With this change it's possible to set curve to symmetrical slopes as a part
of preset.
This appears to be really common workflow when you stabilize shot to make compo
easier (roto, some effects and so) and then re-introduce the motion back.
Surely it's doable with some magic nodes and manual network for transforming
but such workflow is too common in VFX to resist adding one small option in
single node for this.
Opening a tmp window on a 4K display with virtual pixelsize set to double results in a too small window. For Retina this seems to be handled on GHOST level already, so multiply by virtual pixelsize only.
This patch makes independant the override camera zoom and the default camera object zoom. The override zoom is 2.0f and the camera object zoom is computed with the RegionView3D camera zoom.
These features are only used for the embedded BGE, so there's no compatibility issues with the blenderplayer.
I let the override zoom with a setter to allow later to create an override camera in blenderplayer easily.
Tested with the 3 framing modes in embedded and standalone.
I made this patch to declared a python list without converting all elements in python object (too slow) or use a CListValue which required CValue items (too expensive in memory). In the case of a big list of points like a collision contacts points list, to use a CListValue we must implement a new class based on CValue for 3D vector to create a python proxy even if mathutils do it perfectly, we must also convert all points (frequently ~100 points) when fill the CListValue even if the list is not used (in the case of the collision callback). The easy way is to use callback (it doesn't worth to do an inheritance) which convert the item in PyObject only during an acces.
5 callbacks are used :
- Check if the list is valid = allow acces (like PyObjectPlus.invalid)
- Get the list size
- Get an item in the list by index.
- Get an item name in the list by index (used for operator `list["name"]`)
- Set an item in the list at the index position.
All of these callback take as first argument the client instance.
Why do we use a void * for the client instance ? : In KX_PythonInitTypes.cpp we have to initialize each python inherited class, if we use a template (the only other way) we must add this class each time we use a new type with in KX_PythonInitTypes.cpp
To check if the list can be accessed from python by the user, we check if the python proxy, which is the `m_base` member, is still a valid proxy like in PyObjectPlus. But we can use a callback for more control of user access (e.g a list of collision point invalidate a frame later, in this case no real python owner).
This python list is easily defined with :
```
CPythonCallBackList(
void *client, // The client instance
PyObject *base, // The python instance which owned this list, used to know if the list is valid (like in KX_PythonSeq)
bool (*checkValid)(void *), // A callback to check if this list is till valid (optional)
int (*getSize)(void *), // A callback to get size
PyObject *(*getItem)(void *, int), // A callback to get an item
const char *(*getItemName)(void *, int), // A callback to get an item name (optional) use for acces by string key
bool (*setItem)(void *, int, PyObject *) // A callback to set an item (optional)
)
```
To show its usecase i replaced the odd KX_PythonSeq, it modify KX_Gameobject.sensors/controllers/actuators, SCA_IController.sensors/actuators and BL_ArmatureObject.constraints/channels.
Example : {F245193}, See message in console, press R to erase the object and see invalid proxy error message.
Reviewers: brita_, #game_python, youle, campbellbarton, moguri, agoose77, sergey
Reviewed By: campbellbarton, moguri, agoose77, sergey
Subscribers: sergey
Projects: #game_engine
Differential Revision: https://developer.blender.org/D1363
Previously, a warning was added to provide feedback to users trying to change the values
of driven properties why their edits would not have any effect on the propeerty. However,
it turned out that instead of only showing up when the user tried to increment/decrement/slide
the property's value, it was also firing everytime they were trying to edit the expression.
That however is not what we want at all!
This fix assumes that BUTTON_STATE_TEXT_EDITING is used for expression editing, and
BUTTON_STATE_NUM_EDITING (or everything else) refers to the user trying to adjust the
value normally.
When using the "Current Frame" options for these operators, the Cursor X value
will now be used instead of the current frame. Perhaps the labels could be changed
too, but for now, I guess this will be good enough.
When the "Use Y" option in the Copy Rotation constraint is disabled, the constraint
behaves eratically when rotating all the target on all axes at the same time.
This is partially to be expected due to the way that euler rotations work
(i.e. the rotation orders stuff - you should use a rotation order based on most to
least important/significant rotations). Hence, by locking Y, you're causing accuracy
problems for Z.
What was not expected though was that changing the rotation orders on the objects
involved (for the record, it's the constraint owner that counts) did nothing.
It turns out that for objects, the rotation order settings were getting ignored!
This commit fixes this problem, and this particular case can be resolved by using
"XZY".
Notes:
* Since all object constraints were previously working on the assumption that they
used XYZ (default) order, it is possible that this change may have the unintended
consequence of changing the behaviour of some rigs which relied on the buggy
behaviour. Hopefully this will be a rare occurrence.
When working is the Graph Editor it can be very important to be able to work with fractions
(sub integers), especially when working with Drivers. Currently the "Cursor Y" is hooked up
to "cursor_position_y" which allows fractions but "Cursor X" is directly hooked up to
"frame_current" which is an integer.
This commit adds initial support for this feature.
* When in Drivers mode, the x-part of the cursor is mapped to a new "cursor_position_x"
value which can have fractional values. Animation mode however remains mapped to frame_current
* This commit only adds the UI/property/drawing tweaks needed to support this.
Many operators still need to be modified to consider this value instead of the
current frame, for this to be more useful.
Sometimes the timeline header didn't update after time-scrubbing in the graph
editor ends, leaving the "Pause" button visible until the next refresh of the
timeline (e.g. on mouse over)
Load driver dynamically at runtime instead of weak-linking the
3Dconnexion framework. Driver no longer needed at build time!
Works with really old drivers (as in PowerMac old), more recent
versions, and the latest which allows us to process events on a
separate thread.
This patch fixes:
- the call of LibFree on a unfinished loaded library;
- memory leak created on end of game : the async libraries are loaded but not converted, so not freed with the master scene.
Reviewers: campbellbarton, sybren, youle, hg1, moguri, lordloki
Reviewed By: moguri, lordloki
Differential Revision: https://developer.blender.org/D1571
- RegionView3D.view RV3D_VIEW_PERSPORTHO
only ever set on initialization, never checked for.
- Lamp.type LA_YF_PHOTON
from old 2.4x yafray files.
Also iniitalize movie-clip + grease-pencil theme colors.
It was possible to miss some intersection caused by wrong barycentric
coordinates sign.
Cases when one of the coordinate is zero and other are negative was not
handled correct.
The issue was caused by wrong sign check. It originally came from more optimized
Cycles code where because of other reasons it wasn't visible yet. But in fact it
should be solved there as well.
Previously it would save packed file(s),
which would ignore the image.filepath,
making it impossible to set the destination.
Add image.packed_files[...].save() so you can save packed files if its needed.
Idea is, instead of ignoring completely missing linked datablocks, to
create void placeholders for them.
That way, you can work on your file, save it, and find again your missing data once
lib becomes available again. Or you can edit missing lib's path (in Outliner),
save and reload the file, and you are done.
Also, Outliner now shows broken libraries (and placeholders) with a 'broken lib' icon.
Future plans are also to be able to relocate missing libs and reload them at runtime.
Code notes:
- Placeholder ID is just a regular datablock of same type as expected linked one,
with 'default' data, and a LIB_MISSING bitflag set.
