existing "Equirectangular". This projection is useful to create light probes
from a chrome ball placed in a real scene. It expects as input a photograph of
the chrome ball, cropped so the ball just fits inside the image boundaries.
Example setup with panorama camera and mixing two (poor quality) photographs
from different viewpoints to avoid stretching and hide the photographer:
http://www.pasteall.org/pic/28036
and 5 float image textures. For CPU render this limit will be lifted later
on with image cache support. Patch by Mike Farnsworth.
Also changed color space option in image/environment texture node, to show
options Color and Non-Color Data, instead of sRGB and Linear, this is more
descriptive, and it was not really correct to equate Non-Color Data with
Linear.
- Fixed convex quad detection (in some special cases non-convex quad was detecting as convex)
- Do not add faces with zero area to the output object.
This should resolve#30395: Degenerated triangles from BMesh + Difference
environment map, by enabling the Panorama option in the camera.
http://wiki.blender.org/index.php/Doc:2.6/Manual/Render/Cycles/Camera#Panorama
The focal length or sensor settings are not used, the UI can be tweaked still to
communicate this, also panorama should probably become a proper camera type like
perspective or ortho.
undesired dark rings, and give more accurate lighting when the light is
behind the object. As a bonus, the code is simpler & faster.
Patch by Yasuhiro Fujii, detailed explanation here:
http://mimosa-pudica.net/improved-oren-nayar.html
The changes mostly center around two new structures, InternalNode and
LeafNode. These provide an explicit representation of the Octree
nodes, which formerly were manipulated as opaque byte arrays.
A fair amount of commented out/unused code was also removed. This
includes the "CINDY" code, which may yet be useful, easy to bring back
if so.
There should be no difference in the output of the remesh modifier,
but memory usage may be slightly different. The flood fill bytes are
no longer optional; they will be allocated whether or not the 'remove
disconnect components' flag is set. The leaf node is probably not as
tightly packed due to alignment issues; this could be fixed with the
__attribute__((packed)) flag in gcc (probably there's an MSVC
equivalent), but not sure it's worth it. The internal nodes should
take up less space on 32-bit systems, allocating sizeof(pointer) now
rather than constant eight bytes.
These changes were made in persuit of bug #30158 (remesh crashes on
PowerPC). There's still a fair amount of bitwise stuff in the Octree,
so may still be endian issues and not yet sure if this fixes the bug,
but should be much easier to track down problems now.
made some small edits
- removed changes to AVI reading since the data types are apart of the format spec.
- absf -> abs for a double value in render code.
- Fixed memory lead in Carve_getIntersectedOperandMeshes
- Union manifolds only if they intersects second operand, leave manifolds
which doesn't intersect second operand as-is.
Unioning of intersecting manifold tried to perform as little union operations
as possible, but there were some not totally correct assumption which lead to
cases when unioning of manifolds of some mesh might be happened when one of
mesh sets already had got intersecting manifolds.
This commit corrects this incorrect behavior.
Discovered this when was looking into #30175: Boolean Difference causes 2.62 RC1 crash.
Issue was caused by merging triangles into quads policy which used to think
triangulation of non-planar/non-concave quads happens by 1-3 diagonal which
isn't actually correct in some OpenGL implementations.
Added check for non-concave faces when merging triangles. It will work fine if
original faces are flat. In case if original faces aren't flat this check might
fail and triangulate face when it's not actually needed or merge triangles in
a way which leads to OpenGL artifacts.
very well for colors that can be outside of the 0.0..1.0 range, giving +/- infinity
results.
Now we just use a simple linear contrast factor as proposed by Paolo Sourvinos, and
clamp values to be >= 0, and also make the parameters work more in the 0..1 range
instead of the 0..100 range, to be more consistent with other nodes.
