Took quite some effort to get this working with undo, and not everything is working yet. Open todos:
* Fix un-indent undo (pretty easy fix, just to lazy now ;) )
* Fix redo (didn't check details yet)
* Fix un-comment undo for line with nothing but a comment char
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
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.
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.
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)?
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
`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!
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)
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:
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...
(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)),
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))
>>> 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 ", ".
@@ -322,8 +422,11 @@ template<typename T, int _Cols, int _Options> class DenseStorage<T, Dynamic, Dyn
}
}
m_rows=nbRows;
m_rows=nbRows;
}
}
inlineconstT*data()const{returnm_data;}
constT*data()const{returnm_data;}
inlineT*data(){returnm_data;}
T*data(){returnm_data;}
private:
DenseStorage(constDenseStorage&);
DenseStorage&operator=(constDenseStorage&);
};
};
}// end namespace Eigen
}// end namespace Eigen
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.
