Previous fix for T53430 caused T54200.
The edge case for soft & hard cuts weren't working,
where the strip used start/end-still & the frame was placed exactly on
the start/end of of the sequence content.
T54200 fixed the end-still case but broke hard-cuts for all other cases.
This fixes the case for soft/hard cuts with/without start/end-still.
Now repeating the operator will use the previously chosen offset, either with
the modal operator or typed in. The modal operator will still start at zero.
The check to see if `use_advanced_hair` was enabled was actually in two places
(render panel `draw` function and physics panel `poll` function). As these
properties are only in one place now the check in `draw` isn't needed anymore.
Related: T53513, a6c69ca57f
We shouldn't mix image pool acuisition with and without user provided,
the fact that internally image.c uses last frame from Image datablock
confuses the logic.
This can be very slow if it contains a big texture, and it's not
necessarily setup in a useful way anyway, and materials can be used
in multiple scenes.
Was happening when viewport visibility on the particle system is disabled.
This became an issue after c45afcf, but the actual issue goes a bit deeper
and the following aspects were involved:
- Relations builder for particle system was ignoring particle system if
it's visibility is not enabled for viewport. This is something what
shouldn't have been done -- depsgraph relations are supposed to be the
same no matter if it's viewport or render.
- Relation builder was only dealing with duplication set to object, but
was ignoring group duplication.
This is NOT a regression since 2.79, but a regression since 2.79a-rc1.
A single diagonal axis was used for sorting coordinates,
the algorithm relied on users not having vertices axis aligned.
Use BLI_kdtree to remove doubles instead.
Overall speed varies, it's more predictable than the previous method.
Some typical tests gave speedup of ~1.4x - 1.7x.
FFMPEG uses int for the numerator, while Blender uses a short. So in
cases people gave weird exotic framerate values and we cannot reduce
enough the numerator, we'd get totally weird values (even negative frame
rates sometimes!)
Now we add checks for short overflow and approximate as best as possible
in that case (error should not matter unless you have shots of at least
several hundreds of hours ;) ).
This reverts commit dc2617130b, which disabled
writing of previews for undo. While this uses some memory, re-rendering all
previews is very expensive, especially if for example you have lots of materials
using high-res image textures.
The idea is to avoid any threading overhead when we start pushing tasks in a
loop. Similarly to how we do it from the new dependency graph. Gives couple of
percent of speedup here, but also improves scalability.
We can not store pointers to elements of collection property in the
case we modify that collection. This is like storing pointers to
elements of array before calling realloc().
The issue was caused by light sample being evaluated to nan at some point.
This is root of the cause which is to be fixed, but is very hard to trace down
especially via ssh (the issue only happens on AVX2 release build). Will give it
a closer look when back to my AVX2 machine.
For until then this is a good check to have anyway, it corresponds to what's
happening in regular radiance sum.
Our own implementation was behaving different comparing to OSL and GPU,
namely on the border pixels OSL and CUDA was doing interpolation with
black, but we were clamping coordinate.
This partially fixes issue reported in T53452.
Similar change should also be done for 3D interpolation perhaps, but this
is to be investigated separately.
This would lead to sock.default_value pointing to the wrong data type,
possibly causing crashes. Unfortunately, this bug will still exist for
older Blender versions that try to load newer files, which makes
changing the type of a node socket problematic.
Drawing hair weights read before the hair array start.
This code could be improved since it currently copy-pastes,
from do_particle_interpolation, but this would need larger changes.
For now just correct existing logic.
Solves these security issues from T52924:
CVE-2017-12102
CVE-2017-12103
CVE-2017-12104
While the specific overflow issue may be fixed, loading the repro .blend
files may still crash because they are incomplete and corrupt. The way
they crash may be impossible to exploit, but this is difficult to prove.
Differential Revision: https://developer.blender.org/D3002
Solves these security issues from T52924:
CVE-2017-12081
CVE-2017-12082
CVE-2017-12086
CVE-2017-12099
CVE-2017-12100
CVE-2017-12101
CVE-2017-12105
While the specific overflow issue may be fixed, loading the repro .blend
files may still crash because they are incomplete and corrupt. The way
they crash may be impossible to exploit, but this is difficult to prove.
