This adds an initial (and already working), generic API for defining, handling and drawing of table UIs.
It supports two types of tables:
* Vertical flow: Rows are all in a vertical stack.
* Horizontal flow: Rows are stacked vertically, until some threshold height is reached. The table is then split into another vertical stack that is drawn next to the former one. Such drawing could be used for the file browser in non-thumbnail mode.
A table is built out of - guess what - rows and columns. The API allows defining new collumns (identified by an id-name) and inserting new rows at any point. When drawing, we draw a cell for each column/row combination, using a custom cell-drawing callback.
The uiTable API will probably get some support for drawing buttons (uiBut) into custom layouts (uiLayout).
The API is written with big data sets in mind. In future we may get a spreadsheet view where data like all vertices of a mesh is displayed. That would result in thousands of rows, the uiTable API should be ready for this.
Pending:
* UI template for those settings (showing USE)
* Depsgraph evaluation of them (to flush into objects)
* RNA to see if a settings is being used
There is still a warning because of `DST.context = C;` which discards
'const' qualifier. I find this a legit problem, I suspect we are not
suppose to store bContext at all.
This is extracted from the layer-manager branch. With the following
changes:
* Renamed references of layer manager to collections manager
* I didn't include the editors/space_collections/ draw and util files.
I still need to bring the drawing code here, so we see something.
Note: when in edit mode this depsgraph update is not being called. We are using DerivedMesh in those cases, so it is fine. I would like to investigate this though
The idea was to link something to a parent, but the point is:
we must not pass owner deep and then have any parent-type-related
logic implemented in the "children".
This is much more flexible solution which will allow doing some
more procedural features.
Reviewers: brecht, dfelinto, mont29
Reviewed By: mont29
Subscribers: Severin
Differential Revision: https://developer.blender.org/D2403
The freestyle data was never freed when removing a renderlayer.
```
blender -b --factory-startup --debug-memory --python-expr "import bpy;bpy.ops.scene.render_layer_add();bpy.context.scene.render.layers.active_index=0;bpy.ops.scene.render_layer_remove()"
```
Currently the tests don't run on windows for the following reasons
1) render_graph_finalize has an linking issue due missing a bunch of libraries (not sure why this is not an issue for linux)
2) This one is more interesting, in test/python/cmakelists.txt ${TEST_BLENDER_EXE_BARE} and ${TEST_BLENDER_EXE} are flat out wrong, but for some reason this doesn't matter for most tests, cause ctest will actually go out and look for the executable and fix the path for you *BUT* only for the command, if you use them in any of the parameters it'll happily pass on the wrong path.
3) on linux you can just run a .py file, windows is not as awesome and needs to be told to run it with pyton.
4) had to use the NAME/COMMAND long form of add_test otherwise $<TARGET_FILE:blender> doesn't get expanded, why? beats me.
5) missing idiff.exe for msvc2015/x64 in the libs folder.
This patch addresses 1-4 , but given I have no working Linux build environment, I'm unsure if it'll break anything there
5 has been fixed in rBL61751
Reviewers: juicyfruit, brecht, sergey
Reviewed By: sergey
Subscribers: Blendify
Tags: #cycles, #automated_testing
Differential Revision: https://developer.blender.org/D2367
This includes a few fixes in the MBC_ api.
The idea here is for this to be the only interface the render engines
will deal with for the meshes.
If we need to expose special options for sculpting engine we refactor
this accordingly. But for now we are shaping this in a per-case base.
Note:
* We still need to hook up to the depsgraph to force clear/update of
batch_cache when mesh changes
(I'm waiting for Sergey Sharybin's depsgraph update for this though)
* Also ideally we could/should use BMesh directly instead of
DerivedMesh, but this will do for now.
Note 2:
In the end I renamed the `BKE_mesh_render` functions to `static
mesh_render`. We can re-expose them as BKE_* later once we need it.
Reviewers: merwin
Subscribers: fclem
Differential Revision: https://developer.blender.org/D2476
Blenders baking system currently doesn't support the topology used by
adaptive subdivision and primitive ids will be wrong or out of range
leading to crashes. Updating the baking system to support other
topologies would be a bit involved, so for now we simply disable
subdivision while baking to avoid crashes.
We started to run out of bits there, so now we separate flags
which came from __object_flags and which are either runtime or
coming from __shader_flags.
Rule now is: SD_OBJECT_* flags are to be tested against new
object_flags field of ShaderData, all the rest flags are to
be tested against flags field of ShaderData.
There should be no user-visible changes, and time difference
should be minimal. In fact, from tests here can only see hardly
measurable difference and sometimes the new code is somewhat
faster (all within a noise floor, so hard to tell for sure).
Reviewers: brecht, dingto, juicyfruit, lukasstockner97, maiself
Differential Revision: https://developer.blender.org/D2428
Cycles add-on did not actually support reloading correctly.
When you want to correctly reload sub-modules (i.e. modules of an add-on
which is a package), you need to use importlib, a mere import will do
nothing with already loaded modules (RNA classes are sort of
pre-registered when they are evaluated, through the meta-class system).
New options to define the style of the animation paths in order to get
better visibility in complex scenes.
Now is possible define the color, thickness and several options relative
to the style of the lines used to draw motion path.
This way we can stop traversing BVH node early on.
Gives about 2-2.5x times render time improvement with 3 BVH steps.
Hopefully this gives no measurable performance loss for scenes with
single BVH step.
Traversal is currently only implemented for QBVH, meaning old CPUs
and GPU do not benefit from this change.
Similar to the previous commit, the statistics goes as:
BVH Steps Render time (sec) Memory usage (MB)
0 46 260
1 27 373
2 18 598
3 15 826
Scene used for the tests is the agent's body from one of the barber
shop scenes (no textures or anything, just a diffuse material).
Once again this is limited to regular (non-spatial split) BVH,
Support of spatial split to this feature will come later.
The idea is to create several smaller BVH nodes for each of the motion
curve primitives. This acts as a forced spatial split for the single
primitive.
This gives up render time speedup of motion blurred hair in the cost
of extra memory usage. The numbers goes as:
BVH Steps Render time (sec) Memory usage (MB)
0 258 191
1 123 278
2 69 453
3 43 627
Scene used for the tests is the agent's hair from one of the barber
shop scenes.
Currently it's only limited to scenes without spatial split enabled,
since the spatial split builder requires some changes to work properly
with motion steps coordinates.
Also fixed some issues with motion keys calculation:
- Clamp lower and upper limits of curves so we can safely call those
functions for the very first and very last curve segment.
- Fixed wrong indexing for the curve radius array.
- Fixed wrong motion attribute offset calculation.
Mimics how regular triangles are working and makes it more clear where
the stuff is located in the kernel.
Needed to have some forward declarations because of the current placement
of things in the kernel.
Following @AlonDan's feature request and @hjalti's screenshot yesterday,
I've decided to implement support for this to make it easier to scan which
keyframes correspond with which set of controls, especially when faced with
a large wall of keyframes.
In retrospect, I should've done this a long time ago!
Was a bit confusing to have transparent and translucent depth
exposed but no diffuse or glossy.
Reviewers: brecht
Subscribers: eyecandy
Differential Revision: https://developer.blender.org/D2399
This is important for the reliable behavior or isnan/isfinite/min/max
functions to work with nan and non-finite values. Some of the issues
with fast math are possible to work around, but didn't find a way to
have reliable min/max implementation yet.
Please NEVER EVER use such a statement, it's only causing HUGE
issues. What is even worse: it's not always possible to immediately
see that the hell is coming from such a statement.
There is still some statements in the existing code, will leave
those for a later cleanup.
- flushing hidden state ran when it didn't need to.
- flushing checks didn't early exit when first visible element found.
- low level BM_*_hide API calls like this can use skip iterators
can loop over struct members directly.
No user-visible changes.
Took pieces from D2316 and D2359, changed a few more things.
- use new immediate mode
- use new matrix stack
- remove state push/pop
Part of T49043 and T49450
- face-create-extend option could add hidden verts and edges into
the selection history (invalid state).
- faces could be created that included existing hidden edges
that remained hidden (invalid state too).
