Blender Git Loki

Make OpenSubdiv working on a low-end video cards

This commit changes the way how mesh is being displayed when
using OpenSubdiv backend. It does no longer use neither effect
factories nor fragment shaders in order to make all the shaders
compatible with at least GLSL 1.3 compatible cards.

In fact, seems the shaders now goes down to GLSL 1.1 even.

Apart from this now the first material from the material stack
is used for colors.

Still requires our own OpenSubdiv branch:
https://github.com/Nazg-Gul/OpenSubdiv/tree/cuda-dynload

(it actually needs updated version of this branch)

Switch to a much more simplified shader

Useful to make it a base point for making stuff working on
the Intel GPU i'm having here.

Code cleanup: use Blender code style

Fix for assert failure in CUDA when using Nvidia Optimus

Merge branch 'master' into opensubdiv-modifier

Conflicts:
source/creator/CMakeLists.txt

Update CUEW in order to solve compilation error on Windows

Avoid some really bad memory usage

- Need to count DM primitives properly in order
to avoid bad memor write in ORCO DM.

- Skip GPU if subsurf result will be converted to CDDM.

Material and textured viewport is not quite working yet,
but at least it doesn't crash.

Solve crash when doing viewport render

The issue was caused by the render pipeline creating
DerivedMesh for purposes of it's export to the render
database. In this case SubSurf modifier shouldn't be
applied on GPU because render pipeline would want to
access hires data from the CPU side.

Added a flag to modifier apply bitmask which indicates
that the result could be stored in the GPU side. The
flag is called MOD_APPLY_ALLOW_GPU. If this flag is
set modifiers are allowed to keep their result on GPU.

We still need to implement CPU-side hi-res mesh unpack
even if it was originally evaluated on GPU, but the
current change is useful anyway because it'll avoid
possible overhead of copying data back-n-forth.

Merge branch 'master' into opensubdiv-modifier

Use updated version of CUEW

This wrangler is now automatically generated from
cuda.h and cudaGL.h which is really handy from the
maintenance point of view: with new CUDA toolkit
release it should be rather simple to update the
wrangler.

Currently it's generated from CUDA toolkit 5 since
it's the version which comes in Debian Wheezy.

Also bundling the generator scripts, so others
can tweak/improve it.

Works fine on Linux, didn't verify it on Windows.

Merge branch 'master' into opensubdiv-modifier

Merge branch 'master' into opensubdiv-modifier

Conflicts:
source/blender/blenkernel/intern/subsurf_ccg.c
source/blender/windowmanager/intern/wm_init_exit.c

Revert "Update Ceres to latest upstream version"

This reverts commit 39f575106d0fc3e2142bf79132a24a90fc399608.

Was just testing WIP patch, didn't mean to commit
it yet. Still need a bit of cleanup.

Compilation fixes on Windows

Now both scons and cmake are tested building
with msvc2013 64bit.

Merge branch 'opensubdiv-modifier' of git.blender.org:blender into opensubdiv-modifier

Fix crash when quitting blender without OpenCL initialized

Bring new OpenSubdiv with CUEW support

WARNING: Latest fork of OpenSubdiv is to be used to be able to compile
this revision:

https://github.com/Nazg-Gul/OpenSubdiv/tree/cuda-dynload

We'll work on pull-request when our branch would be proved to work on
Linux (which seems to work already) and Windows (need to test this,
that's why commit is happening now -- stashing changes, creating a
milestone, easier access from Windows!).

NOTE: Would ask platform maintainers to not bother trying to make this
revision to work yet. Lemme test things on Windows first and it it
all work i'll prepare libs for windows and do a pull-request to an
upstream.

Main idea is to make OpenSubdiv to detect OpenCL and CUDA on runtime.
Before this Blender was linking against libOpenCL and libcuda which
is not good for portability.

OpenCL part was already kind of done, but wasn't properly integrated.
Not sure why it compiled even :)

CUDA is now does the same trick as Cycles does with dynamic load of
cuda driver api library.

That would have been real straightforward change if we wouldn't try
to de-duplicate CUDA/CL wrappers in the Blender sources. The thing
is, Cycles, Compositor and OpenSubdiv all requires wrappers for this
libraries and it was real stupid to have two clew (for Cycles and
Compositor) already.

The idea to solve this is: move clew/cuew to extern/ and make all
the bits which requires cuda/cl to use those wrappers instead of
shipping their own. This is real straightforward change to make
Cycles and Compositor happy since it's easy to re-arrange libraries
we're maintaining in the sources. Making it so OpenSubdiv libraries
can use symbols from our extern_{clew,cuew} was rather a black
magic but it appeared to work pretty well here on Linux. Let's
see how it works on Windows now.

TODO:
- Make sure Windows works
- Remove cuda/cl wrappers from intern/cycles
- Make sure Cycles stadalone works

As for Cycles standalone it's real easy to make it compiled when
building from inside Blender. When/if we'll move Cycles to own
repo it's still easy to make it use clew/cuew by putting them to
extern/ or third_party/ folder in Cycles repo.

Merge remote-tracking branch 'origin/master' into opensubdiv-modifier

OSX: fix compile after clew rework, include fix and whitespace cleanup

Correction to previous commit