- To allow creation of such datablocks, creation of datablocks in BKE was split in two step:
+ Allocation of memory itself.
+ Setting of all internal data to default values.
See also the design task (T43351).
Reviewed by @campbellbarton, thanks a bunch!
Differential Revision: https://developer.blender.org/D1394
This expose the capability of handling offscreen drawing. The initial
support lays the barebones for addons to work with framebuffer objects
and implement 3d viewport offscreen drawing. This can be used by script
writers to make fisheye lens preview, head mounted display support, ...
The complete documentation is here: http://www.blender.org/api/blender_python_api_2_76_1/gpu.offscreen.html
Review and many changes by Campbell Barton (thank you :)
https://developer.blender.org/D1533
All image source objects now accept a pixel format argument in addition to
the buffer argument in the refresh() method. This can be used to change the
format of the image copied in the buffer. The only allowed values are "RGBA"
and "BGRA". This is useful when transferring images to external application
that do not support the RGBA format.
Ex:
im = bge.texture.ImageRender(...)
im.refresh(buffer, "BGRA")
Note: the transfer is optimal (i.e. no intermediate copy) for ImageRender and
ImageViewport objects when no processing of the image is needed in VideoTexture.
This is the case if no filter is set and the attributes are set as follow:
flip=False, alpha=True, scale=False, depth=False, zbuff=False.
It fix T46381. Normally BL_Action::Update (manage action time, end, loop…) should be called the same number of times as BL_Action::UpdateIPO (update action position, scale ect… in the game object).
But the bug report shows that UpdateIPO is called one less time than Update. To fix it i revert the commit 362b25b382 and implement a mutex in BL_Action::Update.
Example file : {F245823}
Reviewers: lordloki, kupoman, campbellbarton, youle, moguri, sybren
Reviewed By: youle, moguri, sybren
Maniphest Tasks: T39928, T46381
Differential Revision: https://developer.blender.org/D1562
Same case as with space char really, one should not use those special chars in
filenames, but they are globally supported by all current FS/OS, so no real reason
to enforce that behvior on users here.
To be backported to 'a' release.
All optional image format are not #define'd in submodules like DDS read/write code.
This means values of `eImbTypes` would not always be the same in all contexts, yuck!
This is a regression and should be backported to 'a' release.
Even if the weights are normalized, the weighted sum of normalized vectors
usually does **not** give a normalized vector (unless all source vectors
are aligned).
This probably was not a big issue in most cases, since we usually interpolate
similar vectors here - but still!
The final stage of the process (copying/interpolating new dst cddata from src cddata)
was simply broken in normal case, where we need to convert from source to destination
object space.
This patch is a bit verbose, but I cannot see how to avoid it really.
To think this code is in master since over 6 months and it only gets reported now... :/
'thumbnail_size' is now used in all cases, it controlls column width in other viewmodes
of filebrowser. We cannot (easily) rename that DNA member, but I also renamed RNA
property, and fixed its tooltip...
As with cdderivedmesh, performance here is still CPU-limited if material
needs tangents/UVs/vcolors. Draw calls have much less overhead though.
Also, as with derivedmesh, kept an exception for old drawing for NVIDIA
+OSX+VBO off or setDrawOptions callback not being NULL.
setDrawOptions should be ommitable and fully VBOfialbe (?) in the
future, usually those just check for hidden flag of poly or similar.
Filenames over 128 chars would crash.
Move BLI_newname into file_ops,
this was only used in one place and isn't all that re-usable.
Also remove special behavior for 4 digits.
linking API funcs would use 'name, idcode', when all other code here uses
(more sensible) 'idcode, name'.
Also, use 'name' arg name when we expect a bare name, without the prepended ID code, and
'idname' arg name when we expect a complete ID name.
And here too, idcode shall be short, not int!
Used old (2.49 era) filled style for drawing here, with white color and
alpha blending.
Also changed drawing to do linear interpolation between samples instead
of ugly square wave in high zoom.
This could be improved upon, with real waveform drawing in higher zoom
levels, but I'll leave this for later since it may need some hacking on
audaspace level.
This adds an option to control at what time relative to the current frame
the shutter is fully opened. Supported options are:
- Shutter is starting to open at the current frame
- Shutter is fully opened at the current frame
- Shutter is fully closed at the current frame
Custom shutter time offset is possible, same as custom curve for shutter
openness but those are considered nice things to have rather than something
crucial.
Reviewers: juicyfruit, dingto
Subscribers: venomgfx, hjalti
Differential Revision: https://developer.blender.org/D1380
Camera scale was previously ignored in the BGE.
It is now injected in the modelview matrix as a scale of the scene
coordinates in the camera local reference. This is equivalent to a zoom:
A scale of 2 multiplies the coordinates by 2 => only the points with
coordinates less then 0.5 before the scale will fall in the frustrum =>
equivalent to a zoom.
Anisotropic scale is supported (different scale in x, y, z)
Negative scale is also supported. As an odd number of negative scale flips the
normals of the objects, the OGL front face setting is also flipped to compensate.
A Y negative scale of -1 produces a vertical flip at OGL level.
Title says pretty much everything, we now have BKE and RNA funcs to get vertex, poly and
loop normals of a given shapekey.
This will be used e.g. in FBX exporter (shapekeys need normal data too).
Reviewed By: campbellbarton
Differential Revision: https://developer.blender.org/D1510
New code loops much less, does re-open & close .blend files for each data type,
and is much much more flexible - it is also ready for id-remap & co work being done in branches.
Main idea is to store libraries & datablocks to link in a dedicated struct, in a way
that avoids too much looping, and also allows to search for a single datablock in several libraries.
Here again, no change is expected in current behavior of link/append tool, please report
if anything goes different!
This commits does mostly two things:
* Get rid of bContext parameter: I can see no real good reason to pass such a high-level data
to such low-level code... It also makes it more difficult to call when you do not have
a context available.
* Cleanup the instantiating part.
Last point is the most risky - previous code was sometimes quite confusing and hard to follow,
from tests nothing behaves differently in new code, but some hidden corner case may show up.
Anyway, no change in behavior is expected from this commit, if it happens please file a bugreport!
It turns out libGL from Intel crashes when calling glxewInit (where mesa, nvidia work fine),
unfortunately the only option without making larger changes to glew,
is to inline the parts of glew we're using - before the glx context is created.
Root of the issue is that we do not re-read lib data blocks and ID placholders (ID_ID bheads)
in undo context (in `blo_read_file_internal`), because `BLO_read_from_memfile` copies
lib datablocks and Main data directly from oldmain.
The idea being, linked data do not change from undo/redo.
This is valid as long as linked data was not changed by the undo step - but if some
was deleted or localized, it will be missing from oldmain, leading to data loss
(note that does not only concern objects, all linkable data types can be affected,
at least in theory).
This commit addresses that issue by carefully mixing reuse of needed data from oldmain,
and "normal" re-reading of missing one. Makes us swimming in some rather dark waters,
and gives a rather non-easy-to-follow code, but it seems to work quite well,
and only other solution would be to get rid of that optimization
(not re-reading all libs on undo/redo), which is not acceptable.
Also, thanks to @carlosdp for initial investigation of the issue.