- newly created faces could copy hidden flag from surrounding faces,
giving very confusing results (looks as if face creation failed).
Surprising nobody noticed these years old bugs!
Experimental option for the Reproject Strokes operator to project strokes on to
geometry, instead of only doing this in a planar (i.e. parallel to viewplane) way.
The current implementation is quite rough, and may need to be improved before it
is really ready for use. Potential issues:
* Loss of precision (i.e. stairstepping artifacts) from the 3D -> 2D -> 3D conversion
as we don't have float version of one of the projection funcs
* Jagged depth if there are gaps, since it will default back to the 3d-cursor plane
if no geometry was found (instead of doing some fancy interpolation scheme)
* I'm not sure if it's that useful for adapting GP strokes to deforming geometry yet...
Now the eraser checks if there's an active frame with some strokes in it
before creating a new frame. There's no point in creating a new frame if
there are no strokes in the active frame (if one exists).
This still doesn't help much if there were strokes but they weren't touched though...
This operator adds a new frame with nothing in it on the current frame.
If there is already a frame there, all existing frames are shifted one frame later.
Quite often when animating, you may want a quick way to get a blank frame,
ready to start drawing something new. Or maybe you just need a quick way to
add a "placeholder" frame so that a suddenly-appearing element does not show
up before its time.
If the layers or the colors were renamed, the animation data was wrong
because the data path was not updated.
I also have fixed a possible stroke color name update if the name was duplicated moving
the rename function call after checking unique name.
Avoids possible jumps when one is trying to do some really preciese tweak.
Quite striaghtforward change for mouse input initialization: take Shift
state into account. However, this will interfere with the axis exclusion
which is currently also uses Shift (the feature to move something in a
plane which doesn't have selected axis). This is probably not so commonly
used feature (nobody in the studio even knew of it) and the only downside
now would be that such a constrainted movement will become accurate by
default. That's easy to deal from user side by just unholding Shift key.
Reviewers: brecht, mont29, Severin
Differential Revision: https://developer.blender.org/D2418
Util function to check if a SceneCollection is linked to a SceneLayer
This is needed for corner cases of bpy.context.scene_collection when the context render_layer mismatches the context scene_collection.
To make it faster to try different interpolation curves, there's a new operator
"Remove Breakdowns" which will delete all breakdowns sandwiched by normal
keyframes (i.e. all the ones that the previous run of the Interpolation op created)
This commit introduces the ability to use the Robert Penner easing equations
or a Custom Curve to control the way that the "Interpolate Sequence" operator
interpolates between keyframes. Previously, it was only possible to get linear
interpolation between the gp frames.
Workflow:
1) Place current frame between a pair of GP keyframes
2) Open the "Interpolate" panel in the Toolshelf
3) Choose the interpolation type (under "Sequence Options")
4) Adjust settings (e.g. if you're using "Custom Curve", use the curvemap widget
to define the way that the interpolation proceeds)
5) Click "Sequence" to interpolate
6) Play back/scrub the animation to see if you've got the result you want
7) If you need to make some tweaks, undo, or delete the generated keyframes,
then repeat the process again from step 4 until you've got the desired result.
The "gp_sculpt" settings should be strictly for stroke sculpting, and not abused by
other tools. (Similarly, if other general GP tools need one-off options, those should
go into the normal toolsettings->gpencil_flag)
Furthermore, this paves the way for introducing new settings for controlling the way
that GP interpolation takes place (e.g. with easing equations, or a custom curvemap)
* Reshuffled some blocks of code for better ease of navigation/flow in the file
* Improved some tooltips
* Removed "Helper" tag from some functions that serve bigger roles
* Fixed some errant formatting
The interpolation operators (and their associated code) occupied a significant
portion of gpencil_edit.c (which was getting a bit heavy). So, it's best to split
these out into a separate file to make things easier to handle, in preparation
for some further dev work.
After revert the commit rB4b99958ca12642, the line added at the end of the enum is not necessary anymore because it is replaced by the corresponding element in the list in the right position.
Things like `BLI_uniquename` had nothing, but really nothing to do in
BLI_path_util files!
Also, got rid of length limitation in `BLI_uniquename_cb`, we can use
alloca here to avoid overhead of malloc while keeping free size (within
reasonable limits of course).
Just store bones that could not get renamed to desired flipped name on the
first try into a temp list, and try to rename them a second time.
This is rather simple solution, will induce 'over numbering' in case you
flip a bone to another unselected bone's name (since number will be
incremented in both rename attempts), but think this is acceptable minor
glitch, for a corner case situation that does not have any good
resolution anyway.
Also, set `strip_numbers` option of `BKE_deform_flip_side_name` to
false, otherwise chains of bones with same names would get their numbers
completely messed up after name flipping.
Based on work by @dfelinto in D2456 (https://developer.blender.org/D2456), thanks.
This is part of T49043
fixed up some color/rect calls
fixed up ANIM_channel_draw()
Reviewers: krash, merwin
Reviewed By: merwin
Tags: #bf_blender_2.8
Maniphest Tasks: T49043
Differential Revision: https://developer.blender.org/D2377
Had to add a few utility functions to replace existing functions. Let me know if these are duplicates.
Reviewers: merwin
Reviewed By: merwin
Tags: #bf_blender_2.8
Maniphest Tasks: T49043
Differential Revision: https://developer.blender.org/D2434
This adds two functions to project 3d coordinates onto a 3d plane,
to get 2d coordinates, essentially eliminating the plane's normal axis
from the coordinates.
Reviewed By: mont29
Differential Revision: https://developer.blender.org/D2460
Also adding a new flag value for ObjectBase to store visibility.
I still need this to be synced, but the idea is to centralize the
logic of tree evluation, and keep the visibility cached.
When we make a collection invisible or unselectable, selected object have to be re-evaluated
Same goes for when we add a new object, its base visibility has to be
refreshed.
TODO also add a call for this when we remove a scene collection, or
unlink a scene layer.
It is quite likely in a triangulated mesh that the actual island edge
belongs to a different triangle than the current pixel; for example
consider corners of a triangulated axis aligned rectangle face that
have the additional edge: a pixel there will have to be assigned to
one of the triangles, but one of the edges of the original rectangle
can only be accessed through the other triangle.
Thus for robust operation it is necessary to do a recursive search.
The search is limited by requiring that it only goes through edges
that bring it closer to the target point, and also by depth as a
safeguard.
Differential Revision: https://developer.blender.org/D2409
The code requires the pixel on the other side of the seam to be assigned
precisely to the expected triangle. This can cause false negatives around
vertices, where a pixel is likely to touch multiple triangles and thus
cannot be said to unambiguously belong to any one of them, so check
distance to the intended triangle and accept the result if it's close.
1. Forcibly symmetrize the neighbor relations, so that if A is neighbor
of B, B is neighbor of A. The existing code is guaranteed to violate
this if texture resolution is different between the sides of a seam.
2. In texture mode dynamic paint adds a 1 pixel wide border around the
islands. These pixels aren't really part of the dynamic paint domain
and thus by design can't have symmetrical neighbor relations. This
means they can't be treated by effects like normal pixels.
The simplest way to handle it in a consistent way is to exclude
them from effects, but add an additional pass that recomputes them
as average of their non-border neighbors, located on both sides of
the seam.
This avoids intersection AABB of different curve primitives
which makes it less ray-to-primitive intersections.
This gives about 30% speedup of hair rendering in the barber
shop scenes here. There is still some work to be done on those
files to solve major speed issues on certain frames.
This officially makes the viewport not draw anything, until we get the
new ObjectBase * to use.
This is easily revertable, but for now I prefer to make sure this is not
in the way of refactoring.
This way we can have different limits for regular and motion curves
which we'll do in one of the upcoming commits in order to gain some
percents of speedup.
The reasoning here is that motion curves are usually intersecting
lots of others bounding boxes, which makes it inefficient to have
single primitive in the leaf node.
Maximal number of elements is supposed to be inclusive. That is what
it was always meant in this file and what @brecht considered still
the case in 6974b69c61.
