In rB78c491e the `initialize_particle` function was split into 2 parts for particle texture initialization.
The texture init part however also initializes birth times, which is now missing in the main init function
in some cases (notably when setting start/end directly without a subsequent time step).
in threaded depgraph updates and effector list construction.
Gathering effectors during depgraph updates will call the
psys_check_enabled function. This in turn contained a DNA alloc call
for the psys->frand RNG arrays, which is really bad because data must be
immutable during these effector constructions.
To avoid such allocs the frand array is now global for all particle
systems. To avoid correlation of pseudo-random numbers the psys->seed
value is complemented with random offset and multiplier for the actual
float array. This is not ideal, but work sufficiently well (given that
random numbers were already really limited and show repetition quite
easily for particle counts > PSYS_FRAND_COUNT).
Root of the issue goes to the order of particle initialization which does
texture evaluation (which does depend on particle coordinate) and particle
birth coordinate calculation. So basically what happened is:
* Changing number of particles re-allocated all the particles,
which sets their coordinate to (0,0,0)
* Texture evaluation used this non-initialized coordinate
* Coordinates were calculated for particles
Reshuffled code a bit so now texture evaluation happens after particles.
coordinate calculation. Basically moved texture evaluation to particle
reset function. Reset happens after initialization anyway and it does
know particle coordinates. Also, if reset is being called without init
then it's also kind of logical to re-evaluate texture because particle
coordinates might change.
Lattice deformation used to store some runtime data
inside of lattice datablock itself. It's something
which is REALLY bad. Ideally DNA shouldn't contain
and runtime data.
For now solved it in a way that initialization of
lattice deform will create a structure which contains
lattice object for which deformation is calculating
and that runtime data which used to be stored in
lattice datablock itself.
It works really fine for mesh deform modifier, but
there's still runtime data stored in particle system
DNA, It didn't look something easy to be solved, so
leaving this as-is for now.
--
svn merge -r58277:58278 -r58795:58796 ^/branches/soc-2013-depsgraph_mt
rendering. This used to happen in an unneeded frame change update which was
removed. For heavy particle systems this could have a bad impact on viewport
performance after rendering.
generator with a local one. It's not thread safe and will not give repeatable
results, so in most cases it should not be used.
Also fixes#34992 where the noise texture of a displacement modifier was not
properly random in opengl animation render, because the seed got reset to a
fixed value by an unrelated function while for final render it changed each
frame.
This was caused by a floating point precision error. During collision detection, Newton-Raphson iteration is used to find the exact time of the collision. But when using subframes, the initial Newton step was too small. Now the initial step is given in absolute units. When subframes = 0, this should behave almost the same as before.
Thanks to Janne Karhu, Lukas Toenne and Ton Roosendaal for their help with this patch, and to AutoCRC for funding.
http://projects.blender.org/tracker/index.php?func=detail&aid=29681&group_id=9&atid=127
The solver was mostly implemented by John Mansour at VPAC, with help from me and with funding from the AutoCRC. The SPH formulation is due to Gingold and Monaghan, and the smoothing kernel is due to Wendland.
This solver does not replace the old one; it is available as an option. Note that the new solver uses different units than the old one. The patch page has a couple of attachments that can be used to test the new solver, particularly sphclassical_dam_s0.01_grav.blend (ignore the earlier tests). The simulation in that file compares well with a physical experimental dam break; details in a paper by Changhong Hu and Makoto Sueyoshi, also referred to on that page.
- define array sizes for functions that take vectors.
- quiet some -Wshadow warnings.
- some copy/paste error in readfile.c made it set the same particle recalc flag twice.
In addition the billboards can be scaled by the particle velocity with optional head and tail factors (similar to line drawing options). This allows for pseudo-motionblur effects.
* The old collisions code detected particle collisions by calculating the
collision times analytically from the collision mesh faces. This was
pretty accurate, but didn't support rotating/deforming faces at all, as
the equations for these quickly become quite nasty.
* The new code uses a simple "distance to plane/edge/vert" function and
iterates this with the Newton-Rhapson method to find the closest particle
distance during a simulation step.
* The advantage in this is that the collision object can now move, rotate,
scale or even deform freely and collisions are still detected reliably.