Differential Revision: https://developer.blender.org/D1485
Change the mathutils callback from vector to color for mistColor, backgroundColor and ambientColor.
Reviewers: lordloki, campbellbarton, panzergame, moguri, sybren
Reviewed By: panzergame, moguri, sybren
Projects: #game_engine, #game_logic
Differential Revision: https://developer.blender.org/D1432
Actually we only have a Python API that allows to change the max jumps value.
The patch also allows non programmers to change the maximum numbers of jumps.
Reviewers: panzergame, sybren, campbellbarton, lordloki, moguri, agoose77
Reviewed By: lordloki, moguri
Projects: #game_engine
Differential Revision: https://developer.blender.org/D1302
Title says it all, based on feedback of artists from gooseberry team.
This mainly affects cases when going to a local view from layers setup
when some lamps were on invisible layers. Those lights are no longer
becoming visible to the object in local view.
Reviewers: brecht, juicyfruit, dingto
Reviewed By: juicyfruit, dingto
Subscribers: maxon, eyecandy, venomgfx
Differential Revision: https://developer.blender.org/D1326
The issue was caused by non-continuous tangent space calculated for triangles.
This commit adds a Tangent input to Hair BSDF node which can be used to hook up
Tangent calculated form UV as an input to the node in order to make sure the
tangent space is continuous.
Doing this as an input instead of using default tangent layer from UV because of
several reasons:
- This way it's really easy to preserve compatibility with existing setups.
- Default UV map is not necessarily giving continuous space, one might want to
use other tangent space sources or distort the space for some artistic
decision.
Reviewers: juicyfruit, dingto
Reviewed By: dingto
Differential Revision: https://developer.blender.org/D1428
Currently only image loading benefits of this and will give magenta color
when image manager detects it's running out of memory.
This isn't ideal solution and can't handle all cases. For example, OOM
killer might kill process before it realized it run out of memory, but
in other cases this could prevent some crashes.
Reviewers: juicyfruit, dingto
Differential Revision: https://developer.blender.org/D1502
* Change the character jumping variables and methods from int to char.
* Limit the maxJumps integer value from 0 to 255.
* Allow to set the minimum jump amount to 0.
Reviewers: panzergame, lordloki, moguri
Reviewed By: lordloki, moguri
Subscribers: agoose77
Projects: #game_engine
Differential Revision: https://developer.blender.org/D1305
Currently OpenCL devices are packing images into a single texture,
which means technically number of textures is not limited here.
Now OpenCL will use same number of textures as CPU. If we want
to bump number of textures further, this values are to be modified
in sync.
NOTE OpenCL still does not support float textures.
Original patch from a guy called bliblubli in the tracker with
some own modifications.
Reviewers: brecht, dingto, sergey
Differential Revision: https://developer.blender.org/D1530
The icon16 and icon32_mesh_capsule.dat files are committed with 0 Kb, because I used the patch from the differential from the Phabricator which I done with a regular .diff file.
This patch adds icons to the physic collision shapes.
Adding a new capsule shape 'mesh_capsule' icon which represent the shape better then the metaballs icon.
And replace the metaballs icon for the Blender collision shape.
{F206628}
Reviewers: moguri, sybren, agoose77, lordloki, mont29, panzergame, campbellbarton
Reviewed By: lordloki, panzergame, campbellbarton
Projects: #game_engine, #game_ui, #user_interface
Differential Revision: https://developer.blender.org/D1403
This patch will delete all associated objects from a group instance (dupli group) if the are deleted.
Reviewers: brita_, sybren, lordloki, moguri
Reviewed By: lordloki, moguri
Projects: #game_engine
Maniphest Tasks: T36388
Differential Revision: https://developer.blender.org/D1205
Extending the existing game actuator with a screenshot function, to give also non programmers the ability to take screenshots in the BGE.
Reviewers: lordloki, campbellbarton, moguri
Reviewed By: lordloki, moguri
Subscribers: lordloki, Genome36
Projects: #game_engine
Differential Revision: https://developer.blender.org/D651
Use new interp_m4_m4m4 instead of blend_m4_m4m4.
Note that maybe we could replace other usages of blend_m... by interp_m...,
but this should be investigated on a case-by-case basis.
mat3_polar_decompose gives the right polar decomposition of given matrix,
as a pair (U, P) of matrices.
interp_m3_m3m3 uses that polar decomposition to perform a correct matrix interpolation,
even with non-uniformly scaled ones (where blend_m3_m3m3 would fail).
interp_m4_m4m4 just adds translation interpolation to the _m3 variant.
Ray direction is assumed to be normalized in such areas as scaling intersection
distance on instance push/pop when doing ray-scene intersection, but it was
possible that some closures wouldn't give normalized direction which could cause
wrong intersection checks.
Now normalization will happen on surface bounce, which could be a bit of a waste
if closure actually gives normalized direction, but currently only transparent
BSDF seems to give guaranteed normalized direction.
Issue was caused by wrong intersection distance scaling on instance pop,
which could cause intersection distance to become zero, confusing following
intersection checks.
Looks like instancing of smoke sim is not supported at all
(was fake-working in 3DView in 2.74, but not rendered).
But it should not crash - code was adding temp 'fromdupli' base to the delayed
drawing list...
Nice to backport this to 2.76 I think.
This patch makes it possible to zero out angular velocity. tiny angular
velocities may cause instabilities, according to the discussion in T41943,
so they are mapped to (0, 0, 0) instead.
It also applies the same reasoning to the linear velocity, unifying the
different approaches.
Differential revision: D952
The render pass is enabled by default but it can be disabled with
bge.logic.setRender(False).
Once disabled, the render pass is skipped and a new logic frame starts
immediately. Note that VSync no longer limits the fps when render is off
but the 'Use Frame Rate' option in the Render Properties still limits it.
This commit exposes the interpolation parameter for environment textures (requested by DolpheenDream on IRC), just as it already is for image textures.
Reviewers: sergey
Differential Revision: https://developer.blender.org/D1544
Transition strips in the NLA should always stick to whatever strips are beside it,
and are allowed to grow/shrink as needed to achieve this. Previously the code here
was only checking if the neighbouring strips started encroaching on the transition,
but not whether the transition needed to grow to fill a gap. It was also just
moving all strips when there was insufficient space, even though that would alter
timing down the track.
Now transition strip resizing works as follows:
* It will grow/shrink as necessary to absorb any changes in the length of its neighbours
instead of shunting everything around to maintain its length
* If the neighbour has been resized by an amount greater than the transition's length,
all the strips will need to be shunted away to make way for the neighbour. In this
case, the transition will shrink down to being 1 frame long to ensure that it is
still visible (so that it can be removed if necessary).
The "tweak user" flag used to flag strips using the same action as the active strip
was not getting set on other strips that live on the same track as the active one.
Strips with this flag set are shown with a red colour to indicate that editing the
action may have the unintended consequence of modifying another strip.
Duplicate wasn't updating links,
so duplicatinvg a objects would still point to the originals for curve-taper, texmesh, drivers.
Use generic id-looper to handle replacing data.
- BLI_current_working_dir's return value must be checked, since it may fail.
- BLI_current_working_dir now behaves like getcwd, where a too-small target will return failure.
