This patch is the root of the GHash rework, all other diff will be based on it:
Reduce average load from 3.0 to 0.75
----------------------------------
This is the big performance booster part, e.g. makes tracing a dyntopo stroke between 25% and 30% faster.
Not much to say about it, aside that it obviously increase memory footprint (about 25% - 30% too).
Add optional shrinking
----------------------------------
I.e. ghashes/gsets can now shrink their buckets array when you remove enough entries. This remains optional and OFF by default.
Add code to use masking instead of modulo
----------------------------------
Buckets indices are obtained from hashes by “reducing” the hash value into the valid bucket range. This can be done either by bit-masking, or using modulo operation.
The former is quicker, but requires real hashes, while the later is slower (average 10% impact on ghash operations) but can also be used as a 'fake' hashing on raw values, like e.g. indices.
In Blender currently not all ghash usages actually hash their keys, so we stick to modulo for now (masking is ifdef’ed out), we may however investigate the benefits of switching to masking with systematic very basic hashing later…
Add various missing API helpers
----------------------------------
I.e. a way to deep-copy a ghash/gset, and a way to (re-)reserve entries (i.e. manually grow or shrink the ghash after its creation).
Various code refactoring
----------------------------------
* Get rid of the 'hack' regarding ghash size when used as gset (it’s simpler and safer to have two structs defined here, and cast pointers as needed).
* Various re-shuffle and factorization in low-level internal code.
* Some work on hashing helpers, introducing some murmur2a-based hashing too.
Thanks a bunch to Campbell for the extensive review work. :)
Reviewers: sergey, campbellbarton
Subscribers: psy-fi, lukastoenne
Projects: #bf_blender
Maniphest Tasks: T43766
Differential Revision: https://developer.blender.org/D1178
This commit introduces a few ready made effects for the 3D viewport
and OpenGL rendering.
Included effects are Depth of Field, accessible from camera view
and screen space ambient occlusion. Those effects can be turned on and
tweaked from the shading panel in the 3D viewport.
Off screen rendering will use the settings of the current camera.
WIP documentation can be found here:
http://wiki.blender.org/index.php/User:Psy-Fi/Framebuffer_Post-processing
Summary: Just some extra comparison between native and 32bit structure length was needed.
Reviewers: brecht, campbellbarton
Reviewed By: brecht
Differential Revision: http://developer.blender.org/D131
Summary:
This commit removes ~10000 strlen calls and ~100000 memcpy calls in blender (profiled with blender --background), ~10000 memcpy calls in makesdna.
There is no need to create null-terminated strings for atoi, because it converts only the initial portion of the string anyway.
Also it was noticed that DNA_elem_array_size and arraysize functions work only with full strings, so there is no point to calculate strlen.
Reviewers: campbellbarton
Reviewed By: campbellbarton
Differential Revision: http://developer.blender.org/D105
Release builds will now use lock-free allocator by
default without any internal locks happening.
MemHead is also reduces to as minimum as it's possible.
It still need to be size_t stored in a MemHead in order
to make us keep track on memory we're requesting from
the system, not memory which system is allocating. This
is probably also faster than using a malloc's usable
size function.
Lock-free guarded allocator will say you whether all
the blocks were freed, but wouldn't give you a list
of unfreed blocks list. To have such a list use a
--debug or --debug-memory command line arguments.
Debug builds does have the same behavior as release
builds. This is so tools like valgrind are not
screwed up by guarded allocator as they're currently
are.
--
svn merge -r59941:59942 -r60072:60073 -r60093:60094 \
-r60095:60096 ^/branches/soc-2013-depsgraph_mt
Instead of having ifdef __GNUC__ all over the headers
to use special compiler's hints use a special file where
all things like this are concentrated.
Makes code easier to follow and allows to manage special
attributes in more efficient way.
Thanks Campbell for review!
- Re-arrange locks, so no actual memory allocation
(which is relatively slow) happens from inside
the lock. operation system will take care of locks
which might be needed there on it's own.
- Use spin lock instead of mutex, since it's just
list operations happens from inside lock, no need
in mutex here.
- Use atomic operations for memory in use and total
used blocks counters.
This makes guarded allocator almost the same speed
as non-guarded one in files from Tube project.
There're still MemHead/MemTail overhead which might
be bad for CPU cache utilization
- Revert some stylistic changes made in branch
They weren't directly related on freestyle project
and better be handled in trunk. Helps reading patches
ad prevents possible merge conflicts.
- Solved issue introduced with recent deprecation of
RenderData.maximize,
and RNA for it independent of the build flag for enabling Freestyle. Suggested
by Sergey Sharybin through a code review of the branch.
* Many #ifdef WITH_FREESTYLE blocks were removed to always have Freestyle-specific
DNA file specification and RNA for it built in Blender. This will allow Freestyle
setting survive even when a non-Freestyle build is used for loading and saving
files. It is noted that operations are still conditionally built through #ifdef
WITH_FREESTYLE blocks.
* To this end, new blenkernel files BKE_freestyle.h and intern/freestyle.c have
been added. All API functions in FRS_freestyle_config.h as well as some of those
in FRS_freestyle.h were moved to the new files. Now the relocated API functions
have BKE_ prefix instead of FRS_.
Resolved conflicts:
release/datafiles/startup.blend
source/blender/editors/space_nla/nla_buttons.c
Also updated source/blender/blenkernel/intern/linestyle.c as a follow-up of
recent changes for the use of bool.
Automatic switching for drawing pixel buffers via glDrawPixels or using GPU textures
It works with a User Preference limit, in megapixels, to define whether to use
GPU or direct pixel drawing. Default is now initialized to 10 MP (4k buffers).
Especially for zooming out (draw smaller) texture drawing is much smaller. Also
Nvidia cards typically draw much faster with textures in general.
Added to node backdrop first now, the other editors follow in a next commit.
For coders: added new DNA function to initialize new struct variables, so you
don't have to sub-version files anymore.
DNA_struct_elem_find(fd->filesdna, "structname", "typename", "varname")
"filesdna" is the sdna description of the current file being versioned.
Scons used to fail because of couple of reasons:
- CMake does have WITH_DNA_GHASH enabled, Scons not
Which is a good question why?!
- This makes Cmake including BLI_utildefines in dna_genfile.c,
but scons missed this include
- This damn "bool" thing requires BLI_utildefines
And last question here is why would we want to use bool here =\
It's just always ends up in headache without any visible benefit
of using it.
Well, at least now svn shall be compilable again..
This is just the basic structure, the simulation isn't hooked up yet.
Scenes get a pointer to a rigid body world that holds rigid body objects.
Objects get a pointer to a rigdid body object.
Both rigid body world and objects aren't used directly in the simulation
and only hold information to create the actual physics objects.
Physics objects are created when rigid body objects are validated.
In order to keep blender and bullet objects in sync care has to be taken
to either call appropriate set functions or flag objects for validation.
Part of GSoC 2010 and 2012.
Authors: Joshua Leung (aligorith), Sergej Reich (sergof)
