I've been going over the zbuf.c code, which is indeed very ancient,
with a load of old optimizing and redundant code in use.
Added more 'modern' Span support, which fills per face two arrays
with the scanline information in it. That way you can zbuffer a quad in one
run as well. It was also exactly that code that's copied all over in zbuf.c
For now, to prevent issues for the release, the 'render a quad in 1 run' is
only in use for the strand render. Tests reveil a speedup of about 33%.
Will work on this recode later... which would also result in making zbuf.c
threadsafe.
And: bugfix #3398
When using the new 'render emitter' for particles, the orco array for
particles was accidentally used by mesh too.
doc;
http://www.blender3d.org/cms/Hair_Strand_Rendering.722.0.html
- added width control for strands
- made tangent (anisotropic) render an option
(so you can render strands more solid, like metal/wood)
Also:
- ALT+A anim playback with static particles made cursor flashing
http://www.blender3d.org/cms/New_Particle_options_a.721.0.html
There's no doubt this patch had a lot of good ideas for features, and I
want to compliment Janne again for getting it all to work even!
A more careful review of the features and code did show however quite some
flaws and bugs... partially because the current particle code was very much
polluted already, but also because of the implementation lacked quality.
However, the patch was too good to reject, so I've fixed and recoded the
parts that needed it most. :)
Here's a list of of most evident changes in the patch;
- Guides support recoded. It was implemented as a true 'force field',
checking all Curve path points for each particle to find the closest. Was
just far too slow, and didn't support looping or bends well.
The new implementation is fast (real time) and treats the paths as actual
trajectory for the particle.
- Guides didn't integrate in the physics/speed system either, was added as
exception. Now it's integrated and can be combined with other velocities
or forces
- Use of Fields was slow code in general, made it use a Cache instead.
- The "even" distribution didn't work for Jittered sample patterns.
- The "even" or "vertexgroup" code in the main loops were badly constructed,
giving too much cpu for a simple task. Instead of going over all faces
many times, it now only does it once.
Same part of the code used a lot of temporal unneeded mallocs.
- Use of DerivedMesh or Mesh was confused, didn't work for Subsurfs in all
cases
- Support for vertex groups was slow, evaluating vertexgroups too often
- When a vertexgroup failed to read, it was wrongly handled (set to zero).
VertexGroup support now is with a name.
- Split up the too huge build_particle() call in some parts (moving new code)
- The "texture re-timing" option failed for moving Objects. The old code used
the convention that particles were added with increasing time steps.
Solved by creating a object Matrix Cache.
Also: the texture coordinates had to be corrected to become "OrCo".
- The "Disp" option only was used to draw less particles. Changed it to
actually calculate fewer particles for 3D viewing, but render all still.
So now it can be used to keep editing realtime.
Removed;
The "speed threshold" and "Tight" features were not copied over. This
resembled too much to feature overkill. Needs re-evaluation.
Also the "Deform" option was not added, I prefer to first check if the
current particle system really works for the Modifier system.
And:
- Added integration for particle force fields in the dependency graph
- Added TAB completion for vertexgroup names!
- Made the 'wait cursor' only appear when particles take more than 0.5 sec
- The particle jitter table order now is randomized too, giving much
nicer emitting of particles in large faces.
- Vortex field didn't correctly use speed/forces, so it didn't work for
collisions.
- Triangle distribution was wrong
- Removed ancient bug that applied in a *very* weird way speed and forces.
(location changes got the half force, speed the full...???)
So much... might have forgotten some notes! :)
from 0 - 1, instead of -1 to 1. Thats fixed.
The error made tests i did with alpha make nice though, apparently hairs
are nicer when rendered with alpha range 0.0 to 0.5.
This means the diffuse and specular shaders don't use the normal
for hair (which is actually undefined, a hair is micro cylinder) but
it uses the tangent vector (vector in direction of hair).
For Diffuse, it computes a fake normal now, representing the optimal
hair normal pointing towards the light. All current builtin shaders
work with this, including ramps.
For Specular, it uses another formula to remap dot products for all
lines that now use the tangent vector instead of the normal:
dot = vector * tangent
dot = sqrt(1.0 - dot*dot)
Gives better results than using the 'fake' normal for diffuse. Officially
(according the papers) this could be used for diffuse too, but then hair
becomes very flat. Now you can control the flatness easily with ramps or
using Oren-Nayer for example.
Example image (disappears in some weeks)
http://www.blender.org/bf/rt9.jpg
- Added new texture channel "Strand" to apply textures on hairs over the
length of hair (1 dimensional). Orco now gives 1 fixed coordinate for
the entire hair, based on where it starts.
Note; UV doesn't work yet. Nor vertexcolor.
http://www.blender.org/bf/rt10.jpg
Thanks to testing in studio orange (thnx andy, matt!) I've found the
simple way to code it.