- avoid buffer overrun with BLI_path_cwd, by taking a maxlen arg.
I have removed the m_pHitObject, m_xray and m_testPropName and replace them by a temporary struct "RayCastData" which contains these datas and a collision mask. Finally i add a collision mask argument in the python function "rayCast" :
```
rayCast(to, from, dist, prop, face, xray, poly, mask)
```
It can be useful to hit only object which are on the right colision layer. for example if you have hitbox for a charater or vehicle you don't want to hit it with raycast.
test file : {F237337}
left mouse click on two planes and see console messages.
Somewhat more elaborate test file by @sybren: {F237779}
Look around and click on the cubes. One cube lamp responds, the other doesn't, based on their collision groups.
Reviewers: moguri, hg1, agoose77, campbellbarton, sybren
Reviewed By: agoose77, campbellbarton, sybren
Subscribers: campbellbarton, sergey, blueprintrandom, sybren
Projects: #game_engine, #game_physics
Differential Revision: https://developer.blender.org/D1239
This func is long enough, there's no real need to make it even longer with
verbose local varnames and multi-line for loops...
Also, avoid mono-leters names as well for data pointers.
We now keep actions around when they are finished playing so scripts can
still get access to information such as the current frame. Playing a new
action in the same layer still overwrites the previous action as before this
commit. Using an explicit KX_GameObject.stopAction() will free the memory. The
action is also freed when the KX_GameObject is freed as before.
We now have to explicitely enure tesselation of DMs when we need it.
Notes: Maybe we could use looptris here as well?
Not a regression (bug already present in 2.75, but not 2.74), nice to backport to 2.76 nontheless.
New raterizer method:
fbo = bge.render.offScreenCreate(width,height)
width, height: size of the FBO, need not be power of two
Returns a PyRASOffScreen object that encapsulates the FBO.
It has just 2 attributes: width and height to return the size of the FBO.
Currently, this object can only be used in the ImageRender constructor.
New optional argument on ImageRender constructor:
ir = bge.texture.ImageRender(scene, camera, fbo)
If present, fbo is an object of type PyRASOffScreen.
The returned ImageRender object holds a reference to it, which ensures
that the FBO livespan is at least as long as the ImageRender object.
One PyRASOffScreen object can be shared by multiple ImageRender objects.
The FBO is automatically destroyed when all refences to the PyRASOffScreen
object are released.
Note: the whole and capSize attributes of the ImageRender object have no
effect when an FBO is used: the render and the capture is automatically
done on the full size of the FBO.
Own regression caused by fix for T46067,
edit-mode bvh only contained unselected faces.
This commit adds support for an edit-mode bvh containing all faces.
Depsgraph customdata flag was set on the parent,
then cleared once the parent object was handled.
This is initialized to zero for new nodes, no need to clear.
Expose our `BKE_image_load_exists` feature through an optional parameter to `Image.load()`.
Extend `image_utils.load_image()` with two optional parameters, to return existing image datablock
if possible, and in that case, to force reloading said image.
Needed by incomming 'import images as planes' addon enhancement.
Also, fix own stupidity, no need to define ID callback in case we only have objects,
calling code knows to fallback to `foreachObjectLink()` when `foreachIDLink()` is missing...
Our version of clang fails with latest SDK. It's not really clear if such
change will disable openmp or not (-fopenmp doesn't throw an error, but
it might be a silent fail).
In any case, builds without OpenMP is better than no builds at all.\
- Match layout from stamp buffer (note last so it can wrap)
- Correct vertical spacing (didn't take decender into account)
- Use U.pixelsize when setting font-size.
It would simply remove default units (in most case), cannot see any good reason for such behavior?
Might work in case default unit is the only one present, but breaks consistency (and it may not be
always obvious for user which unit it is). Comes from original patch from five years ago, rB7d8f0fce.
This will break keyboard-setting of values, e.g. '2m 33.4cm' would become '2 33.4cm',
totally useless and invalid entry!
Issue was caused by changed function signature. This is still not really full
support of new OSL API since we don't store anything in the derivatives which
could confuse mipmapping.
Would write 1.04 seconds as `00:00:01,40` instead of `00:00:01,040`...
Anyway, we already have BLI API for timecodes, much better to add
SubRip timecode format there, heavily simplifies code.
To be backported to final 2.76.
Instead of offsetting along normals, smooth positions are now
calculated on a sphere defined by the vertices and their normals.
This removes visible seams along original edges, which were common previously.
E.g. trackball-arrows (R-R) were no more visible.
Caused by wrong cast of an array from int to float, we need an int version of those helpers (`glTranslatenxv`) too.
Broken by font wrap commit rBf2341f829654c4dc97, there was actually two things here:
* Using non-initialized color (which lead to transparent drawing...), we need to use new
`blf_draw_buffer__start/end` helpers here too, made them shared internally.
* Using `draw_str_i18n_nbr` as `draw_str[i]` length, ugh! That's the number of utf8 glyphs of
translated string, not the length of untranslated string!
This fix must be backported to final 2.76.
Cleanup and rework a bit text_cursor_set_to_pos_wrapped(), it did not handle correctly negative y valuesi,
because it was only checking lines starting from first visible one.
Mostly, we now directly get line matching given y position (be it an invisible one) from the helper
(renamed `get_line_pos_wrapped()`), instead of first visible one, which allows us to get rid of
additional complexity of looping over next lines until we find correct one.
This code remains rather complex to follow, added some asserts to ensure everything works as expected,
and tested it rather seriously, but DO NOT backport this to 2.76!
If t->mode remains edge/vert slide, restoreTransObjects() ends up calling
projectVert/EdgeSlideData(), which tries to access invalid customdata...
Not sure why we call again restoreTransObjects() and resetTransRestrictions() here tbh,
but safer not to change that for now.
Should be backported to 2.76 if possible.
More precise description of the issue: Clicking on the right side of a num-slider button with Continuous Grab enabled sometimes caused cursor to jump to the inner value indicator even though text editing was started.
Happened because data->ungrab_mval wasn't reset correctly after dragging.
Windows-only bug, mmap & co are not threadsafe by default on this platform, so we have to add a dedicated
spinlock for them in win32.
Note that we may try to get rid of those mmap later, but not for 2.76!
To be backported to final 2.76...
The issue was introduced by original Cycles point density support commit,
it lead to a constant density of 1 for object verticies point density source.
Mostly added missing :arg: clauses -- these caused the :type:-clauses
that were already in the RST files to be hidden from the output HTML.
I also fixed some argument types from 'list' to 'Vector' where applicable
and corrected some obvious mistakes.
We need custom context here, those are often very short names so context collision is pretty easy.
Also some minor changes (and avoid shadowing varnames)...
This is a mere bandage, that whole area is known broken anyway, but at least it should prevent the crash.
Note that that kind of stuff (the efd->index being a pointer) is really bad practice imho...
Should be backported to final 2.76.
If provided, the parameter must be an object that supports the buffer protocol
(bytearray, memoryview, etc) otherwise a runtime error is generated. If the
buffer is sufficiently large, the image is copied in the buffer before being
refreshed (i.e. invalidated). If the image was not already available, it will
be updated first. In the later case, it is possible that the image is loaded
directly in the user's buffer wihtout an intermediate copy in the internal
image buffer. This is currently the case for ImageViewport and ImageRender
sources when the OGL format matches the buffer format (alpha, no filter,
no flip, no scaling). Note that the image format in the buffer is always RGBA.