In fact, the commit message to that change mentions that we allowed
up to 2 curve primitives per leaf while in fact it was doing up to 1
curve primitive.
Making it real 2 primitives at a max gives about 5% slowdown for the
koro.blend scene. This is a reason why BVHParams.max_curve_leaf_size
was changed to 1 by this change.
Since the beginning of times hair settings in cycles were global for
the whole scene but were located in the particle context. This causes
quite some trickery to get shots set up for the movies here in the
studio by forcing artists to create dummy particle system to change
settings of hair on the shot.
While ideally this settings should be properly become per-particle
system for the time being it will save sweat and blood to move the
settings to scene context.
Reviewers: brecht
Subscribers: jtheninja, eyecandy, venomgfx, Blendify
Differential Revision: https://developer.blender.org/D2287
Made them closer to how GTest shows the output, so reading test logs
is easier now (at least feels more uniform).
Additionally now we know how much time tests are taking so can tweak
samples/resolution to reduce render time of slow tests.
It is now also possible to enable colored messages using magic
CYCLESTEST_COLOR environment variable. This makes it even easier to
visually grep failed/passed tests using `ctest -R cycles -V`.
Reusing PROP_TEXTEDIT_UPDATE instead of adding a new property flag just for search strings. Currently it's only used for search strings anyway so seems fine for now.
Fixes T50336.
This splits `interp_weights_face_v3` into `interp_weights_tri_v3` and
`interp_weights_quad_v3`, in order to properly handle three sided polygons
without needing a useless extra index in your weight array. This also
improves clarity and consistency with other math_geom functions, thus
reducing potential future errors.
Reviewed By: mont29
Differential Revision: https://developer.blender.org/D2461
The issue was that we used to compare number of vertices for mesh after the auto
smooth was applied (at the center of the shutter time) with number of vertices
prior to the auto smooth applied. This caused false-positive consideration of a
mesh as changing topology.
Now we do autosplit as early as possible and do it from blender side, so Cycles
does not need to re-implement splitting on it's side.
This way render engine can request mesh to be auto-split and not
worry about implementing this functionality on it's own.
Please note that this split is to be performed prior to tessellation.
Other than implementing a `mid_v3_v3_array` function, this removes
`cent_tri_v3` and `cent_quad_v3` in favor of `mid_v3_v3v3v3` and
`mid_v3_v3v3v3v3` respectively.
Reviewed By: mont29
Differential Revision: https://developer.blender.org/D2459
When layout has only small buttons (buttons with icon and without label)
its size should be fixed. Code was modified to be able to add a new UI_ITEM_MIN
flag which indicates that the layout has only small fixed-width buttons.
Patch by @raa, with minor style edits by @mont29.
Reviewers: Severin, mont29
Reviewed By: mont29
Tags: #bf_blender, #user_interface
Differential Revision: https://developer.blender.org/D2423
This does not address stapling shader in 2.8, though the solution can be
similar (own shader, not polutting interlace shader).
part of T49043
Reviewers: merwin
Differential Revision: https://developer.blender.org/D2440
Am pretty sure node update should not touch to Main database like that,
but for now let's allow it, I guess the hack is needed for things like
Sverchok. ;)
-Remove NPOT check as it should be supported by default with OGL 3.3
-All custom texture creation follow the same path now
-Now explicit texture format is required when creating a custom texture (Non RGBA8)
-Support for arrays of textures
If the active object is in weight paint mode, but some armatures in pose mode, 'manipulate center points' still affects the transformation. See bd2034a749.
Also removed redundant check, we basically did the same check for paint modes twice.
If a very low wetness absolute alpha brush is used with spread and
drying effects enabled, some pixels will rapidly accumulate paint.
This happens because paint drying code applies a minimal wetness
threshold that causes the paint to instantly dry out.
Specifically, every frame the brush adds paint at the specified
absolute alpha and wetness set to the minimal threshold, spread
drops it below threshold, and finally drying moves all paint to
the dry layer. This drastically accelerates the rate of flow of
paint into the affected pixels.
Fortunately, the reason paint spread actually ends up decreasing
wetness turns out to be a simple floating point precision problem,
which can be easily fixed by restructuring the affected expression.
Reported on IRC by dfelinto, thanks.
Root of the issue was that opening a new text file would create
datablock with one user, when Text editor is actually a 'user one' user.
This was leaving Text datablocks in inconsitent user count, and
generating asserts in BKE_library area.
Also changed a weird piece of code related to that extra user thing in
main remapping func.
- Calculate normals with dfdy while waiting for a proper way to draw mesh normals.
- Added initial support for 2d texture arrays.
- Better API naming.
- Generate Matcap texture arrays and draw with it.
Main issue here was that in old usercount system 'user_real' did simply
not allow that kind of thing to work. With new pait of 'USER_EXTRA'
tags, it becomes possible to handle the case correctly, by merely refining
checks about indirectly use objects whene removing them from a scene.
Incidently, found another related bug, 'link group objects to scene' was not
incrementing objects' usercount - bad, very very bad!
The settings.frame_start rna was clamping frame start to frame end when frame start was bigger than frame end.
The fix is simply to set frame end first
- Added temporary draw_mesh function to render edit mesh
- DRW_draw_batch_list allows to render a list of objects with optimal state change
- All viewport rendering is done offscreen for the moment
Signed-off-by: Clément Foucault <foucault.clem@gmail.com>
This is a hacky fix for a regression introduced sometime after 2.76.
The "Strip Time" setting on NLA Strips could not be edited without the
value immediately jumping back to the current FCurve value (or 0.0 if no
keyframes existed); even enabling autokey wouldn't let you key the property.
Until we have proper overrides (that only lose their values on frame change),
it's best that this setting is editable, even if it does mean it you have to
manually change the frame to see the updated values.
Sometimes it can be useful to be able to keep onion skins visible in the
OpenGL renders and/or when doing animation playback. In particular, there
are two use cases where this is quite useful:
1) For creating a cheap motion-blur effect, especially when the before/after
values are also animated.
2) If you've animated a shot with onion skinning enabled, the poses may end
up looking odd if the ghosts are not shown (as you may have been accounting
for the ghosts when making the compositions).
This option can be found as the small "camera" toggle between the "Use Onion Skinning"
and "Use Custom Colors" options.
This is a regression introduced in rB5bd9e832
It looks more like a hack than a proper fix, but the shader logic
changed a lot for blender2.8, so I would rather do the elegant fix
there, while leaving master working.
If we ever do a 2.78b (or 2.79) this should get in.
That code was a joke, letting some invalid utf8 bytes pass, returning
wrong offset for some invalid sequences, not to mention length and
pointer easily going out of sync, NULL final byte being 'forgotten' by
memcpy, etc. etc.
The miracle here is that we could survive using this for so long!
Probably because we do not use utf-8 sanitizing enough in Blender,
actually... :/
This test should ensure we correctly detect all invalid utf-8 sequences in a given string.
DISCLAIMER:
Do not run this with current code - you'll either laugh or cry, nearly *all* checks fail!
Based on utf-8 decoder stress-test (https://www.cl.cam.ac.uk/~mgk25/ucs/examples/UTF-8-test.txt)
by Markus Kuhn <http://www.cl.cam.ac.uk/~mgk25/> - 2015-08-28 - CC BY 4.0
Please **DO NOT** add changes from master when it's totally uneeded!
Changes to BLI_ area most certainly shall *always* be done in master,
there is absolutely no point in adding more diff between the two
branches than needed, will only makes merging more cumbersome!
Conflicts:
CMakeLists.txt
source/blender/blenlib/intern/math_vector_inline.c
This is the same issue as was fixed with T39486: the adjustment pass
that tries to equalize different widths at either end of an edge
sometimes causes the widths to get bigger and bigger.
The previous fix was to let "clamp_overlap" do double duty as a way
to limit this behavior. But clearly this is undiscoverable, as the
current bug report shows. So I put in an "auto-limiting" mode that
detects when adjustments are going crazy and then acts as if
clamp_overlap were set.
The reason we can't always act as if clamp_overlap is set is that
certain models (e.g., Bent_test in regression tests) look bad if
that is enabled.
This reverts commit 5aa19be912 and b4a721af69.