Differential Revision: https://developer.blender.org/D3002
CUDA 9.0.176 apparently caused some slow down on high-end Pascal cards that can be mitigated by increasing the number of registers. See https://developer.blender.org/F1142667 for a detailed comparison.
We tried to do as much as possible in a single threaded callback, which
lead to using some nasty tricks like fake atomic-based spinlocks to
perform some operations (like float addition, which has no atomic
intrinsics).
While OK with 'standard' low number of working threads (8-16), because
collision were rather rare and implied memory barrier not *that* much
overhead, this performed poorly with more powerful systems reaching the
100 of threads and beyond (like workstations or render farm hardware).
There, both memory barrier overhead and more frequent collisions would
have significant impact on performances.
This was addressed by splitting further the process, we now have three
loops, one over polys, loops and vertices, and we added an intermediate
storage for weighted loop normals. This allows to avoid completely any
atomic operation in body of threaded loops, which should fix scalability
issues. This costs us slightly higher temp memory usage (something like
50Mb per million of polygons on average), but looks like acceptable
tradeoff.
Further more, tests showed that we could gain an additional ~7% of speed
in computing normals of heavy meshes, by also parallelizing the last two
loops (might be 1 or 2% on overall mesh update at best...).
Note that further tweaking in this code should be possible once Sergey
adds the 'minimum batch size' option to threaded foreach API, since very
light loops like the one on loops (mere v3 addition) require much bigger
batches than heavier code (like the one on polys) to keep optimal
performances.
This is a bit annoying to have per-DM locking, but it's way better (as in, up to
4 times better) for playback speed when having lots of subsurf objects,
This a small cleanup of something which I think is just a typo anyway.
With all the recent talks of harrassment and groping, I think we better avoid
that within our source code! :)
Reviewers: sergey
Reviewed By: sergey
Tags: #motion_tracking
Differential Revision: https://developer.blender.org/D2979
Technically this was introduced in 01b547f993 when
exposing size and randomness for particles.
This "fixes" makes sure particle size and size randomness is always in the
Render panel when it affects the particle system (i.e., always unless using
advanced hair or hair that is not rendering groups/objects).
The issue was caused by SpinLock implementation in old pthreads we ar eusing on
Windows. Using newer one (2.10-rc) demonstrates same exact behavior. But likely
using own atomics and memory barrier based implementation solves the issue.
A bit annoying that we need to change such a core part of Blender just to make
specific CPU happy, but it's better to have artists happy on all computers.
There is no expected downsides of this change, but it is so called "works for
me" category. Let's see how it all goes.
There was some changes about namespaces, which causes ambiguities.
Replaces using namespace with an explicit symbols we need. Is good idea to NOT
pull in the whole namespace anyway!
Now we replace O(N^2) computational complexity with O(N) extra memory penalty.
Memory is much cheaper than CPU time. Keep in mind, memory penalty is like
4 megabytes per 1M vertices.
The issue here is that we can not read scale from socket when determining
dependent area of interest. This area will depend on current pixel. Now fall
back to more stupid but reliable thing: if scale size input is connected to some
nodes, we use the whole frame as area of interest.
The issue was caused by threading conflict around looptris: it was possible
that DM will return non-NULL but non-initialized array of looptris.
Thanks Campbell for second pair of eyes!
Use more watertight and robust intersection test.
It uses now ray to triangle intersection, but it's all fine because segment was
covering the whole bounding box anyway.
Mainly when object origin is not at the geometry bounding box center.
Seems to be straightforward to fix, hopefully it doesn't break some obscure case
where this was a desired behavior.
This is fully unpredictable for artists when one damaged object makes the whole
scene to render incorrectly. This involves two main changes:
- It is not enough to check triangle bounds to be valid when building BVH.
This is because triangle might have some finite vertices and some non-finite.
- We shouldn't add non-finite triangle area to the overall area for MIS.
Best guess is that cuInit() somehow interferes with the AMD graphics driver
on Windows, and switching the initialization order to do OpenCL first seems
to solve the issue.
This reverts commit d749320e3b.