Static particle systems, when not set to wire or halo, now render 1 pixel
wide 'strands', which are actually just faces with vertexnormals and
proper orco texture. Check for quick fun;
http://www.blender.org/bf/rt5.jpg
(and rt6, rt7, rt8)
<blush>Missing "2*" caused AO tables to be only initialized half</blush>
Result was bad AO quality in render, and unpredictable brightness. This bug
happened in previous commit, when fixing random table issues.
- AO and soft shadow AreaLight tables were generated without fixed seed,
causing animations to give unwanted amounts of noise.
- Made sure these tables now are calculated before render, with fixed seed
- Then found out the BLI_rand() has very bad seeding... it showed up as
patterns. After some experimenting, found a nice method using noise.c
hash tables. For compatibility with old code, named it BLI_srandom() to
use this next to the BLI_srand(). This follows libc rand() and random()
naming convention.
- Then of course threading should work... so made a BLI_thread_rand version
of the calls. Now supports up to 16 threads, comments added in .h and .c
Result is stable animation render with AO and soft shadow. But, please
test and feedback!
to get rid of faces with MFace.v3==0
- change all Mesh's to have ->medge now. This is forced by make_edges
on readfile, and in the various exotic important routines, and on
conversion back in python.
- make python NMesh structure always have medges now (needs testing)
- with above two changes it is guarenteed that mf->v3 is never ==0
in main blender code (i.e., all MFace's are actually triangles
or quads) and so I went through and removed all the historic tests
to deal with MFace.v3==0. Equals lots of deleting, I am in heaven!
- removed MEdge edcode flag, no longer needed
- added experimental replacement for edge flag system
Still are some inconsistencies in FACESELECT mode edge drawing to
be ironed out.
NOTE: This commit adds an experimental edge flag calc system, based
on 10-seconds-of-thought algorithm by yours truly. Would appreciate
feedback on how this system works, esp compared to old one and esp
on complex or interesting models.
To Use: New system is enabled by setting G.rt to a value between
1 and 1000 (Value of 0 uses old system). Value 1000 is reserved for
"auto" edge, which is more or less identical to old system but also
makes sure that at least 10% of edges are drawn (solves errors for
super subdivided meshes). Values between 1 and 999 act as percent
(out of 1000) of edges that should be drawn, starting with "most
interesting" edges first. Please try it and comment!
- removed {lattice,curve}_modifier functions
- changed render code to use displist for curve rendering
instead of making its own. required adding a bevelSplitFlag
field to DispList. I also fixed the bevel face splitting
which did not work correctly in many situations.
- changed so all curve data creation happens in makeDispListCurveTypes,
includes making bevel list and filling polys
- changed render code to use displist for surface rendering
- removed Curve.orco variable, built as needed now
- removed stupid BLI_setScanFill* functions... why use a function
argument when you can use a global and two functions! Why indeed.
(this fixed crash when reloading a file with filled curves and
toggling editmode)
- bug fix, setting curve width!=1 disabled simple bevel for no
apparent reason
- cleaned up lots and lots of curve/displist code (fun example:
"if(dl->type==DL_INDEX3 || dl->type==DL_INDEX3)"). Hmmm!
- switched almost all lattice calls to go through lattice_deform_verts,
only exception left is particles
- added DBG_show_shared_render_faces function in render, just
helps to visualize which verts are shared while testing (no
user interface).
- renamed some curve bevel buttons and rewrote tooltips to be
more obvious
- made CU_FAST work without dupfontbase hack
Also by the way I wrote down some notes on how curve code
works, nothing spiffy but it is at:
http://wiki.blender.org/bin/view.pl/Blenderdev/CurveNotes
really work at all. Fortunately no one actually *USES* nurbs so
no one noticed. This is bug from NaN days (strubi, I am looking at
you here) which I guess says a lot (or a little) about NURBS usage.
And as usual, the correct code is shorter.
- added do_version copy of ME_OPT_EDGES flag
- added ME_EDGERENDER flag, barely changes things atm except makes
sure plain meshes with FasterDraw/etc set still render all edges.
The edge drawing system needs a bit of a revamping - it is a cool
feature but could use several improvements:
(1) The algorithm could be better in choosing the best edges to
draw.
(2) The drawflags should interact well with modifiers. It is wierd
to have a large grid with a deformer that draws no edges because
flags are only calculated based on base mesh.
(3) Drawflags should not be destroyed by editmode. Better design
would be a "Draw % of edges" button.
Of course, could also be the feature is not worth it and we
should just drop. Feel free to comment if you have an opinion.