* For some extreme movements the calculation errors could stack up so much
that the detection fails, but this can be easily fixed by increasing the
particle size or simulation substeps.
* As a side note the algorithm doesn't really do point particles anymore,
but uses a very small radius as the particle size when "size deflect" isn't
selected.
* I've also updated the collision response code a bit, so now the particles
shouldn't leak even from tight corners.
All in all the collisions code is now much cleaner and more robust than before!
* Grid distribution isn't really suited for hair, so this is now disabled.
* Setting a jittered distribution with particles/face = 1 now creates particles on the center of faces.
* Quite a bit of cleanup of the whole particle distribution code.
* Renamed the old split uv's animate option "time" to "age" and added a new option to change the used split frame by frame. These are good changes were suggested/implied by Hannu Hoffren over 3 years ago in his tutorial video! (oh my how time flies)
* Also cleaned up the billboard ui quite a bit. For example now the uv channels can be properly chosen from the existing channels.
* Effecting particle properties with textures was possible in 2.49,
but not in 2.5 anymore.
* Now particles have their own textures (available in texture panel
for objects with particle systems), which are totally separate from
the material textures.
* Currently a basic set of particle properties is available for
texture control. Some others could still be added, but the whole
system is not intended as an "change anything with a texture" as
this kind of functionality will be provided with node particles in
the future much better.
* Combined with the previously added "particle texture coordinates"
this new functionality also solves the problem of animating particle
properties through the particle lifetime nicely.
* Currently the textures only use the intensity of the texture in
"multiply" blending mode, so in order for the textures to effect
a particle parameter there has to be a non-zero value defined for
the parameter in the particle settings. Other blend modes can be
added later if they're considered useful enough.
* The SPH fluid particle algorithm was implemented a bit wrong. This problem could for example result in the fluid moving sideways after being dropped straight to a horizontal collision surface, a very big no-no as far as real world physics are concerned!
* After some extensive code shuffling the algorithm is now much more true to the paper it was implemented from, and more importantly now the physics should be correct too!
* The main thing was that fluids calculations can effect many particles simultaneously, so just a single loop through all particles can't work properly. As a side note this also means that the actual fluid algorithm can't be made threaded :(
* To make things work I also had to reshuffle some general particle physics code, but there should be no functional changes what so ever to other physics types, so poke me immediately if something strange happens.
Note to users: these changes will most probably effect the way previously done sph fluid simulations look, so some parameter tweaking will be needed to get things back looking the way they were.
Also use const char in many other parts of blenders code.
Currently this gives warnings for setting operator id, label and description since these are an exception and allocated beforehand.
* This caused nearly all particles to leak through the collision surface if simulation subframes were used and the collision object was moving.
* In addition to fixing this I also did some more cleanup of the collision code and refined some of the comments.
* Although not strictly a bug it is the expected behavior and won't mess anything else up.
* Note: the lattice is applied to the actual hair keys instead of the calculated strands so the applied result will differ a bit from the original.
This commit and other commits attempting to fix it broke various things. The
main thing that changed was that instead of computing children/paths in
advance as part of particle_system_update, this was moved to do it just before
drawing or rendering. I've changed back that behavior and tried to keep the
other fixes in the commit.
When the new particle system was just committed, it also worked this way but
gave various problems, and I had to remove that behavior to get things working
stable. Basically it meant that you could get have a path cache that was
outdated in various situations, and it doesn't fit well with dependency graph
evaluation order.
This fixes:
#22823: Children Particle Rendering is broken
#22733: Particle objects not displayed
#22888: SigSegV when rending hair particles
#22820: Another SigSegV when undo adding hairs in particel edit mode
Some particle setups in dupligroups.
The three bugs that the original commit fixed are now also still working in
my tests:
#21316: Hair weight drawing is wrong
#21923: Consistent Crash When Rendering Particle Scene.
#21950: Path rendering option for particles causes crash
* Path drawing now works for non hair particles.
* Should fix the following bugs too:
[#21316] Hair weight drawing is wrong
[#21923] Consistent Crash When Rendering Particle Scene.
[#21950] Path rendering option for particles causes crash