It's possible the container struct is larger,
we could do sizeof checks that falls back to memmove
but rather avoid complicating things.
Recent addition of 'reinsert' didn't match logic for ghash API.
Rename to BLI_heap_node_value_update,
also add BLI_heap_insert_or_update since it's a common operation.
Was using an edge hash for triangle -> edge lookups,
updating triangle indices for each edge-rotation.
Replace this with half-edge which can rotate edges much more simply,
writing triangles back once the solution has been calculated.
Gives ~33% speedup in own tests.
This causes render differences in some scenes, for example fishy_cat
and pabellon scenes render brighter in a few spots. This is an old
bug, not due to recent RR changes.
SVM nodes need to read all data to get the right offset for the following node.
This is quite weak, a more generic solution would be good in the future.
The loc one (shift-alt-G) was same as 'remove selected from active group'
action... Clear delta transform is not a common operation, so we can
live without a default shortcut for it.
Note that using same key (G) in same space for two completely different
kind of operations is probably a rather bad thing, nice topic for future
keymap work. ;)
Probably nice to have in 2.79a.
by the transform constraint lines
Ported over e7395c75d5 from the
greasepencil-object branch. I should've fixed this ages ago, but
couldn't figure out why at the time.
When the mesh changed topology but kept the vertex count the same, it would
result in a corrupt mesh. By checking the face & loop counts too, this has
become less likely.
I've checked IPolyMeshSchema::isConstant(), but it returns true even when
we see that the mesh changed topology.
Before this patch, the XBlur/YBlur compositor nodes would crash for me when run in a MSVC 2015 debug build (test scene: BMW27_cpu). I added the compiler instructions to explicitly align the local variables that the SSE instructions are accessing.
This was wrong since it's concenption in 28ee0f9218.
The if statement was returning true when pinid was NULL, and false otherwise.
However when scene is pinned we also want to run this code.
Code snippet by Brecht Van Lommel.
Since in Alembic the loop order seems to be reversed when exporting and
importing, and this was the only place where it was not, I was thinking
to match this to the convention of reversing the loop order as well.
Reviewers: sybren, kevindietrich
Tags: #alembic
Differential Revision: https://developer.blender.org/D2968
Two issues here:
- Checking table size to be non-zero is not a proper way to go here. This is
because we first resize the table and then fill it in. So it was possible that
non-initialized table was used.
Trickery with using temporary memory and then doing table.swap() might work,
but we can not guarantee that table size will be set after the data pointer.
- Mutex guard was useless, because every thread was using own mutex. Need to
make mutex guard static so all threads are using same mutex.
One crucial thing here: OpenVDB shoudl be compiled WITHOUT
OPENVDB_ENABLE_3_ABI_COMPATIBLE flag. This is how OpenVDB's Makefile is
configured and it's not really possible to detect this for a compiled library.
If we ever want to support that option, we need to add extra CMake argument and
use old version 3 API everywhere.
There is absolute no reason to have such an indentation level, it only causes
readability and maintainability issues. It is really simple to make code more
"streamlined".
The issue actually goes a bit deeper, converting curve to mesh will
change texture space just because font and bezier curves are using CV
to calculate texture space.
So now when those objects are converted to mesh, we disable auto
texture space and copy evaluated space over.
Previously, hitting Shift-LMB will first invoke selection operator, which
then later on is transformed to mouse tweak used for reroute operator.
This was causing problems extending selection with Shift-LMB when clicking
fast or from a tablet.
This was expected behavior for over-exposured lamps when the mode was originally
created for Tears of Steel. Turns out, there could be really bad green screen in
real production which will only have green (or rather screen) channel over
exposured.
Tweaked condition now so we use least bright channel to see if the area has
proper exposure or not.
Seems to work fine in tests, but further tweaks are possible.
Was a mistake in optimization commit which was disconnecting closures and nodes
which does not make sense for volume output.
OSL script we can't ignore and can't currently know in advance if it's a proper
volume shader or not. So we never disconnect OSL nodes from volume output.
This is a good candidate for corrective release.