DLM to share data from DerivedMesh (reduces some copying/memory allocation)
- added displistmesh_copyShared function to copy a DLM but not duplicate any
internal data
- changed crease drawing to use DerivedMesh functions... this means varying
edge width style of creases had to go, I replaced by using varying color to
show crease weight instead. Don't think this is a big loss since the subsurf
result gives you a much better indication of the crease weight anyway.
- bug fix in mirror modifier, didn't copy edge creases from editmesh correctly
- XCode project
- Scons :
scons stopped working for Os X in the last month (dont know when) :
* the '.' and '..' keywords in CCPPATH
are not recognized anymore for sconscripts compiling files at
a sub level
* when doing a scons clean, the 3 subdirs in build dir are removed
and scons then fail to recreate them (exten, intern, source)
this commit solve the first problem as a temp workaround
for the latter simply recreate the dirs manually
I will investigate that further when on vacations, which i should already
be.
;(
all 3 build systems are now working on Os X
- remove python access to Optimal and Subsurf flags (they don't
work this way anymore, I suppose need to replace with python
access to modifiers but not going to do right now).
- removed interface access to OPTIMAL mode, needs to be rethough...
this means at the moment subsurfs outside editmode always draw
and render all edges
- Bug #2857: Spin didn't create nice consistant normals
- removed unnecessary call to where_is_object() in init-render phase.
- Added DAG_scene_sort() calls when objects were removed (join cases)
- When using texture fonts, the file window header didn't display OK
- Saving a file didn't set the 'wait cursor' anymore (oldie!)
and passed over faces twice
- changed init_render_mesh to *always* call calc_vertexnormals. Vertex normal
calculation is insignificant time wise in comparison to modern renders so
it wasn't a big savings anyway, and vertex normals were sometimes incorrect.
Still issue remaining regarding what to do with wire normals.
based on time change. would be nice if dep graph could handle this.
- made dep check if modifiers need update on time change
- fix render crash (access null)
- added new Build Effect modifier type. compared to old one works as
a full member of modifier system, means can apply subsurf, etc on
it, reorder, what have you. and it is all nice and self contained.
- removed old Build effect, old files convert to new style on load
- couldn't help myself, added a randomize feature to build effect
- removed Python BuildEffect support
builds new DerivedMesh... caching can come later)
- split DerivedMesh returning functions into editmesh and mesh groups
- got rid of DL_NORS displist type (get built on fly for mesh when
needed)
- got rid of Mesh.disp (yay!)
- started to punch DerivedMesh returning functions into shape to introduce
modifier stack
- switch renderer to store orco's in a hash table instead of
caching in mesh (I don't like the renderer caching data
in the actual mesh structure)
- added mesh_create_orco[_render] function
- changed mesh_get_derived_render to always return a DerivedMesh (even if
no subsurf)
- changed init_render_mesh to always get the mesh data through a
DerivedMesh
situations where data can be shared easily.
- added convertDisplistToMesh function for regular mesh DerivedMesh
interface (how many times can *you* use mesh in one sentence?)
- do_puno was uninitialized in init_render_mesh
- added mesh_get_derived_final (temporary), difference from
mesh_get_derived is it always returns a derived mesh, even if
no subsurf.
- appropriate callers of makeDispList replaced with depgraph calls
- unappropriate places just killed... small chance this gives some
errors in corner cases if dep graph isn't notified (example, font
family displists) but these can be tracked down as they show up.
- still a large number of callers of makeDispListCurveTypes, but
makeDispListMesh has just a few.
still a makeDispList that dispatches to the appropriate one.
makeDispList is on the way out and this makes it easier to track down
exactly which places use makedispList and for what types of objects.
- switch calls to makeDispList to appropriate more specific function (if
the object type is known by caller).
- added mesh_changed function that invalidates cached mesh data (but does
not rebuild, mesh data gets rebuilt on access). Most old calls to
makeDispListMesh use this instead now.
Aim was to get a total refresh of the animation system. This
is needed because;
- we need to upgrade it with 21st century features
- current code is spaghetti/hack combo, and hides good design
- it should become lag-free with using dependency graphs
A full log, with complete code API/structure/design explanation
will follow, that's a load of work... so here below the list with
hot changes;
- The entire object update system (matrices, geometry) is now
centralized. Calls to where_is_object and makeDispList are
forbidden, instead we tag objects 'changed' and let the
depgraph code sort it out
- Removed all old "Ika" code
- Depgraph is aware of all relationships, including meta balls,
constraints, bevelcurve, and so on.
- Made depgraph aware of relation types and layers, to do smart
flushing of 'changed' events. Nothing gets calculated too often!
- Transform uses depgraph to detect changes
- On frame-advance, depgraph flushes animated changes
Armatures;
Almost all armature related code has been fully built from scratch.
It now reveils the original design much better, with a very clean
implementation, lag free without even calculating each Bone more than
once. Result is quite a speedup yes!
