This commit enables new unaligned BVH builder and traversal for scenes
with hair. This happens automatically, no need of manual control over
this.
There are some possible optimization still to happen here and there,
but overall there's already nice speedup:
Master Hair BVH
bunny.blend 8:06.54 5:57.14
victor.blend 16:07.44 15:37.35
Unfortunately, such more complexity is not really coming for free,
so there's some downsides, but those are within acceptable range:
Master Hair BVH
classroom.blend 5:31.79 5:35.11
barcelona.blend 4:38.58 4:44.51
Memory usage is also somewhat bigger for hairy scenes, but speed
benefit pays well for that. Additionally as was mentioned in one
of previous commits we can add an option to disable hair BVH and
have similar render time but have memory saving.
Reviewers: brecht, dingto, lukasstockner97, juicyfruit, maiself
Differential Revision: https://developer.blender.org/D2086
This commit implements traversal of unaligned BVH nodes.
QBVH traversal is fully SIMD optimized and calculates orientation
for all 4 children at a time, regular BVH might probably be optimized
a bit more.
This is a special builder type which is allowed to orient nodes to
strands direction, hence minimizing their surface area in comparison
with axis-aligned nodes. Such nodes are much more efficient for hair
rendering.
Implementation of BVH builder is based on Embree, and generally idea
there is to calculate axis-aligned SAH and oriented SAH and if SAH
of oriented node is smaller than axis-aligned SAH we create unaligned
node.
We store both aligned and unaligned nodes in the same tree (which
seems to be different from what Embree is doing) so we don't have
any any extra calculations needed to set up hair ray for BVH
traversal, hence avoiding any possible negative effect of this new
BVH nodes type.
This new builder is currently not in use, still need to make BVH
traversal code aware of unaligned nodes.
This seems to be straightforward way to support heterogeneous nodes
in the same tree.
There is some penalty related on 4gig limit of the address space now,
but here's are the thing:
Traversal code was already using ints to store final offset, so
there can't be regressions really.
This is a required commit to make it possible to encode both aligned
and unaligned nodes in the same array. Also, in the future we can use
this to get rid of __leaf_nodes array (which is a bit tricky to do since
trickery in pack_instances().
There are several internal changes for this:
First idea is to make __tri_verts to behave similar to __tri_storage,
meaning, __tri_verts array now contains all vertices of all triangles
instead of just mesh vertices. This saves some lookup when reading
triangle coordinates in functions like triangle_normal().
In order to make it efficient needed to store global triangle offset
somewhere. So no __tri_vindex.w contains a global triangle index which
can be used to read triangle vertices.
Additionally, the order of vertices in that array is aligned with
primitives from BVH. This is needed to keep cache as much coherent as
possible for BVH traversal. This causes some extra tricks needed to
fill the array in and deal with True Displacement but those trickery
is fully required to prevent noticeable slowdown.
Next idea was to use this __tri_verts instead of __tri_storage in
intersection code. Unfortunately, this is quite tricky to do without
noticeable speed loss. Mainly this loss is caused by extra lookup
happening to access vertex coordinate.
Fortunately, tricks here and there (i,e, some types changes to avoid
casts which are not really coming for free) reduces those losses to
an acceptable level. So now they are within couple of percent only,
On a positive site we've achieved:
- Few percent of memory save with triangle-only scenes. Actual save
in this case is close to size of all vertices.
On a more fine-subdivided scenes this benefit might become more
obvious.
- Huge memory save of hairy scenes. For example, on koro.blend
there is about 20% memory save. Similar figure for bunny.blend.
This memory save was the main goal of this commit to move forward
with Hair BVH which required more memory per BVH node. So while
this sounds exciting, this memory optimization will become invisible
by upcoming Hair BVH work.
But again on a positive side, we can add an option to NOT use Hair
BVH and then we'll have same-ish render times as we've got currently
but will have this 20% memory benefit on hairy scenes.
It was initially unsupported because initial idea of checking visibility
of all children was slowing scenes down a lot. Now the idea has changed
and we only perform visibility check of current node. This avoids huge
slowdown (from tests here it seems to be withing 1-2%, but more tests
would never hurt) and gives nice speedup of ray traversal for complex
scenes which utilized ray visibility.
Here's timing of koro.blend:
Without visibility check With visibility check
Original file 4min 20sec 4min 23sec
Camera rays only 1min 43 sec 55sec
Unfortunately, this doesn't come for free and requires extra data in
BVH node, which increases memory usage of BVH nodes by 15%. This we
can solve with some future trickery of avoiding __tri_storage created
for curve segments.