If no parameter is provided, the method works as before: the image is
invalidated without any attempt to updated it first.
The function returns False if a buffer was provided but could not be updated
for any reason (source not ready, buffer too small). It returns True in all
other situations.
The purpose of this function is to efficiently retrieve the OGL frame buffer
directly into a user buffer by skiping an extra copy to the internal image
buffer if it's not needed.
Stereo mode disturbs ImageRender by changing some OGL state.
It was already fixed for Quadbuffer mode, but not yet for anaglyph and
interlaced mode. This fix simply clears the color mask and stipple mode
that are specific to these stereo mode. In any case, the correct state
is restored prior the render of each eye.
Certain GPU (intel) will not allow MSAA together with swap copy.
Previously, swap copy had priority over MSAA: fewer AA samples would be
chosen if it was the condition to get swap copy. This patch reverse the
logic: swap copy will be abandonned if another swap method (undefined or
exchange) will provide the number of AA samples requested. If no AA
samples is requested, swap copy still has the priority of course.
Use mlock() to lock memory in case of pinned memory extension.
Use get/setrlimit to set limit so thqt locking will work.
However, only privilege process can increase the limit, so ignore error.
AMD_pinned_memory method implemented according to BMD sample file but
not tested (would need an ATI card).
Fix OGL method: needed to allocate texture data first.
The Windows and asm variant were returning the value, of the variable
before the add or sub operation. All the other variants were returning
the value after the operation. Now all variants return the new value.
Capture works: I can see the frame coming in but passing them to the GPU
doesn't work yet. I'm using standard OGL function though.. TBC.
them to the GPU.
ImageViewport will now skip the unnecessary RGBA32 filtering if the capture
is already in the RGBA format. This will be the case if alpha=True and no
filter is set. This saves a great deal of CPU.
Add swap attribute to DeckLink to skip pixel format conversion between
VideoTexture internal (RGBA) and DeckLink (BGRA). This is to test the
benefit of passing the image directly to the board without intermediate
copy => great boost of performance.
However, it will be necessary to change the VideoTexture internal pixel
format from RGBA to BGRA.
3D stream is now fully supported on DeckLink, the right eye image is
passed via a new 'right' attribute (name not definitive). To generate this
image one can simply put a second camera to figure the right eye and
capture its view with ImageRende.
A new option '-a' can be passed to the blenderplayer. It forces the
framebuffer to have an alpha channel. As the background color is applied
with 0 alpha, ImageViewport will return an image with transparent
background (provided alpha is set to True on the ImageViewport object).
This is useful in combination with DeckLink to generate a key frame
with transparent background.
Without the -a option, the frame buffer has no alpha channel and
ImageViewport always returns an opaque image, no matter what.
This should be working in Windows and Linux, but only Windows is tested at
present.
Note that although the frame buffer has alpha, the player window is still
rendered opaque. Transparency of the window is possible using the
'compositing' functions of Windows and the code is already there for that
(look for WIN32_COMPOSITING) but is not enabled because 1) it doesn't work
so well 2) it requires a DLL that is only available on Vista and up.
The DVP library allows direct memory transfer with the GPU to send or
receive textures. It's used in bge.Texture.VideoDeckLink to sent a
captured video stream efficiently to the GPU, only for Windows.
The user will need to download the SDK version 10.3.1 from BlackMagicDesign
and copy dvp.dll where blender can find it (e.g. next to blender.exe)
Implement the DeckLink object to output on DeckLink card. It takes a
source like the Texture object and gets the image the same way.
API:
dl = bge.texture.DeckLink(0, "HD1080p24") # card nb, mode
dl.source = bge.texture.ImageViewport()
dl.source.whole = True
dl.refresh(True)
Known issue: using ImageViewport as source will not set the alpha channel
right. This is because the framebuffer by default will not have an alpha
channel. This can be solved either by inserting a 'green' filter that will
convert the background color to alpha, or enable the alpha channel on the
framebuffer (require change in Ghost).
This object is analoguous to Texture and is used to send frames out to a
DeckLink card. It is primarly intended to mix computed frames with a video
stream looping through the card for real-time compositing. This is known
as 'keying'.
Currently the implementation is only a stub.
Fix a bug in setUniformEyef implementation where the caching system for
uniform was getting in the way: a 'eye' uniform needs to be updated on
every frame and for each eye in stereo mode.
BL_Shader.setUniformEyef() python method to define a uniform that will
receive the eye value as a float: 0.0 when rendering the left eye, 0.5
when rendering the right eye.
This can be used in a shader to split a 3D texture sent to the GPU by
VideoDeckLink as the left+right eye frame is sent as a single texture in
bottom/top order.
Work around a bug in VideoDeckLink with the the DVP library: the call to
dvpBindToGLCtx() fails for the first frame. This needs to be investigated
but for now catch the exception thrown by this condition and release the
decklink frame anyway as otherwise it corrupts the reference count of the
decklink device.
The VideoDeckLink object is still a stub but now it compiles and links
with DeckLink COM API and nVidia DVP library.
Only Windows & CMake working at the moment.
The nVidia DVP library is added in SVN win64_vc12 repo.
Decklink capture cards from BlackMagicDesign are capable of capturing
high speed video stream such as FullHD on HDMI, also 3D. It will
be available in the VideoTexture module as a new type of texture source.
The capture data will be sent to the GPU using GPUDirect for best
performance.
It will be available for Linux and Windows.
This first commit is only a stub for Linux. Windows to be added.
2015-03-21 14:50:30 +01:00
3223 changed files with 181134 additions and 130956 deletions
(BoolVariable('WITH_BF_PYTHON','Compile with python',True)),
(BoolVariable('WITH_BF_PYTHON_SAFETY','Internal API error checking to track invalid data to prevent crash on access (at the expense of some effeciency)',False)),
(BoolVariable('WITH_BF_PYTHON_SECURITY','Disables execution of scripts within blend files by default',True)),
('BF_PYTHON','Base path for python',''),
('BF_PYTHON_VERSION','Python version to use',''),
('BF_PYTHON_INC','Include path for Python headers',''),
('BF_PYTHON_BINARY','Path to the Python interpreter',''),
('BF_PYTHON_LIB','Python library',''),
('BF_PYTHON_LIBPATH_ARCH','Additional Python library under some linux flavors (arch-dependent)',''),
('BF_PYTHON_DLL','Python dll - used on Windows only',''),
(BoolVariable('WITHOUT_BF_PYTHON_UNPACK','dont remove and unpack Python modules everytime if true',False)),
(BoolVariable('WITHOUT_BF_OVERWRITE_INSTALL','dont remove existing files before breating the new install directory (set to False when making packages for others)',False)),
(BoolVariable('BF_FANCY','Enable fancy output if true',True)),
(BoolVariable('BF_QUIET','Enable silent output if true',True)),
(BoolVariable('BF_LINE_OVERWRITE','Enable overwriting of compile line in BF_QUIET mode if true',False)),
Right eye being used during stereoscopic rendering.