The issue here was that removing datablock from main database will poke editors
update, which includes buttons context to free users of texture. Since Cycles
will free datablocks from job thread, it might crash Blender since main thread
might be in the middle of drawing.
Solved by exposing extra arguments to bpy.data.foo.remove() which indicates
whether we want to perform ID user count and interface updates. While scripts
shouldn't be using those normally, this is the only way to allow Cycles to skip
interface update when removing datablock.
Reviewers: mont29
Reviewed By: mont29
Differential Revision: https://developer.blender.org/D2840
Was caused by numeric overflow when calculating preview dimensions.
Now we try to avoid really insance preview resolutions by fitting
aspect into square.
The issue was caused by operator redo which frees all object's evaluated data,
including bounding box. This bounding box can not be reconstructed properly
without full curve evaluation (need to at least convert font to nurbs, which is
not cheap already).
In fact, any type of baking might have caused holes in mesh.
The issue was caused by zspan_scanconvert() attempting to get order of traversal
'a-priori', which might have failed if check happens at the "tip" of span where
`zspan->span1[sn1] == zspan->span2[sn1]`.
Didn't see anything bad on making it a check when iterating over scanlines and
pick minimal span based on current scanline. It's slower, but unlikely to cause
measurable difference. Quality should stay the same unless i'm missing something.
Reviewers: brecht, dfelinto
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D2837
Fix T52977: Parent bone name disappeared in the UI in pose mode.
Regression caused by own rBc57636f060018. So instead of changing widget
type, just flag it as disabled.
Note that core of the issue is elsewhere though - there is absolutely no
reasons to have a search widget for pointers we cannot change nor
search! But fixing this is not really top priority, one of the many
glitches of our UI code, so think we can live with current code.
To be backported to 2.79a.
Inner DAG code would not check against NULL pointer, and in case of an
active linked scene, scene pointer will be NULL here, so we have to
check it ourself. ;)
No need to print status for basic & reliable operations,
build systems can output operations they run if needed,
or debug output changed in the source if developers are debugging.
Nice for ninja, so any printed text hints at a problem to fix.
Was using cursor position from within menu,
clicking on the same position for every selected item (toggling).
Now operate on each selected outliner element, without toggling.
When 2x loops have different number of vertices,
the distribution for vertices fan-fill depended on the loop order
and was often lop-sided.
This caused noticeable inconstancies depending on the input
since edge-loops are flipped to match each others winding order.
Pixel size was not initial early enough. For first time this was not a problem
because the bevel amount starts at 0 then, and after the mouse moves the pixel
size is initialized. For the second time the bevel amount starts at a non-zero
value, and it failed then.
Use triple buffer by default now on all platforms, remaing ones where:
* Mesa: seems to have been working well for a long time now, and not using
it gives issues with the latest Mesa 17.2.0.
* Windows software OpenGL: no longer supported since OpenGL 2.1 requirement
was introduced.
* OS X with thousands of colors: this option was removed in OS X 10.6, and
that's our minimum requirement.
Camera clipping was left to default values, which won't work well for
very large (or small) objects. Now recompute valid clipping start/end
based on boundingbox of rendered data, and final location of camera.
Tentative fix, since I cannot reproduce thenissue for some reason here
on linux.
Core of the problem is pretty clear though, thanks to Germano Cavalcante
(@mano-wii): another thread could try to use looptris data after worker
one had allocated it, but before it had actually computed looptris.
So now, we use a temp 'wip' pointer to store looptris being computed
(since this is protected by a mutex, other threads will have to wait on
it, no possibility for them to double-compute the looptris here).
This should probably be backported to 2.79a if done.
Should have been done before ahoy, sorry about that. Means 2.79 tag will
be one (no functionnal changes) commit ahead from our 2.79 builds, think
we can live with that.
Previous fix wasn't working correct for certain compiler and CPU intrinsics
mode, causing quite some crashes.
This should be a safer fix, which is closer in behavior to previous release
but which should still fix issues with robust curve intersection.
Note: this commit seems to work as expected (also with transform
snapping etc.). However, it is rather unsafe - not enough for 2.79 at
least, unless we get much more testing on it. It also depends on three
previous ones.