Due to postponement of particle system rewrite it was decided to put particle code
back into the 2.8 branch for the time being.
Include idea that Blender may fail to launch it even if path is correct,
in some cases (dear Windows...).
Based on idea from @lijenstina and @blendify (D2349), thanks.
Over time roll and orbit would scale the quaternion
which is documented as unit length.
In practice any errors would be subtle,
but better normalize as other operators do.
Basic idea is to store fileversion in Library datablock, and split again
Main by libraries after lib linking, do_versions_after_liblink on
those separated Mains, and merge again.
This allows to still have correct versions for each data-block in that
second do_versions step.
Note that this is not used currently in master (might be soon, though),
but is needed for 2.8 work.
Main scheduler would be created way before `-t` argument would be
parsed, since it was on forth pass! Moved it to first pass of argparse,
that kind of stuff should be initialized asap on startup.
otherwise I could not get different iterators based on a flag (SELECT), which is used everywhere in object_relations.c
The alternative would be to split every function in object_relations.c into _all, and _selected
When linking data-blocks from same library in several steps, the already
linked data-blocks of same lib would go again through versionning code...
Note: only fixed for libraries, I can't imagine how this could happen
with local data...
This is not the ideal iterator (it loops over the scene collection tree 3x).
One solution (I want to discuss with Bastien Montagne @mont29) is whether to store the *parent of a SceneCollection to help with that. That would speed things up, and cost less memory.
We do not even need to store it in the file, since it can be re-generated at read time
Data transfer was not checking if the required geometry existed, thus
causing a segfault when it didn't. This adds the required checks, and
reports errors if geometry is missing.
This also replaces instances of the words "polygon" and "loop" in error
messages with "face" and "corner" respectively, to be consistent with
the rest of the existing UI.
Reviewed By: mont29
Differential Revision: http://developer.blender.org/D2410
When append a datablock the default brushes were not created and only
were created when draw new strokes. Now the default brushes are created
when draw strokes if necessary.
It's now possible to change the shortcut that enables planar transformation with the transform manipulators (shift+LMB on axis).
This actually fixes the workaround added in rB20681f49801fd. Thing is that we needed to allow using the manipulators, even if a modifier key is held so things like snapping work right away. That's why normal LMB behavior uses KM_ANY. However, event handling would always execute the KM_ANY keymap handler because it's iterated over first. Simply solved this by registering the KM_SHIFT keymap item first, so it has priority over the KM_ANY one.
Enum properties with icon only flag should use minimum/fixed width in expanded layouts (alignment=UI_LAYOUT_ALIGN_EXPAND).
Differential Revision: https://developer.blender.org/D2415 by @raa (only made some really minor corrections)
Shuffle existing code, hook it up to the new (& old) viewport.
Also the 3D mouse rotation guide for NEW viewport only. Minor feature not worth enabling for legacy 3D view.
Instead of exposing the base I think this may be nicer for the API. We still need a way to change select for a layer though (expose the ObjectBase perhaps? :/)
We may want to re-use part of this struct (or concept) for groups and armatures. But filter is something specific to SceneCollections, so may as well keep it in a separate struct, and re-evaluate that once/if we get to it.
Basically all this does is drawing layout previews into the opened layout search menu.
https://youtu.be/RHYWtZP7pyA
The previews are drawn using offscreen rendering so they can't use multi-threading (yet!). But that shouldn't be an issue since only a handful of previews are drawn at the same time. Normally we only need to redraw the preview if a screen layout was changed. Would be nice if PreviewImage could store if it supports threaded rendering.
Previews are saved in files, might be useful if you later want to support appending layouts.
Adds a new file screen_draw.c.
This fixes multiple problems on latest Mac OS + Xcode. Hopefully does not cause any on other platforms.
The Xcode detection logic could use further cleanup. It's checking several old versions that are unsupported for Blender 2.8+ development.
Needed because deps graph can destroy objects from any thread. We ran into the same problem & solved it in GPU_buffers.
Implemented in C++11 since it provides the needed machinery. The interface is in C like the rest of Gawain.
This means editing a property will now always affect all selected objects, bones or sequencer strips. Support for this was added in rBdfbb876d4660 but you had to hold the Alt-key to use it. The old behavior of only editing the active object will not be kept like decided in the 2.8 workflow meeting (reports comming). If you only want to edit the active object, you have to deselect others.
There are still a couple of issues to be resolved (listed below), but having it enabled by default helps testing and getting used to it and should motivate us to fix them ;)
To be fixed:
* Give users hint when edits are applied to all objects/bones/strips ("Applying to x objects") - there are ideas but we need to finalize and implement them
* Make it work better in corner cases (material editing, modifier property editing, etc)
Note: Values usually override the initial value of the object/bones/strips, except of number buttons where it depends if you enter the value (absolute override) or drag the value (add value change). This behavior is consistent with multi-button editing.
Updated the GL calls to the new immediate mode.
I left some glcolor calls which I'm not sure wether thats right?
Part of T49043
warm regards,
Sebastian Witt
Reviewers: merwin
Reviewed By: merwin
Tags: #bf_blender_2.8
Maniphest Tasks: T49043
Differential Revision: https://developer.blender.org/D2305
Just as mentioned in title. Need new functions for calls found in `transform.c`
T49043
Reviewers: merwin
Tags: #bf_blender_2.8
Differential Revision: https://developer.blender.org/D2358
This conversion is pretty straightforward.
The code for debug drawing is not great, but it does the job.
Rewriting it is for another day, if it becomes more widely used.
Convert UI_view2d_constant_grid_draw to new immediate mode.
Part of T49043.
Reviewers: merwin
Reviewed By: merwin
Tags: #bf_blender_2.8
Differential Revision: https://developer.blender.org/D2298
Convert ED_region_grid_draw to new immediate mode.
Part of T49043
Reviewers: merwin
Reviewed By: merwin
Tags: #bf_blender_2.8
Differential Revision: https://developer.blender.org/D2289
all is in the title too..
Reviewers: merwin
Reviewed By: merwin
Subscribers: Blendify, Severin
Tags: #bf_blender_2.8, #opengl_gfx
Maniphest Tasks: T49043
Differential Revision: https://developer.blender.org/D2337
Consumes much less memory (1/3 for both normals = 32 bytes less per edge). Same visual result.
We can pack normals for other draw modes to get similar savings.
Part of T49165
Most useful for packed normals, which take 1/3 the space of float32 normals.
2-bit alpha|w component is ignored for now.
Batch API can use these now, will add support to immediate mode API if desired.
Enabling on Windows first. Will enable on all platforms after we switch Blender to core profile.
Drop SciTech support & workarounds for WinCE and OpenGL ES.
AMD_debug_output is still in the code but disabled. Once I verify the newer extensions are available on all the GPU + OS combos we support we can delete this disabled code.
Motivations:
1) GLenum is too broad; tightly-defined enum just for this is safer.
2) enable a Vulkan future
New code should use these instead of GL_FLOAT etc. When all existing code has been updated to use new enum, we can drop compatibility with GLenum values.
Early work towards 10_10_10 format, more to come soon.
Caused by 4811b2d356 which caused the event handler hack that is used to fire up the file browser from other operators to fail. Basically the context from before the file browser is opened gets stored and used later for executing the actual file read/write operation (in this case, saving image). This context storage is cleared when exiting an editor since 4811b2d356, which is technically correct, but causes usage of NULLed context data in this case, because the file browser is exited before the file read/write operation is executed.
For now I solved this by moving the fileselect handler to list of normal handlers, instead of modal ones. 4811b2d356 only touches list of modal handlers so we avoid the crash. Ideally we'd completely refactor how the file browser opening works to get rid of these event handler hacks.
Note that I wouldn't be suprised if this causes other regressions, but I couldn't find one so worth a try.
As our library of built-in shaders grows, it's important to create, access, and discard them efficiently.
Lookup via GPU_shader_get_builtin is now constant time instead of linear (# of built-in shaders). This is called very often with our new immediate mode.
Creation and discard are unified.
Adding a new shader requires fewer steps.
365 lines shorter :D
all is in the title.