..data:: RAS_OFS_RENDER_BUFFER
The pixel buffer for offscreen render is a RenderBuffer. Argument to :func:`offScreenCreate`
..data:: RAS_OFS_RENDER_TEXTURE
The pixel buffer for offscreen render is a Texture. Argument to :func:`offScreenCreate`
*****
Types
*****
..class:: RASOffScreen
An off-screen render buffer object.
Use :func:`offScreenCreate` to create it.
Currently it can only be used in the :class:`bge.texture.ImageRender` constructor to render on a FBO rather than the
default viewport.
..attribute:: width
The width in pixel of the FBO
:type:integer
..attribute:: height
The height in pixel of the FBO
:type:integer
..attribute:: color
The underlying OpenGL bind code of the texture object that holds the rendered image, 0 if the FBO is using RenderBuffer. The choice between RenderBuffer and Texture is determined by the target argument of :func:`offScreenCreate`.
:type:integer
*********
Functions
@@ -113,7 +150,9 @@ Functions
..note:: Only works in the standalone player, not the Blender-embedded player.
:arg width:width in pixels
:type width:integer
:arg height:height in pixels
:type height:integer
..function:: setFullScreen(enable)
@@ -122,6 +161,7 @@ Functions
..note:: Only works in the standalone player, not the Blender-embedded player.
:arg enable:``True`` to set full screen, ``False`` to set windowed.
:type enable:bool
..function:: getFullScreen()
@@ -134,9 +174,12 @@ Functions
..function:: getDisplayDimensions()
Get the actual display dimensions, in pixels, of the physical display (e.g., the monitor).
Get the display dimensions, in pixels, of the display (e.g., the
monitor). Can return the size of the entire view, so the
combination of all monitors; for example, ``(3840, 1080)`` for two
side-by-side 1080p monitors.
:type dimension:list [width,heigh]
:rtype:tuple (width,height)
..function:: makeScreenshot(filename)
@@ -161,13 +204,14 @@ Functions
..function:: enableVisibility(visible)
Doesn't really do anything...
Deprecated; doesn't do anything.
..function:: showMouse(visible)
Enables or disables the operating system mouse cursor.
:arg visible:
:type visible:boolean
@@ -175,15 +219,15 @@ Functions
Sets the mouse cursor position.
:arg x:X-coordinate in screen pixel coordinates.
:type x:integer
:arg y:Y-coordinate in screen pixel coordinates.
:type y:integer
..function:: setBackgroundColor(rgba)
Sets the window background color. (Deprecated: use KX_WorldInfo.background_color)
:type rgba:list [r, g, b, a]
Deprecated and no longer functional. Use :py:meth:`bge.types.KX_WorldInfo.backgroundColor` instead.
..function:: setEyeSeparation(eyesep)
@@ -227,6 +271,7 @@ Functions
Set the material mode to use for OpenGL rendering.
:arg samples:the number of multisample for anti-aliasing (MSAA), 0 to disable MSAA
:type samples:integer
:arg target:the pixel storage: :data:`RAS_OFS_RENDER_BUFFER` to render on RenderBuffers (the default), :data:`RAS_OFS_RENDER_TEXTURE` to render on texture. The later is interesting if you want to access the texture directly (see :attr:`RASOffScreen.color`). Otherwise the default is preferable as it's more widely supported by GPUs and more efficient. If the GPU does not support MSAA+Texture (e.g. Intel HD GPU), MSAA will be disabled.
Refresh video - get its status and optionally copy the frame to an external buffer.
:arg buffer:An optional object that implements the buffer protocol. If specified, the image is copied to the buffer, which must be big enough or an exception is thrown.
:type buffer:any buffer type
:arg format:An optional image format specifier for the image that will be copied to the buffer. Only valid values are "RGBA" or "BGRA"
:type format:str
:arg ts:An optional timestamp (in seconds from the start of the movie) of the frame to be copied to the buffer.
:type ts:float
:return:see `FFmpeg Video and Image Status`_.
:rtype:int
*************
@@ -244,12 +256,15 @@ Image classes
*:class:`FilterRGB24`
*:class:`FilterRGBA32`
..method:: refresh()
..method:: refresh(buffer=None, format="RGBA")
Refresh image, i.e. load it.
Refresh image, get its status and optionally copy the frame to an external buffer.
:value:see `FFmpeg Video and Image Status`_.
:arg buffer:An optional object that implements the buffer protocol. If specified, the image is copied to the buffer, which must be big enough or an exception is thrown.
:type buffer:any buffer type
:arg format:An optional image format specifier for the image that will be copied to the buffer. Only valid values are "RGBA" or "BGRA"
:type format:str
:return:see `FFmpeg Video and Image Status`_.
:rtype:int
..method:: reload(newname=None)
@@ -411,9 +426,14 @@ Image classes
:type::class:`~bgl.Buffer` or None
..method:: refresh()
..method:: refresh(buffer=None, format="RGBA")
Refresh image - invalidate its current content.
Refresh image - render and copy the image to an external buffer (optional) then invalidate its current content.
:arg buffer:An optional object that implements the buffer protocol. If specified, the image is rendered and copied to the buffer, which must be big enough or an exception is thrown.
:type buffer:any buffer type
:arg format:An optional image format specifier for the image that will be copied to the buffer. Only valid values are "RGBA" or "BGRA"
:type format:str
..attribute:: scale
@@ -498,9 +518,14 @@ Image classes
:type::class:`~bgl.Buffer` or None
..method:: refresh()
..method:: refresh(buffer=None, format="RGBA")
Refresh image - invalidate its current content.
Refresh image - calculate and copy the image to an external buffer (optional) then invalidate its current content.
:arg buffer:An optional object that implements the buffer protocol. If specified, the image is calculated and copied to the buffer, which must be big enough or an exception is thrown.
:type buffer:any buffer type
:arg format:An optional image format specifier for the image that will be copied to the buffer. Only valid values are "RGBA" or "BGRA"
:type format:str
..attribute:: scale
@@ -545,14 +570,17 @@ Image classes
:type:bool
..class:: ImageRender(scene, camera)
..class:: ImageRender(scene, camera, fbo=None)
Image source from render.
Image source from render. The render is done on a custom framebuffer object if fbo is specified, otherwise on
the default framebuffer.
:arg scene:Scene in which the image has to be taken.
:type scene::class:`~bge.types.KX_Scene`
:arg camera:Camera from which the image has to be taken.
Render the scene but do not extract the pixels yet. The function returns as soon as the render commands have been send to the GPU. The render will proceed asynchronously in the GPU while the host can perform other tasks. To complete the render, you can either call :func:`refresh` directly of refresh the texture of which this object is the source. This method is useful to implement asynchronous render for optimal performance: call render() on frame n and refresh() on frame n+1 to give as much as time as possible to the GPU to render the frame while the game engine can perform other tasks.
:return:True if the render was initiated, False if the render cannot be performed (e.g. the camera is active)
:rtype:bool
..method:: refresh()
..method:: refresh(buffer, format="RGBA")
Refresh video - render and optionally copy the image to an external buffer then invalidate its current content. The render may have been started earlier with the :func:`render` method, in which case this function simply waits for the render operations to complete. When called without argument, the pixels are not extracted but the render is guaranteed to be completed when the function returns. This only makes sense with offscreen render on texture target (see :func:`~bge.render.offScreenCreate`).