Note that using a global lock here is far from ideal, we should rather
have a lock per DM, but that will do for now, whole DM thing is doomed
to oblivion anyway in 2.8.
Also, we may need a `DM_DIRTY_LOOPTRIS` dirty flag at some point. Looks
like we can survive without it for now though... Probably because cached
looptris are never copied accross DM's?
This was... horribly wrong, CDDM will often *not* need to allocate
anything to return arrays of mesh items! Just check whether array
pointer is NULL.
Also, remove `DM_get_looptri_array`, that one is useless currently,
`dm->getLoopTriArray` will always return cached array (computing it if
needed).
Only the camera from View3D.localvd is used,
other pointers may be invalid.
Longer term we should probably clear these to ensure no accidents.
For now just follow the rest of Blender's code and don't access.
Baking rigid body cache was broken if some cached frames already
existed.
This is just a band aid for release, the logic need to be looked into
further.
scripts being "Artwork" which is your sole property and free to license.
I've removed the reference to scripts in this text.
This was from 2002! With our Python scripts becoming part of how Blender runs,
such scripts now are officially required to be compliant with GNU GPL.
For more information; check the FAQ or consult foundation@blender.orghttps://www.blender.org/support/faq/
We have a hardcored limit of 1000 images to be baked.
However anything anove 100 would be leading to overflow in the code.
Caught by warning from builder bot (my compiler doesn't even complain
about this, but it should).
Apparently with Maya in a certain configuration, it's possible to have an
Alembic object without schema in the Alembic file. This is now handled
properly, instead of crashing on a null pointer.
For example, if you have two keyframes:
k1 = 1px, k2 = 10px
it was doing:
1px, 9px, 8px, ..., 3px, 2px, 10px
instead of:
1px, 2px, 3px, ..., 8px, 9px, 10px
- Vertex only meshes never restored their selection history.
- Select history was cleared on the source instead of the target.
Simple Optimizations:
- Avoid O(n^2) linked list looping that checked the entire list before
adding elements (NULL values in the source array to prevent dupes).
- Re-use vert & edge lookup tables instead of allocating new ones.
For some specific pipelines (e.g., holographic rendering) you can easily
need over a million frames (1k * 1k view angles).
It seems a corner case, but there is no real reason not to allow users
doing that.
That said we do loose subframe precision in the highest frame range. Which can
affect motionblur. The current maximum sub-frame precision we have is 16.
While the previous limit of 500k frames has a precision of 32.
Thanks to Campbell Barton for the help here.
To be backported to 2.79
Issue was nasty hidden one, the dual status (mix of local and linked)
of proxies striking again.
Here, remapping process was considering obdata pointer of proxies as
indirect usage, hence clearing the 'LIB_TAG_EXTERN' of obdata pointer.
That would make savetoblend code not store any 'lib placeholder' for
obdata data-block, which was hence lost on next file read.
Another (probably better) solution here would be to actually consider
obdata of proxies are fully indirect usage, and simply reassign proxies
from their linked object's obdata on file read...
However, that change shall be safer for now, probably good for 2.79 too.
Would be nice to be able to catch this with assert as well, will see what would
be the best way to do this/.\
Need to verify with Mai that this solves crash for her and maybe consider
porting this to 2.79.
We need to make sure we can store all volume closures for all objects in volume
stack. This is a bit tricky to detect what would be the "nestness" level of
volumes so for now use maximum possible stack depth. Might cause some slowdown,
but better to give reliable render output than to fail quickly.
Should be safe for 2.79 after extra eyes.
We were showing "search for unknown menutype WM_MT_button_context" messages in terminal which were not helpful for users, so now they are disabled.
To be backported to 2.79
It is possible to have same image used multiple times at different frames,
which means we can not free it's buffers without any guard. From quick tests
this seems to be doing what it is supposed to.
Need more testing and port this to 2.79.
Regression from rBfed853ea78221, calling this inside thread worker was
not really good idea anyway, and we already have all the code we need in
pre-threading init function, was just disabled for vertex particles
before.
To be backported to 2.79.
The documentation for the bpy.app.handlers.scene_update_{pre,post}
handlers states that they're called "on updating the scenes data".