Reviewers: merwin
Tags: #bf_blender_2.8, #opengl_gfx
Maniphest Tasks: T49043
Differential Revision: https://developer.blender.org/D2336
This one is for the straight line (white with width 2.0 over a black with width 4.0) drawn when you use the gradient tool.
To test: Image editor, create / open an image, choose image paint mode and on the tool shelf: choose the Fill brush and enable "Use Gradient" for it. Then click and drag on the image.
From what I checked, calls to glLineWidth are not being removed yet, so I kept them.
Reviewers: dfelinto, Severin, merwin
Reviewed By: merwin
Tags: #bf_blender_2.8, #opengl_gfx
Maniphest Tasks: T49043
Differential Revision: https://developer.blender.org/D2312
D2311 by @ianwill
This is the radial control that appears when we change the size of a brush in sculpt and vertex and texture painting modes, by pressing "f".
Also includes a new built-in shader that can be useful in other places.
Part of T49043
Gawain does very strict runtime checking to help us catch coding errors. Final release should disable most of these checks, so I'm disabling now for all non-debug builds.
When writing Blender code that uses Gawain, always make debug builds and test there! "make lite debug" is my favorite.
Right now this only affects other objects in wireframe. The idea is to do something similar for other draw modes, and keep focus on the edit object.
As seen at #bcon16
Most of this was already in place, just enabling & adding comments.
One fix was needed to make batch uniforms stick between multiple draws.
Added comments to selection outline; no functional changes there.
As seen at #bcon16
Geometry shader version is automatically used on modern GL runtimes. Legacy version is used on pre-3.2 systems (Mac, Mesa compat profile). They have the same inputs and visual result.
TODO: specialized versions that are less flexible -- draw ALL edges or draw JUST silhouette edges.
Part of T49165
(also, fix warning regarding const float being written)
You only see the color if you use the "modern" viewport option
(otherwise I believe Blender is drawing the old on top of the new outline).
That said, in the "modern" viewport we have unfreed mem. To be
investigated separately.
Totally WIP.
Started with copies of legacy routines, modified to use the new shaders & batch cache. Not all features are implemented; this is why we keep legacy viewport around during development!
Gawain batches are built on demand while drawing, then kept in this per-DerivedMesh cache.
A mesh's batches try to share vertex buffers as much as possible.
Not sure if this file is the best home for this code, but functions in this file are the only users of the cache. So maybe.
Big part of T49165
Surveying buffer usage & clears for new viewport. Not yet perfect, but closer. Committing from Mac so I can test this on Windows.
Using new matrix API (T49450) for gradient background.
Removed some of my earlier glActiveTexture calls. After reviewing the
code I now trust that GL_TEXTURE0 is active by default. Fewer GL calls,
same results.
Fixed some misuse of glActiveTexture & glUniformi, mostly my fault.
Caught by --debug-gpu on Windows. Don't know why this appeared to be
working previously!
Plus some easy cleanup nearby.
Works great on Mac now. Will test on Windows & Linux (Mesa) tomorrow. Related to T49505
Main fix is glActiveTexture and immUniform1i.
TEXTURE_2D vs TEXTURE_RECTANGLE is now a compile-time option. Both are available starting in GL 3.1 so there's no need for a run-time check.
Removed glClears that I don't think are necessary.
Prevent TEXTURE_2D from creating extra mipmap levels. We only need level 0.
Some minor cleanup: booleans and variable declarations.
At the moment this already shows that the depth is the same after the solid plates and in the very end of drawing, while they should be different. Later on we can adapt this to show different buffers we want to debug.
I am using near=0.1, far=2.0 for my tests. I decided not to make a doversion for near/far because this is for debugging only
The active camera has a solid "up" triangle instead of the usual outline. We want to see both sides of this triangle. Disable face culling only when drawing the active camera, not for every camera.
- any shader program can use matrix state (not only built-in shaders)
- you can mix matrix & begin/end calls, and the bound shader will use the latest matrix state
Part of T49450 & T49043
Mostly the same as before. Except:
- avoid drawing same lines multiple times
- helper functions take "bool filled" argument instead of GLenum
- drawcamera_volume draws its own near & far planes
widget_draw_text was calculating wrong display length when field is too narrow to show entire input string. Gawain assert caught this 11 function calls away!
Thanks to @ianwill for reporting.
Fixes an assert in drawing code.
Might need further work to support variable-thickness strokes (from pressure-sensitive stylus). This all is due for geometry shader overhaul anyway.
- rename image shaders to describe exactly what they do
- rename inputs to match other built-in shaders
- set & use active texture unit
- no need to enable/disable textures with GLSL
- pull vertex format setup out of loops
Turns out CTX_wm_region returns mostly NULL in wm_manipulatormaps_handled_modal_update. Now propertly unsetting area/region data of handlers when deleting area/region.
Non-power-of-two textures are always allowed. Keeping the disabled checks in the code in case we support OpenGL ES in the future. Even then it should be a compile-time check, not at run-time.
Own mistake in 9a9a663f40. Guessed there is a case where we have to rebuild the tree but everything seemed fine... It didn't work in display modes like "Data-Blocks".
Couple of issues here:
* Missing initialization for 3D view keyframe options for "Reset to Default Theme"
* Alpha values not reset correctly on "Reset to Default Theme"
* Alpha values of timeline keyframe options not reset correctly for old files
Also corrected old version patches even though they're overridden later, to avoid more issues in case people copy this code.
Corrections to d7af7a1e04 and 8d573aa0ec
* LMB now replaces selection instead of adding to it. Shift+LMB adds to selection (or removes if already selected). This is usual selection behavior Blender.
* Outliner selection isn't completely separate from object/sequencer-strip/render-layer/... selection anymore, when selecting an outliner item we now always try to select (and activate) the object it belongs to. Previously you had to click the name or icon of an item to select the object (or whatever) and on empty space within the row to set outliner selection.
* Collapsed items may show click-able icons for their children (nothing new). Clicking on such an icon will also select the hidden item it represents now, you'll notice after opening the parent. This valid from a technical POV, I'm not sure if this is wanted from user POV though. Changing would be easy, feedback welcome!
* Code cleanup.
Part of T37430.
GL 3.3 is the new minimum. Compatibility profile for now, core profile eventually. During development, GL 3.0 (on Mesa) and 2.1 (on Mac) will still work.
Part of T49012
Make color values compact. Set color once per primitive. Use new immSkipAttrib to avoid useless color copies.
All of this should make text drawing less CPU hungry.
Now you can explicitly skip a vertex attribute -- you don't give it a value and it won't get a copy of the previous vert's value. Useful for flat interpolated per-primitive values.
This is an advanced feature. Expect garbage in the empty spaces, and copies of garbage if you rely on the attrib copy behavior after skipping.
This was already fast on Apple, but @Severin and @dfelinto noticed slowdowns in user prefs, which is text heavy.
The problem was immBeginAtMost not being smart about VBO write flushing. immBeginAtMost can use all of its allocated range or only a subrange. The previous code was forcing back-to-back draw calls and buffer writes to serialize. This commit lets OpenGL know that our VBO writes never overlap, so there's no need to wait.
Should be much faster now!
1 or 2 draw calls per node instead of 1 per socket (inputs + outputs).
Rearranged draw order so we set uniforms less frequently.
Some style & dead code cleanup.
Part of T49043
Smooth round point with outline (uniform color) and fill (varying color).
Updated shader naming scheme: a shader that doesn't deal with color does not have to say "no color". Vertex shaders do not have to say "uniform color" since their frag counterpart actually has the uniform. Each name should describe what that shader *does*, not what it *doesn't do*.
I use your new point shader to draw the node's soket
Reviewers: Severin, merwin
Maniphest Tasks: T49043
Differential Revision: https://developer.blender.org/D2286
Some little UI polish to get familiar with outliner code (but also because it's a useful feature). Committing to blender2.8 branch but can also port to master (2.7) if wanted.
This basically causes the mouse hovered element to be highlighted. Contrast of the highlight should be fine, even with a non-default theme. Also did some minor cleanup.