:arg buffer:An object that implements the buffer protocol. If specified, the image is copied to the buffer, which must be big enough or an exception is thrown. The transfer to the buffer is optimal if no processing of the image is needed. This is the case if flip=False, alpha=True, scale=False, whole=True, depth=False, zbuff=False and no filter is set.
:type buffer:any buffer type of sufficient size
:arg format:An optional image format specifier for the image that will be copied to the buffer. Only valid values are "RGBA" or "BGRA"
:type format:str
:return:True if the render is complete, False if the render cannot be performed (e.g. the camera is active)
:rtype:bool
..class:: ImageViewport
Image source from viewport.
@@ -689,9 +732,14 @@ Image classes
:type:sequence of two ints
..method:: refresh()
..method:: refresh(buffer=None, format="RGBA")
Refresh image - invalidate its current content.
Refresh video - copy the viewport to an external buffer (optional) then invalidate its current content.
:arg buffer:An optional object that implements the buffer protocol. If specified, the image is copied to the buffer, which must be big enough or an exception is thrown. The transfer to the buffer is optimal if no processing of the image is needed. This is the case if flip=False, alpha=True, scale=False, whole=True, depth=False, zbuff=False and no filter is set.
:type buffer:any buffer type
:arg format:An optional image format specifier for the image that will be copied to the buffer. Only valid values are "RGBA" or "BGRA"
:type format:str
..attribute:: scale
@@ -730,7 +778,174 @@ Image classes
:type:bool
..class:: VideoDeckLink(format, capture=0)
Image source from an external video stream captured with a DeckLink video card from
Black Magic Design.
Before this source can be used, a DeckLink hardware device must be installed, it can be a PCIe card
or a USB device, and the 'Desktop Video' software package (version 10.4 or above must be installed)
on the host as described in the DeckLink documentation.
If in addition you have a recent nVideo Quadro card, you can benefit from the 'GPUDirect' technology
to push the captured video frame very efficiently to the GPU. For this you need to install the
'DeckLink SDK' version 10.4 or above and copy the 'dvp.dll' runtime library to Blender's
installation directory or to any other place where Blender can load a DLL from.
:arg format:string describing the video format to be captured.
:type format:str
:arg capture:Card number from which the input video must be captured.
:type capture:int
The format argument must be written as “<displayMode>/<pixelFormat>[/3D][:<cacheSize>]” where <displayMode>
describes the frame size and rate and <pixelFormat> the encoding of the pixels.
The optional /3D suffix is to be used if the video stream is stereo with a left and right eye feed.
The optional :<cacheSize> suffix determines the number of the video frames kept in cache, by default 8.
Some DeckLink cards won't work below a certain cache size. The default value 8 should be sufficient for all cards.
You may try to reduce the cache size to reduce the memory footprint. For example the The 4K Extreme is known
to work with 3 frames only, the Extreme 2 needs 4 frames and the Intensity Shuttle needs 6 frames, etc.
Reducing the cache size may be useful when Decklink is used in conjunction with GPUDirect: all frames must be locked
in memory in that case and that puts a lot of pressure on memory. If you reduce the cache size too much,
you'll get no error but no video feed either.
The valid <displayMode> values are copied from the 'BMDDisplayMode' enum in the DeckLink API
without the 'bmdMode' prefix. In case a mode that is not in this list is added in a later version
of the SDK, it is also possible to specify the 4 letters of the internal code for that mode.
You will find the internal code in the DeckLinkAPIModes.h file that is part of the SDK.
Here is for reference the full list of supported display modes with their equivalent internal code:
* NTSC 'ntsc'
* NTSC2398 'nt23'
* PAL 'pal '
* NTSCp 'ntsp'
* PALp 'palp'
HD 1080 Modes:
* HD1080p2398 '23ps'
* HD1080p24 '24ps'
* HD1080p25 'Hp25'
* HD1080p2997 'Hp29'
* HD1080p30 'Hp30'
* HD1080i50 'Hi50'
* HD1080i5994 'Hi59'
* HD1080i6000 'Hi60'
* HD1080p50 'Hp50'
* HD1080p5994 'Hp59'
* HD1080p6000 'Hp60'
HD 720 Modes:
* HD720p50 'hp50'
* HD720p5994 'hp59'
* HD720p60 'hp60'
2k Modes
* 2k2398 '2k23'
* 2k24 '2k24'
* 2k25 '2k25'
4k Modes
* 4K2160p2398 '4k23'
* 4K2160p24 '4k24'
* 4K2160p25 '4k25'
* 4K2160p2997 '4k29'
* 4K2160p30 '4k30'
* 4K2160p50 '4k50'
* 4K2160p5994 '4k59'
* 4K2160p60 '4k60'
Most of names are self explanatory. If necessary refer to the DeckLink API documentation for more information.
Similarly, <pixelFormat> is copied from the BMDPixelFormat enum.
Here is for reference the full list of supported pixel format and their equivalent internal code:
* 8BitYUV '2vuy'
* 10BitYUV 'v210'
* 8BitARGB * no equivalent code *
* 8BitBGRA 'BGRA'
* 10BitRGB 'r210'
* 12BitRGB 'R12B'
* 12BitRGBLE 'R12L'
* 10BitRGBXLE 'R10l'
* 10BitRGBX 'R10b'
Refer to the DeckLink SDK documentation for a full description of these pixel format.
It is important to understand them as the decoding of the pixels is NOT done in VideoTexture
for performance reason. Instead a specific shader must be used to decode the pixel in the GPU.
Only the '8BitARGB', '8BitBGRA' and '10BitRGBXLE' pixel formats are mapped directly to OpenGL RGB float textures.
The '8BitYUV' and '10BitYUV' pixel formats are mapped to openGL RGB float texture but require a shader to decode.
The other pixel formats are sent as a 'GL_RED_INTEGER' texture (i.e. a texture with only the
red channel coded as an unsigned 32 bit integer) and are not recommended for use.
Example: “HD1080p24/10BitYUV/3D:4” is equivalent to “24ps/v210/3D:4” and represents a full HD stereo feed at 24 frame per second and 4 frames cache size.
Although video format auto detection is possible with certain DeckLink devices, the corresponding
API is NOT implemented in the BGE. Therefore it is important to specify the format string that
matches exactly the video feed. If the format is wrong, no frame will be captured.
It should be noted that the pixel format that you need to specify is not necessarily the actual
format in the video feed. For example, the 4K Extreme card delivers 8bit RGBs pixels in the
'10BitRGBXLE' format. Use the 'Media Express' application included in 'Desktop Video' to discover
which pixel format works for a particular video stream.
..attribute:: status
Status of the capture: 1=ready to use, 2=capturing, 3=stopped
:type:int
..attribute:: framerate
Capture frame rate as computed from the video format.
:type:float
..attribute:: valid
Tells if the image attribute can be used to retrieve the image.
Always False in this implementation (the image is not available at python level)
:type:bool
..attribute:: image
The image data. Always None in this implementation.
:type::class:`~bgl.Buffer` or None
..attribute:: size
The size of the frame in pixel.
