Updated TODO.md

Author: MartinOpat

simulation/TODO.md

@@ -1,18 +1,8 @@
 ## To discuss in thesis
-- Instanced rendering not beneficial for current setup
-- research objective and res. question should be in intro; Make a contribution subsection.
-- Describe new approach and how to design experiments (methods section) -> explain  how testing what I did answers R.Q.
-- discussion -> discuss exp., impact, (loading different data, features, interactions, what can be done), future work (tried out things, bottlenecks, load distrib.)
-- Describe required dataset(s) either before methods or very first in methods
-- Reuse abstract and intro from proposal
-- background section can be based on globalAppState-of-the-art but needs updating to be relevant
-- Discuss capped frame-rate due to v-sync and its impact on the measurements in discussion sections
-- 
-- Discuss the discrepancies (see status update email thread)
--
 - Check that the uses of "was done" and "will be done" is consistent and makes sense
 - Make sure that it is properly distinguished between the time step of switching files (a day), and a simulation time step (dt)
 - Make sure that it is clear that fluid simulation and particle simulation is the same thing as we are simulating the fluid as particles (probs. in the intro)
+- Update research question
 
 
 ## Android side
@@ -21,31 +11,16 @@
 ## Native side
 - Delete unused functions
 
-## Current bottlenecks
-For a lot of particles, or randomly distributed particles in a dense grid, the bottleneck is
-the grid interpolation, as a lot of different vertices are being accessed, causing cache misses
-and lowering throughput.
-The compute shaders (best) approach might benefit from a dynamic
-grid-wise parallelization -> Not necessarily.
--> Could simply due to the fact that the particles in thread-groups are far apart in memory -> No simple fix
-- -> Read about warps and global memory
-- -> Discuss the CPU equivalent
-Note that the dense grid is pretty much an issue just for the compute shaders, for the other two modes it is negligable
-
-
 ## Stuff to research
 - Try cuberiles after all
 - Direct volume rendering / volume ray casting
 - The outline box approach
 
 ## General project
-- Sort out references/licenses from used code
 - Make sure to use same case (e.g. camelCase)
 - Try better rendering methods when exatra time, e.g. thesis draft is being review ->volume rendering / volume ray casting, find new ones
 - Mention the different (attempted) different ways of rendering stuff - so far: LIC
-- Make a measurement for the "new bottleneck" see how the performance changes with a more random distrib. (either perlin, or bigger double-gyre)
 
-- Put library credit in the readme.md
 - Add apk as a github release
 
 ## Next meeting points