The problem was the function tried to draw a line with one point only. This fix will be replaced by new geometry shaders, but we need while this change is not ready.
Before this change, the stroke was filled only after complete the stroke drawing. For artist is better to get a feedback of the area he is filling while drawing, so this commit draws the filling area while drawing.
The triangulation of the stroke is recalculated every time the function is called because using a cache is not useful because the points information is changing all the time while the stroke is being drawing.
API stays exactly the same.
Attribute names can still be of variable length, as long as the average length does not exceed AVG_VERTEX_ATTRIB_NAME_LEN. Since this includes unused attributes (length = 0) the current avg of 5 might even be too high.
New functions activate & deactivate immediate mode. Call these when switching context and the internal VAO will be handled properly. VAOs are one of the few things *not* shared between OpenGL contexts.
Was multiplying matrices backward, so concatenation was broken. Fixed!
Also a way to mix legacy matrix stacks with the new library. Just during the transition! Anything within SUPPORT_LEGACY_MATRIX will go away after we switch to core profile.
Part of T49450
Reworked logic in the few places that still called this. Deleted the "GLSL not supported" fallbacks.
Also removed some nearby checks for ARB_multitexture and OpenGL 1.1. Blender 2.77 removed checks like this, but game engine still has some.
Built-in shaders now use uniforms instead of legacy built-in matrices. So far I only hooked this up for new immediate mode.
We use the same matrix naming convention as OpenGL, but without the gl_ prefix, e.g. gl_ModelView becomes ModelView.
Right now it can skip the new matrix stack and use the legacy built-in matrices app-side. This will help us transition gradually from glMatrix functions to gpuMatrix functions.
Still some work to do in gpuBindMatrices. See TODO comments in gpu_matrix.c for specifics.
@zeauro reported this issue:
texture2DRect needs the ARB_texture_rectangle extension.
But isn't that an OpenGL 2.1 feature and should be part of GLSL 1.2+?
This should fix it, and future shaders should do something similar.
Got rid of GLU and some matrix manipulation. Everything is shader driven now, drawn with point sprites.
Still plenty to do in this file...
Part of T49042 and T49043
Simple convert of drawing emboss lines to new immediate mode.
Part of T49043
Reviewers: merwin
Reviewed By: merwin
Subscribers: dfelinto
Tags: #bf_blender_2.8
Differential Revision: https://developer.blender.org/D2271
Made function categories more clear & added more notes about how to use this API.
immEndVertex is no longer part of the public API.
Minor cleanup & organizing of recent additions.
I change UI_draw_roundbox_gl_mode to use immediate API.
The rest of the change is the call to the function.
I also make some change in UI_ThemeColor4(int colorid) for eg to make convenience to use.
I would really like to know if it's the good way to do, if yes I will make all the change in the node_daw.c after, else say me what's wrong and how to deal with color else.
Reviewers: merwin, dfelinto, Severin
Reviewed By: merwin
Subscribers: fablefox, Severin
Tags: #bf_blender_2.8, #opengl_gfx
Maniphest Tasks: T49043
Differential Revision: https://developer.blender.org/D2274
@dfelinto reported crash when setting grid subdivisions too low.
Code was setting color twice and Gawain was catching this. Fix is to only set regular grid color when we have regular grid lines to draw. Then emphasized grid lines are free to set their own color further down.
This commit lands the core backend of the Custom Manipulators project onto the blender2.8 branch. It is a generic backend for managinig interactive on-screen controls that can be integrated into any 2D or 3D edito. It's also already integrated into the window-manager and editor code where needed.
NOTE: The changes here should not be visible for users at all. It's really just a back-end patch. Neither does this include any RNA or Python integration.
Of course, there's still lots of work ahead for custom manipulators, but this is a big milestone. WIP code that actually uses this backend can be found in the 'custom-manipulators' branch (previously called 'wiggly-widgets').
The work here isn't completely my own, all the initial work was done by @Antony Riakiotakis (psy-fi) and - although it has changed a lot since them - it's still the same in essence. He definitely deserves a big credit! Some changes in this patch were also done by @Campbell Barton (campbellbarton). Thank you guys!
Merge accepted by @brecht and @merwin.
Patch: https://developer.blender.org/D2232
Code documentation: https://wiki.blender.org/index.php/Dev:2.8/Source/Custom_Manipulator
Main task: https://developer.blender.org/T47343
More info: https://code.blender.org/2015/09/the-custom-manipulator-project-widget-project/
Want to avoid updating code we no longer use anyway.
Comments for areas to investigate or deadlines for deletion.
also some minor bool cleanup
Part of T49165
We can't prevent users from using this branch, so I'd say it's reasonable to be a bit careful about what we store to files. In this concrete case we were storing a bit-flag for temporary use (only during early viewport transition) in a bit-field that's saved in files. Doing so would mean we either can't reuse this bit later or we risk breaking files (admittedly, likely in a pretty minor way). Moved the bit-flag to a new bit-field which can be removed later.
Makes it easier to enable it and avoids jumping of the panel when activating/deactivating it (because some panels disappear then). Also changed how panel title is drawn to make it behave like other panels.
We will keep the old system working as long as we can. At the moment even the visibility flags we are getting from the old system. That will continue like this until we have decided on the new UI
A new option (set in the properties region) allows the user to pick the
"new viewport" for the rendering (in the UI: Modern Viewport).
For now we have a semi-blank file (view3d_draw.c) that can starts to take
over the drawing pipeline.
I can't guarantee we will be able to keep both drawing systems working
through the entire 2.8 development, but it should do for now.
also, we can use branches for some of the viewport development, but it's
better to keep things in 2.8 whenever we can, so people can test it.
Transition away from GLU and legacy matrix stack. Using point sprites eliminated the need for most of the matrix math!
Depends on decent support of large aliased points. NVIDIA is good at this, must test limits on AMD & Intel systems.
Still needs proper scaling based on view zoom.
Part of T49042, touches on T49043 and T49450.
Apparently GL_POINT_SPRITE is important to GL 3.2+ compatibility profile, not just to GL 2.1 as thought.
We'll remove this during the core profile transition.
At context startup, make sure our assumptions about the OpenGL version are true. Should match since we set up the contexts... but this is what asserts are for, to check "should"s!
Part of T49012
patch P397 by @lichtwert + minor const by @merwin
Notes from drawvertsN function:
this used to be called twice (once for selected/active, once for unselected -- guess: to avoid state switches[color]?)
this used to be called in a loop, too (subcurves), moved the loop here to avoid multiple init stuff
Part of T49043
WARNING! Full build is broken, alembic has not been merged in correctly and has some references to particle stuff.
Don't have time to tackle this now (and probably would be better if someone knowing what he's doing does it anyway).
Conflicts:
release/scripts/startup/bl_ui/properties_particle.py
source/blender/blenkernel/intern/library_remap.c
source/blender/blenkernel/intern/smoke.c
source/blender/editors/physics/particle_object.c
source/blender/editors/physics/physics_intern.h
source/blender/editors/physics/physics_ops.c
source/blender/editors/space_outliner/outliner_intern.h
source/blender/editors/space_view3d/drawvolume.c
source/blender/makesrna/intern/rna_smoke.c
Complete (for our needs) 2D & 3D transformation API. Should be easy to port legacy OpenGL matrix stack-based code to this. Still needs testing.
Ported ortho, frustum, lookAt functions from Viewport FX (rB194998766c65). Kept plenty of Viewport FX code from previous commit.
Stack API and 2D routines ported from Gawain. This version uses BLI_math library so everything is licensed under GPL instead of the usual MPL.
Part of T49450
mul_m4_m4m4(R, A, B) gives us R = AB in general. Existing code assumed the worst, that A and B both alias the output R. For safety it makes internal copies of A and B before calculating & writing R.
This is the least common case. Usually all 3 matrices differ. Often we see M = AM or M = MB, but never M = MM.
With this revision mul_m4_m4m4 is called in exactly the same way but copies inputs only when needed. If you know the inputs are independent of the output use the "uniq" variant to skip the saftety checks.
Is actually a redundant cast since Blender uses -funsigned-char, however I think it's fine to be explicit about it in new code so cast is required to make compiler happy. Am not a fan of -funsigned-char anyway...