Stereo frames have double the height of the video frame, i.e. 3D is delivered to the GPU
as a single image in top-bottom order, left eye on top.
:type:(int,int)
..attribute:: scale
Not used in this object.
:type:bool
..attribute:: flip
Not used in this object.
:type:bool
..attribute:: filter
Not used in this object.
..method:: play()
Kick-off the capture after creation of the object.
:return:True if the capture could be started, False otherwise.
:rtype:bool
..method:: pause()
Temporary stops the capture. Use play() to restart it.
:return:True if the capture could be paused, False otherwise.
:rtype:bool
..method:: stop()
Stops the capture.
:return:True if the capture could be stopped, False otherwise.
:rtype:bool
***************
Texture classes
***************
@@ -782,6 +997,7 @@ Texture classes
:type:one of...
*:class:`VideoFFmpeg`
*:class:`VideoDeckLink`
*:class:`ImageFFmpeg`
*:class:`ImageBuff`
*:class:`ImageMirror`
@@ -789,7 +1005,125 @@ Texture classes
*:class:`ImageRender`
*:class:`ImageViewport`
..class:: DeckLink(cardIdx=0, format="")
Certain DeckLink devices can be used to playback video: the host sends video frames regularly
for immediate or scheduled playback. The video feed is outputted on HDMI or SDI interfaces.
This class supports the immediate playback mode: it has a source attribute that is assigned
one of the source object in the bge.texture module. Refreshing the DeckLink object causes
the image source to be computed and sent to the DeckLink device for immediate transmission
on the output interfaces. Keying is supported: it allows to composite the frame with an
input video feed that transits through the DeckLink card.
:arg cardIdx:Number of the card to be used for output (0=first card). It should be noted that DeckLink devices are usually half duplex: they can either be used for capture or playback but not both at the same time.
:type cardIdx:int
:arg format:String representing the display mode of the output feed.
:type format:str
The default value of the format argument is reserved for auto detection but it is currently
not supported (it will generate a runtime error) and thus the video format must be explicitly
specified. If keying is the goal (see keying attributes), the format must match exactly the
input video feed, otherwise it can be any format supported by the device (there will be a
runtime error if not).
The format of the string is “<displayMode>[/3D]”.
Refer to :class:`VideoDeckLink` to get the list of acceptable <displayMode>.
The optional “/3D” suffix is used to create a stereo 3D feed. In that case the 'right' attribute
must also be set to specify the image source for the right eye.
Note: The pixel format is not specified here because it is always BGRA. The alpha channel is
used in keying to mix the source with the input video feed, otherwise it is not used.
If a conversion is needed to match the native video format, it is done inside the DeckLink driver
or device.
..attribute:: source
This attribute must be set to one of the image source. If the image size does not fit exactly
the frame size, the extend attribute determines what to do.
For best performance, the source image should match exactly the size of the output frame.
A further optimization is achieved if the image source object is ImageViewport or ImageRender
set for whole viewport, flip disabled and no filter: the GL frame buffer is copied directly
to the image buffer and directly from there to the DeckLink card (hence no buffer to buffer
copy inside VideoTexture).
:type:one of...
*:class:`VideoFFmpeg`
*:class:`VideoDeckLink`
*:class:`ImageFFmpeg`
*:class:`ImageBuff`
*:class:`ImageMirror`
*:class:`ImageMix`
*:class:`ImageRender`
*:class:`ImageViewport`
..attribute:: right
If the video format is stereo 3D, this attribute should be set to an image source object
that will produce the right eye images. If the goal is to render the BGE scene in 3D,
it can be achieved with 2 cameras, one for each eye, used by 2 ImageRender with an offscreen
render buffer that is just the size of the video frame.
:type:one of...
*:class:`VideoFFmpeg`
*:class:`VideoDeckLink`
*:class:`ImageFFmpeg`
*:class:`ImageBuff`
*:class:`ImageMirror`
*:class:`ImageMix`
*:class:`ImageRender`
*:class:`ImageViewport`
..attribute:: keying
Specify if keying is enabled. False (default): the output frame is sent unmodified on
the output interface (in that case no input video is required). True: the output frame
is mixed with the input video, using the alpha channel to blend the two images and the
combination is sent on the output interface.
:type:bool
..attribute:: level
If keying is enabled, sets the keying level from 0 to 255. This value is a global alpha value
that multiplies the alpha channel of the image source. Use 255 (the default) to keep the alpha
channel unmodified, 0 to make the output frame totally transparent.
:type:int
..attribute:: extend
Determines how the image source should be mapped if the size does not fit the video frame size.
* False (the default): map the image pixel by pixel.
If the image size is smaller than the frame size, extra space around the image is filled with
0-alpha black. If it is larger, the image is cropped to fit the frame size.
* True: the image is scaled by the nearest neighbor algorithm to fit the frame size.
The scaling is fast but poor quality. For best results, always adjust the image source to
match the size of the output video.
:type:bool
..method:: close()
Close the DeckLink device and release all resources. After calling this method,
the object cannot be reactivated, it must be destroyed and a new DeckLink object
created from fresh to restart the output.
..method:: refresh(refresh_source,ts)
This method must be called frequently to update the output frame in the DeckLink device.
:arg refresh_source:True if the source objects image buffer should be invalidated after being
used to compute the output frame. This triggers the recomputing of the
source image on next refresh, which is normally the desired effect.
False if the image source buffer should stay valid and reused on next refresh.
Note that the DeckLink device stores the output frame and replays until a
new frame is sent from the host. Thus, it is not necessary to refresh the
DeckLink object if it is known that the image source has not changed.
:type refresh_source:bool
:arg ts:The timestamp value passed to the image source object to compute the image. If unspecified, the BGE clock is used.
Look from a point/object to another point/object and find first object hit within dist that matches prop.
if poly is 0, returns a 3-tuple with object reference, hit point and hit normal or (None, None, None) if no hit.
@@ -851,6 +851,8 @@ base class --- :class:`SCA_IObject`
* 2: return value is a 5-tuple and the 5th element is a 2-tuple (u, v) with the UV mapping of the hit point or None if no hit, or the object doesn't use a mesh collision shape, or doesn't have a UV mapping.
:type poly:integer
:arg mask:collision mask: The collision mask (16 layers mapped to a 16-bit integer) is combined with each object's collision group, to hit only a subset of the objects in the scene. Only those objects for which ``collisionGroup & mask`` is true can be hit.
:type mask:bitfield
:return:(object, hitpoint, hitnormal) or (object, hitpoint, hitnormal, polygon) or (object, hitpoint, hitnormal, polygon, hituv).
* object, hitpoint and hitnormal are None if no hit.
String formatting. Use this when you're writing string data from floats and ints
``String formatting`` -
use this when you are writing string data from floats and ints.
>>> file.write("%s %s %s\n" % (str1, str2, str3))
Python’s string joining function. To join a list of strings
``String join() function``
use to join a list of strings (the list may be temporary). In the following example, the strings are joined with a space " " in between, other examples are "" or ", ".
* This module provides a sparse matrix representation, and basic associatd matrix manipulations
* This module provides a sparse matrix representation, and basic associated matrix manipulations
* and operations.
*
* See the \ref TutorialSparse "Sparse tutorial"
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