Those unwrap operators are a bit tricky, some are available from both 3DView and UVEditor, others only from 3DView...
Hacked around this by returning Mesh keymap for UV_OT ops for specific 3DView/MeshEditMode context.
This reverts commit 4fd78bb06f.
After further testing, it turns out that these options are less-broken than
I remember them being (and have been hearing about). Specifically, as long
as you disable all 3-axes of a transform component (i.e. all location, all
rotation, all scale) you're not likely to have problems, whereas if you only
disabled one axis (i.e. y-rotation), you may have problems in some cases.
So, restoring these to the UI.
Currently "long keyframes" are only useful for indicating where stationary
holds occur. If however you try to create a "moving hold" (where the values
are slightly different, but in terms of overall effect, it's still a hold)
then it could get tricky to keep track of where these occur.
Now it's possible to tag such keyframes (using the keyframe types - RKEY)
as being part of a moving hold. These will not only be drawn differently
from normal keyframes, but they will also result in a "long keyframe"
being drawn between each pair of them, just like if they had been completely
stationary instead.
Currently the theming/styling of these is a bit rough. They reuse the existing
theme colours for long keyframes.
* Added new operators to take the current transform value (loc/rot/scale or all 3)
and convert/apply that transform to a corresponding delta transform value. By default,
the transform value will be zeroed out again afterwards, so you don't end up with a
double transform.
* These operators can be found in the "Apply" menu (Ctrl-A)
* The "Animated Transforms to Deltas" (which does a similar job, except it adjusts all
existing animation data instead of the current transform) has also been moved to the
Apply menu (it was in the Transform menu instead)
A long requested feature has been to have objects appear in alphabetical order
in the animation editors, so that it is easier to find where they occur. This
commit implements support for this.
The main sticking point has been the performance impact of having this sorting
happening all the time (as the actual list of "bases" cannot be modified, as the
old depsgraph still needs random-looking unsorted order of objects for scheduling
updates). However, it recently occurred to me that perhaps by restricting it to
the one case where the ordering actually matters (i.e. when we're getting the channel
list for drawing all channels, vs operating on them), and adding a toggle to turn the
sorting off in heavy scenes when it might bog down things, that it will probably
be acceptable enough in general. Furthermore, if things get really bad, we can investigate
putting in place some sort of caching scheme for this too - though hopefully the
new depsgraph will make that unnecessary (i.e. it doesn't sort the bases directly anymore).
This commit introduces a scale factor setting for scaling all keyframe indicators
in the Dopesheet Editor up/down, in order to make them easier to select. It is perhaps
most useful for keyframe types which are usually indicated using smaller keyframes
(e.g. breakdown), which may get tricky to quickly select.
To make it easier to synchronise timing across multiple strips, if you add
markers locally to an action, these will show up in the NLA strip in the
NLA Editor. These markings can then be used to line up the start/end of
another strip, or even to make sure that the markers from two different
strips end up lining up.
By default, this is turned on, but it can be disabled (via the View menu)
if it adds too much visual noise.
This breaks any post-versionning (like IPO conversion, python handler, etc.).
rB961ebfa8c40b9909 mentions some Main being do_versionned several times (which is not desired for sure),
will try to reproduce again and find another fix.
Since SDNA was allocated for each undo step,
the new address meant it was considered different and included again.
Add an option not to duplicate the DNA string when calling DNA_sdna_from_data,
as well as avoiding a redundant copy, it writes the same address each time.
Idea is to replace hard-to-track (id->lib != NULL) 'is linked datablock' check everywhere in Blender
by a macro doing the same thing. This will allow to easily spot those checks in future, and more importantly,
to easily change it (see work done in asset-engine branch).
Note: did not touch to readfile.c, since there most of the time 'id->lib' check actually concerns the pointer,
and not a check whether ID is linked or not. Will have a closer look at it later.
Reviewers: campbellbarton, brecht, sergey
Differential Revision: https://developer.blender.org/D2082
Checking if faces exist was a bottleneck,
use a simpler version of this function for triangles.
Gives approx 1.6x overall speedup (when many edges are being collapsed).
When snapping to edge/vert, check the distance to the ray
instead of the screen-space pixel projection.
This also corrects the conversion of `dist_to_ray_sq` to `dist_px` which was being done incorrectly.
While this change was planned, it fixes T48791, caused by error in b01a56ee.