Important to note is;
1) Armature is data containing the 'rest position'
2) Pose is the changes of rest position, and always on object level.
That way more Objects can use same Pose. Also constraints are in Pose
3) Actions only contain the Ipos to change values in Poses.
- Bones draw unrotated now
- Drawing bones speedup enormously (10-20 times)
- Bone selecting in EditMode, selection state is saved for PoseMode,
and vice-versa
- Undo in editmode
- Bone renaming does vertexgroups, constraints, posechannels, actions,
for all users of Armature in entire file
- Added Bone renaming in NKey panel
- Nkey PoseMode shows eulers now
- EditMode and PoseMode now have 'active' bone too (last clicked)
- Parenting in EditMode' CTRL+P, ALT+P, with nice options!
- Pose is added in Outliner now, with showing that constraints are in
the Pose, not Armature
- Disconnected IK solving from constraints. It's a separate phase now,
on top of the full Pose calculations
- Pose itself has a dependency graph too, so evaluation order is lag free.
TODO NOW;
- Rotating in Posemode has incorrect inverse transform (Martin will fix)
- Python Bone/Armature/Pose API disabled... needs full recode too
(wait for my doc!)
- Game engine will need upgrade too
- Depgraph code needs revision, cleanup, can be much faster!
(But, compliments for Jean-Luc, it works like a charm!)
- IK changed, it now doesnt use previous position to advance to next
position anymore. That system looks nice (no flips) but is not well
suited for NLA and background render.
TODO LATER;
We now can do loadsa new nifty features as well; like:
- Kill PoseMode (can be option for armatures itself)
- Make B-Bones (Bezier, Bspline, like for spines)
- Move all silly button level edit to 3d window (like CTRL+I = add
IK)
- Much better & informative drawing
- Fix action/nla editors
- Put all ipos in Actions (object, mesh key, lamp color)
- Add hooks
- Null bones
- Much more advanced constraints...
Bugfixes;
- OGL render (view3d header) had wrong first frame on anim render
- Ipo 'recording' mode had wrong playback speed
- Vertex-key mode now sticks to show 'active key', until frame change
-Ton-
ESC during alt+a playback.
Solved in the 'proper' way, which is not abusing the (while render)
ESC callback for UI draw of stars, this confused everything. Means;
you cannot esc stars drawing anymore. Nice feature, bad hack...
Appending from files also set the string for "last loaded file". That's
a real bad one... so you can save over accidentally libraries.
Another fix: appending from files that have dynamic other files linked
with relative paths didn't work. (Yah, now it should all work!)
renderconverter change is just a more clear comment.
parameters the material preset menu won't be as useful. Both glass presets will look the same
because there is no 'filter' parameter in the old yafray for instance.
So using the new Blender version with an old yafray version should work a bit better,
though the other way around, using the new yafray with an old blender version, will generally
not work as well.
I added a few extra things. In 'yafray' panel re-arranged some buttons, and added a new
button 'Clamp RGB'. This button will be enabled by default and helps to improve AA on
high contrast edges in the image. When using bokeh however, it is best to switch this off,
otherwise lens shaped highlights will be quite a bit less visible.
Changed the 'extinction' parameter name to the probably more correct term 'absorption',
though mathematically it works out the same. Also changed the behaviour of this color,
it no longer specifies a color that will be removed as I wrote in the previous commit,
but instead the actual color at one (blender) unit of distance. The 'Ds' (distance scale)
button below the color sliders controls the scaling of this unit distance.
What this means is that if you take the standard blender cube, which covers two units of
distance by default, setting the distance scale button to 2.0 will make sure that the color
you specified is exactly that color at that distance (provided the base color itself is white
of course, or 'filter' is 0, otherwise it will be filtered by the base color too).
Beyond this distance the color will get darker.
The glow option for point/soft/sphere lights has a new parameter 'GloOfs', or glow offset.
Setting this to a higher value then 0 will soften the central peak of the glow.
Another unreported bug fix: For xml export, when yafray failed to render the xml file
for some unknown reason, or because of other problems, the export code would still load
the previously rendered image, this causes problems however if the image resolution is
not the same as the current Blender buffer, and so could cause memory corruption or crashes.
This is now taken into account.
World image backgrounds now use the blender mapping settings as well, but only the
'AngMap', 'Sphere' and 'Tube' settings. But in yafray those last two, unlike Blender, cover
the whole view, not just the upper half, so is not really fully compatible with yafray.
So now you have to set one of these buttons too when loading a hdr lightprobe image.
btw, something I forgot to mention in previous commits is that the exposure control using
the texture brightness slider is no longer restricted to integer values. It is now a
floating point value, so you're not restricted to the 0 1 and 2 slider positions anymore,
anything in between will work too.
And finally, display updating is now more like Blender, using the mouse cursor as frame
counter for animation, etc.
