There are several internal changes for this:
First idea is to make __tri_verts to behave similar to __tri_storage,
meaning, __tri_verts array now contains all vertices of all triangles
instead of just mesh vertices. This saves some lookup when reading
triangle coordinates in functions like triangle_normal().
In order to make it efficient needed to store global triangle offset
somewhere. So no __tri_vindex.w contains a global triangle index which
can be used to read triangle vertices.
Additionally, the order of vertices in that array is aligned with
primitives from BVH. This is needed to keep cache as much coherent as
possible for BVH traversal. This causes some extra tricks needed to
fill the array in and deal with True Displacement but those trickery
is fully required to prevent noticeable slowdown.
Next idea was to use this __tri_verts instead of __tri_storage in
intersection code. Unfortunately, this is quite tricky to do without
noticeable speed loss. Mainly this loss is caused by extra lookup
happening to access vertex coordinate.
Fortunately, tricks here and there (i,e, some types changes to avoid
casts which are not really coming for free) reduces those losses to
an acceptable level. So now they are within couple of percent only,
On a positive site we've achieved:
- Few percent of memory save with triangle-only scenes. Actual save
in this case is close to size of all vertices.
On a more fine-subdivided scenes this benefit might become more
obvious.
- Huge memory save of hairy scenes. For example, on koro.blend
there is about 20% memory save. Similar figure for bunny.blend.
This memory save was the main goal of this commit to move forward
with Hair BVH which required more memory per BVH node. So while
this sounds exciting, this memory optimization will become invisible
by upcoming Hair BVH work.
But again on a positive side, we can add an option to NOT use Hair
BVH and then we'll have same-ish render times as we've got currently
but will have this 20% memory benefit on hairy scenes.
This commit introduces a SSS-oriented intersection structure which is replacing
old logic of having separate arrays for just intersections and shader data and
encapsulates all the data needed for SSS evaluation.
This giver a huge stack memory saving on GPU. In own experiments it gave 25%
memory usage reduction on GTX560Ti (722MB vs. 946MB).
Unfortunately, this gave some performance loss of 20% which only happens on GPU.
This is perhaps due to different memory access pattern. Will be solved in the
future, hopefully.
Famous saying: won in memory - lost in time (which is also valid in other way
around).
This commit contains all the work related on the AMD megakernel split work
which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus
some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely
someone else which we're forgetting to mention.
Currently only AMD cards are enabled for the new split kernel, but it is
possible to force split opencl kernel to be used by setting the following
environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1.
Not all the features are supported yet, and that being said no motion blur,
camera blur, SSS and volumetrics for now. Also transparent shadows are
disabled on AMD device because of some compiler bug.
This kernel is also only implements regular path tracing and supporting
branched one will take a bit. Branched path tracing is exposed to the
interface still, which is a bit misleading and will be hidden there soon.
More feature will be enabled once they're ported to the split kernel and
tested.
Neither regular CPU nor CUDA has any difference, they're generating the
same exact code, which means no regressions/improvements there.
Based on the research paper:
https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf
Here's the documentation:
https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit
Design discussion of the patch:
https://developer.blender.org/T44197
Differential Revision: https://developer.blender.org/D1200
From more investigation of the numeric failures in the kernel it appears
the check was rather correct. But in theory it;s also needed for the motion
triangles.
tri_shader does no longer need to a float.
Reviewers: dingto, sergey
Reviewed By: dingto, sergey
Subscribers: dingto
Projects: #cycles
Differential Revision: https://developer.blender.org/D789
Root of the issue goes back to the on-fly normals commit and the
latest fix for it wasn't actually correct. I've mixed two fixes
in there.
So the idea here goes back to storing negative scaled object flag
and flip runtime-calculated normal if this flag is set, which is
pretty much the same as the original fix for the issue from me.
The issue with motion blur wasn't caused by the rumtime normals
patch and it had issues before, because it already did runtime
normals calculation. Now made it so motion triangles takes the
negative scale flag into account.
This actually makes code more clean imo and avoids rather confusing
flipping code in mesh.cpp.
Instead of pre-calculation and storage, we now calculate the face normal during render.
This gives a small slowdown (~1%) but decreases memory usage, which is especially important for GPUs,
where you have limited VRAM.
Part of my GSoC 2014.