However, they're called even when the data hasn't changed. Of course
such handlers are useful, but the documentation should reflect the
current behaviour.
Reviewers: mont29, sergey
Subscribers: Blendify
Maniphest Tasks: T46329
Differential Revision: https://developer.blender.org/D1535
Commit b6d7cdd3ce changed how the mesh data
is deformed, which wasn't taken into account yet in this unit test.
Instead of directly reading the mesh vertices (which aren't animated any
more), we convert the modified mesh to a new one, and inspect those
vertices instead.
When a mesh changes its number of vertices during the animation,
Blender rebuilds the DerivedMesh, after which the materials weren't
applied any more (causing the default to the first material slot).
This can happen with Alembic files exported from Maya. I'm unsure as to the
root cause, but at least this fixes the crash itself.
Thanks to @looch for reporting this with a test file. The test file has to
remain confidential, though, so it's on my workstation only.
Simply disabled python tests, they can't be run anyway (since blender target is
not enabled) and we don't have any player-related tests in that folder.
Creating ngons with multiple axis aligned shapes in the middle of a
single face would fail in some cases.
This exposed multiple problems in BM_face_split_edgenet_connect_islands
- Islands needed to be sorted on Y axis when X was aligned.
- Checking edge intersections needed increased endpoint bias.
- BVH epsilon needed to be increased.
Basically, make re-alloc and memcpy from the same lock, otherwise one
thread might be re-allocating thread while another one is trying to
copy data there.
Reported by Mohamed Sakr in IRC, thanks!
The tweakmode flag and the selected-channels flag accidentally
used the same value, due to confusion over where these flags were
supposed to be set. The selected-channels flag has now been moved
to use a different value, so that there shouldn't be any further
conflicts.
To be ported to 2.79.
Own previous fix (rBd5d626df236b) was not valid, curves are actually
supported by SoftBodies. It was rather a mere UI bug, which was not
including Surfaces and Font obect types in those valid for softbody UI.
Thanks to @brecht for the head up!
Also, fix safe for 2.79, btw.
Adds thin/default/thick modes to add -1/0/1 to the auto detected line width,
while leaving the overall UI scale unchanged.
Also tweaks the default line width threshold, so thicker lines start from
slightly high UI scales.
Differential Revision: https://developer.blender.org/D2778
*Changed categories of some keywords
*reordered some longer keywords that didn't appear
*Activated another color (reserved builtins) by Leonid
*added some HGPOV and UberPOV missing keywords
Patch by Maurice Raybaud (@mauriceraybaud). Thanks to Leonid for additions, feedback and Linux testing.
Related diffs: D2754 and D2755.
While not a regression, this is new feature and would be nice to have it
backported to final 2.79.
As part of the fix for T51587, I removed the Depth output for non-Multilayer
images since it seemed weird that PNGs etc. that don't have a Z pass still get
a socket for it.
However, I forgot about non-multilayer EXRs, which are a special case that can
actually have a Z pass.
Therefore, this commit brings back the Depth output for non-multilayer images
just like it was in 2.78.
New code was handling correctly ID's internal references to self, but
not references between 'made real' different objects...
Regression, to be backported in 2.79.
@@ -69,4 +67,3 @@ base class --- :class:`CPropValue`
..warning::
The id can't be stored as an integer in game object properties, as those only have a limited range that the id may not be contained in. Instead an id can be stored as a string game property and converted back to an integer for use in from_id lookups.
# Layout of the device tuples: (Name, Type, Persistent ID)
device_list=_cycles.available_devices()
# Make sure device entries are up to date and not referenced before
# we know we don't add new devices. This way we guarantee to not
# hold pointers to a resized array.
self.update_device_entries(device_list)
# Sort entries into lists
cuda_devices=[]
opencl_devices=[]
fordeviceindevice_list:
entry=self.find_existing_device_entry(device)
ifentry.type=='CUDA':
cuda_devices.append(entry)
elifentry.type=='OPENCL':
opencl_devices.append(entry)
returncuda_devices,opencl_devices
returncuda_devices,opencl_devices
defget_num_gpu_devices(self):
import_cycles
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.