Changed drawing to use smooth lines, and to fade away when axis points toward / away from screen. (transform manipulators do this already)
Also fixed a nearby (but unrelated) missing immUnbindProgram.
Part of T49043
Ignore texture matrix in the shader, stop messing with texture matrix in BLF code.
Use linear screen-space interpolation instead of perspective.
Avoid redundant call to glMatrixMode.
With USE_GLSL enabled, GPU_basic_shader(TEXTURE|COLOR) always rendered black. New shader uses a solid color + alpha channel of texture (which in our case is a font glyph). See fragment shader for details.
I prefer this approah -- multiple shaders that each do one thing well (and are easy to read/write/understand), instead of one shader that can do many things given the right options.
Fixed compile error in debug build (thanks mont29)
Renamed some functions for consistency.
New features:
Create a Batch with immediate mode! Just use immBeginBatch instead of immBegin. You can keep the result and draw it as many times as you like. This partially replaces the need for display lists.
Copy a VertexFormat, and create a VertexBuffer using an existing format.
Resize a VertexBuffer to a different number of vertices. (can only resize BEFORE using it to draw)
Was already done for immediate mode, but rearranged code to make a clean separation. Cleaned up #includes for code that uses this feature.
Added same for batched rendering.
Follow-up to rBddb1d5648dbd
API is nearly complete but untested.
1) create batch with vertex buffer & optional index buffer
2) choose shader program
3) draw!
This function modifies the GL program object, which reduces our ability to share a shader among meshes with different vertex formats. Recommended approach is to use get_attrib_locations.
Vertex Buffer to store vertex attribute data.
Element List (AKA Index Buffer) to select which vertices to use.
Batch combines these into an object that can be built once then drawn
many times.
Porting over from the C++ version… Most of this C code is compiled but
unused. Some of it is not even compiled. Committing now in case I’m
lost at sea.
Put Gawain source code in a subfolder to make the boundary between the
library and the rest of Blender clear.
Changed Gawain’s license from Apache to Mozilla Public License. Has
more essence of copyleft — closer to GPL but not as restrictive.
Split immediate.c into several files so parts can be reused (adding
more files soon…)
immBegin requires us to know how many vertices will be drawn. Most times this is fine, but sometimes it can be tricky. Do we make the effort to count everything in one pass, then draw it in a second?
immBeginAtMost makes this simple. Example: I'll draw at most 100 vertices. Supply only 6 verts and it draws only 6.
Any unused space is reclaimed and given to the next immBegin.
Was using GL_NONE to mean "no primitive" but GL_NONE and GL_POINTS are both defined as 0x0000.
Introducing PRIM_NONE = 0xF which does not clash with any primitive types.
Application code can pass ubytes, Gawain converts to float vec4 expected by shader.
For now the conversion is simple linear. We can add sRGB support later if needed.
Significant rewrite with some improvements.
Maintain visual hierarchy of the grid:
- emphasized lines draw atop normal lines
- axes draw atop all lines (same as before)
Draw axes only once, not twice.
Return early if nothing to draw.
Single draw call for the default case (grid floor with X and Y axes).
Z axis needs a second draw call because it uses 3D coordinates.
Part of T49043
Also reduced number of matrix ops by generating final positions directly.
Also removed a display list (deprecated in modern GL).
Tried to reuse sinval & cosval tables but those values are skewed (last value repeats first value, middle values are squished to compensate). Went with sinf & cosf instead.
Part of T49042
The little grabby handle in the corner of an area. Now uses 1 draw call
instead of 6.
Also one version of the (+) icon to show a hidden region. Why do we
have multiple versions of this?
Fixed a harmless signed/unsigned error.
Fixed a GL state error that prematurely disabled blending.
Added imm_draw_filled_circle function, which can be used for drawing
other widgets.
Work toward T49042 and T49043
In addition to pack of conflicts listed below, also had to comment out particle part of new Alembic code... :/
Conflicts:
intern/ghost/intern/GHOST_WindowWin32.cpp
source/blender/blenkernel/BKE_effect.h
source/blender/blenkernel/BKE_pointcache.h
source/blender/blenkernel/intern/cloth.c
source/blender/blenkernel/intern/depsgraph.c
source/blender/blenkernel/intern/dynamicpaint.c
source/blender/blenkernel/intern/effect.c
source/blender/blenkernel/intern/particle_system.c
source/blender/blenkernel/intern/pointcache.c
source/blender/blenkernel/intern/rigidbody.c
source/blender/blenkernel/intern/smoke.c
source/blender/blenkernel/intern/softbody.c
source/blender/depsgraph/intern/builder/deg_builder_relations.cc
source/blender/gpu/intern/gpu_debug.c
source/blender/makesdna/DNA_object_types.h
source/blender/makesrna/intern/rna_particle.c
Scanned Blender code for commonly used glVertex, glColor functions.
Implemented immVertex, immAttrib versions of these to ease transition
away from legacy OpenGL.
Keeping unused gla2D code because it might be useful, or inspire
something useful, for Blender 2.8 development.
Also removed an old Mac driver bug workaround. Disabled this before the
2.77 release and nobody has complained.
Errors are caught & reported by our GL debug callback. This gives us way more useful information than sporadic calls to glGetError.
I removed almost all use of glGetError, including our own GPU_ASSERT_NO_GL_ERRORS and GPU_CHECK_ERRORS_AROUND macros.
Still used in rna_Image_gl_load because it passes unvalidated input to OpenGL functions.
Still used in gpu_state_print_fl_ex as an exception handling hack -- will rewrite this soon.
The optimism embodied by this commit will not prevent OpenGL errors. We need to analyze what would cause GL to fail at certain points and proactively intercept these failures. Or guarantee they can't happen.
Legacy OpenGL has a matching Vertex3fv for every Vertex3f, and so on. Add something similar to Gawain, just for a few common functions. Might add more as the need arises.
These are intended for very simple drawing. No lighting etc.
Shares some fragment code with the 2D shaders.
Similar to their 2D counterparts, but are not combined because of
future plans for separate 2D & 3D matrix stacks.
EXT_gpu_shader4 lets us say “noperspective” in GLSL #version 120 just
like in later GLSL.
Mac shader now matches modern GLSL available on other platforms.
When we ask for GL 3.2 compatibility profile:
AMD (Radeon HD 6970) gives us exactly this version
NVIDIA (Quadro K600) gives at least this version
Still need to check Intel behavior
We want *at least* the version requested, plus more recent features if
available.
Both GPUs tested & mentioned above are capable of GL 4.5. With this
commit they both give 4.5 to Blender.
Helps most when real-world units are used.
Previous code started at the smallest visible unit (e.g. Inches) then
followed to Feet, Yards, Chains, Furlongs, Miles. Always to the largest
unit of the set, even though most would be way off screen.
New code knows whether it skipped any grid lines for the next unit to
fill in, can stop once all lines are on screen.
Previous commit works on Windows, found some issues after trying on Mac.
- benign warnings about && within ||
- replaced nearbyint() with round() to avoid floating point environment
surprises
- remquo function appears to be broken on Mac (!) results were way way
off. Replaced with simple division.
- minor tweaks to debug output
Work toward T49043, with a side of client vertex arrays.
Not a straightforward port from glVertex to immVertex since Gawain needs
to know how many vertices we'll be drawing *before* we start drawing.
Fixed these not-so-great aspects of grid drawing:
- coarse grids would draw atop some lines from the finer grids
- visible axes would draw atop lines from coarse grid
- axes were drawn even if they weren't in view
- terrible misuse of vertex arrays
- each line issued its own draw call
New code draws each line exactly once. The entire grid is one draw call.
Bonus: I had to / got to learn how the units system works!
New value of 4MB should handle our needs without taking up too many GPU
resources.
Old value of 1KB was for observing what happens when the buffer fills up
and we need to flush and start a new one.
Getting this ready for Gawain treatment.
Removed setlinestyle(0) -- solid lines are the default, hope this isn't
really needed.
Eliminated redundant math.
Arithmetic is still double precision, passed to OpenGL as single
precision. Even though it said GL_DOUBLE before, values were converted
to GL_FLOAT internally.
Use C99-isms for declaring variables close to where they're used.
Minor whitespace tweaks.
This serves as a good example of the Gawain API. (I’ve thought of a
better way to draw drop shadows, but that can wait!)
Part of T49043.
This is what I had in mind for D1753.
If you don’t specify a vertex’s color, it will use the color of the
previous vertex. Similar for all other attributes.
This matches the legacy behavior of glColor, glNormal, etc. *except* in
Gawain the first vertex of each immBegin must be fully specified. There
is no “current” color in the new system.
Should be simpler to use now.
Made vertex format structure private. New immVertexFormat() function
clears and returns the format. Devs can start with add_attrib(format...)
and not have to clear it first.
immBindProgram automatically packs the vertex format if needed.
Updated 3D cursor drawing to use new API.
glBindAttribLocation does not take effect until the program is
re-linked. In other words I was doing it wrong!
New code gets attrib locations from program, then remembers the attrib
-> location mapping for subsequent draw calls.
The program and VertexFormat are not modified (makes threading and reuse
easier).
There are older ways to give OpenGL hints about buffer invalidation, but
glInvalidateBufferData does exactly what we want. Use this function when
OpenGL 4.3 is available (Windows and proprietary Linux drivers).
Part of Gawain immediate mode.
When running blender --debug-gpu
Display which debug facilities are available. One of these, in order of preference:
- OpenGL 4.3
- KHR_debug
- ARB_debug_output
- AMD_debug_output
All messages are logged now, not just errors. Will probably turn some of these off later.
GL_DEBUG_OUTPUT_SYNCHRONOUS lets us break on errors and backtrace to the exact trouble spot.
Callers of GPU_string_marker no longer pass in a message length, just the message itself (null terminated).
Apple provides no GL debug logging features.
Old code had a mix of framebuffer error codes from OpenGL ES, EXT_framebuffer_object, and desktop GL. We use desktop GL (or ARB_framebuffer_object which acts just like GL 3.x) so I made it compatible with that.
Changed messages to the actual GL_FRAMEBUFFER_XXX symbols. These are less friendly and more accurate. Can easily look up what an error means, unfiltered by what a Blender dev thinks it means.
Kept ES error codes around in case we support that one day. Just flip the #if or use a compile-time option.
Replace legacy OpenGL with Gawain. Use shaders instead of fixed
function pipeline.
This simple UI element is shown at startup so is easy to verify things
are working. It also serves as a good example for people converting
other parts of the code.
Part of T49043
Use simple alternating colored lines instead of stippled overdraw.
Reimplement circ function to not use deprecated GLU (T49042). It also
leaves matrix stack untouched.
Remove unused circf function.
immBindBuiltinProgram extends Gawain’s immBindProgram to use Blender’s
library of built-in shader programs.
It uses imm prefix instead of GPU_ so people won’t be tempted to call
GPU_unbind_program() afterward.
From my understanding, Apache code is not allowed to call GPL code, so
this function needs to be in the GPU lib.
How to use:
1) set up vertex format
2) bind a shader
3) draw with immBegin … immEnd
4) unbind shader
TODO: expand this a little, so we can send uniform values to the bound
shader.
glMapBufferRange is a wonderful function that doesn’t exist on GL < 3.0.
Use the APPLE_flush_buffer_range extension on Mac. It offers several of
glMapBufferRange’s benefits.
Use older “black arts” method to orphan VBOs when we are done with
them. In modern OpenGL this behavior is more obvious.
Add APPLE_flush_buffer_range to Mac requirements. Every GPU is
supported. T49012
Apple invented VAOs and exposes them via an extension in legacy GL.
Other platforms use at least GL 3.0 which has VAOs built in.
QUADS were removed from core profile but are useful for immediate-mode
drawing. We’ll have to implement our own QUAD drawing before switching
to core profile.
Immediate mode no longer leaves its internals bound after use. Part of
transition from a simple prototype app to non-simple Blender, which has
lots of other parts using OpenGL.
More ways to send values via immAttrib:
2D float vectors
3 & 4 component ubytes (for colors mostly)
New immVertex functions that act more like familiar glVertex. We’ll
find a balance between making this API convenient and keeping it small.
2f and 3f are enough for now.
ARB_framebuffer_object replaces several related EXT extensions. The ARB
version pulls GL 3 FBO features into GL 2.1, useful for Mac platform.
Its functions and enums have no ARB suffix so transition to modern GL
will be seamless!
Extension is checked at startup, so is guaranteed to be true at runtime.
Part of T49012
Mac’s OpenGL version is furthest away from our target of GL 3.2. This
commit brings Mac closer to other platforms, so that our shaders and
other code don’t diverge too much during development.
According to Apple’s OpenGL matrix these useful extensions are
available on all GPUs that will be able to run Blender 2.8.
Only checked in debug builds; we might need something more forceful.
Part of T49012
The first two of several new simple built-in shaders (will test these
before adding more). These are intended for the new immediate mode API,
but you can use them just like any built-in GPUShader.
Due to limitations on different platforms, shaders need to work with
GLSL versions 120, 130 and 150. Final Blender 2.8 will be pure #version
150.
Introducing an immediate mode drawing API that works with modern GL 3.2
core profile. I wrote and tested this using a core context on Mac.
This is part of the Gawain library which is Apache 2 licensed. Be very
careful not to pull other Blender code into these files.
Modifications for the Blender integration:
- prefix filenames to match rest of Blender’s GPU libs
- include GPU_glew.h instead of <OpenGL/gl3.h>
- disable thread-local vars until we figure out how best to do this
This implements Mac part of T49012.
Removed options for EGL, ES2, compatibility profile. None of these
exist on Mac platform.
Create a GL 3.2 core context when requested at build time. Old code
just pretended to support core profile.
Implements the Linux part of T49012.
Simplify the options for context creation. No options for legacy GL or EGL or ES2. Select 3.2 CORE or COMPATIBILITY profile at build time.
If that fails, use a GL 3.0 context. This keeps Mesa supported while we work on full 3.2 core elsewhere in the code.
This greatly simplifies the options for context creation. No options for
legacy GL or EGL or ES2. Select CORE or COMPATIBILITY profile at build
time.
OpenGL 3.2 core profile will be our final target on all platforms. Until
all our code is ready we can use 3.2 compatibility profile or "legacy"
GL 2.1 on platforms that don't support compatibility profile.
At the moment light shading in Blender is produced in viewspace. Apparently, that's why
shader nodes work with normals in camera space. But it is not convenient for artists.
The more convenient approach is implemented in Cycles where normals are represented in world space.
Blend4Web Team designed the engine keeping in mind shader parameters readability,
so normals are interpreted in world space as well. And now our users have to use some tweaks, like
empty node group with the name "Replace", which is replacing one input by another on the engine side
(replacing working configuration in Blender Viewport by the configuration that has the same behavior in the engine).
This patch adds the ability to switch to world space for normals and lamp vector in BI and Viewport.
This patch is very important to us and we crave to see this patch in Blender 2.7 because
it will significantly simplify Blend4Web material creation workflow.
{F315547}
{F315548}
Reviewers: campbellbarton, brecht
Reviewed By: brecht
Subscribers: homyachetser, Evgeny_Rodygin, AlexKowel, yurikovelenov
Differential Revision: https://developer.blender.org/D2046
The purpose of the patch is to replace deprecated glShadeModel.
To decrease glShadeModel calls I've set GL_SMOOTH by default
Reviewers: merwin, brecht
Reviewed By: brecht
Subscribers: blueprintrandom, Evgeny_Rodygin, AlexKowel, yurikovelenov
Differential Revision: https://developer.blender.org/D1958
Note that this only removes the actual dependencies of Cycles on the
particle code in Blender, but not the internal "particle" definition
or the curve type handling inside Cycles. These structures may be in need
of some improvement themselves, but that is out of scope here.
There are a lot of cases here where deciding for removal is a bit tricky.
Many features have options for "use_particles" and similar settings. Only
features which actually store a particle object reference or work on actual
particle data have been removed.
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.
