Fix particle effect randomness

[iche033]

Signed-off-by: Ian Chen ichen@osrfoundation.org

🦟 Bug fix

Summary

The randomness algorithm in the particle effect for has issue when time (which is used to add randomness) becomes large in the shaders. The fix is to just use the fractional part of the time value for adding randomness.

To Test

To reproduce the problem, one way to test this is to artificially increase the time value in the shaders. You can do this by applying the patch below to this branch to simulate the old behavior when time is large:

diff --git a/ogre2/src/media/materials/programs/depth_camera_fs.glsl b/ogre2/src/media/materials/programs/depth_camera_fs.glsl
index 0d3109d5..a22bd4b3 100644
--- a/ogre2/src/media/materials/programs/depth_camera_fs.glsl
+++ b/ogre2/src/media/materials/programs/depth_camera_fs.glsl
@@ -107,7 +107,7 @@ void main()
   if (particle.x > 0 && particleDepth < fDepth)
   {
     // apply scatter effect so that only some of the smoke pixels are visible
-    float r = rand(inPs.uv0 + vec2(rnd, rnd));
+    float r = rand(inPs.uv0 + vec2(rnd + 1e5, rnd + 1e5));
     if (r < particleScatterRatio)
     {
       // set point to 3d pos of particle pixel

Then run this example world:
https://gist.github.com/iche033/bcd3b7d3f4874e1e707e392d6dbb0aa0

The patch adds 100000 seconds (27.x hours) to time and you'll see bands across the rgbd camera :

The changes in this PR makes sure the random number never exceeds 1 and the noise pattern should stay distributed across the image:

Checklist

• Signed all commits for DCO
• Added tests
• Updated documentation (as needed)
• Updated migration guide (as needed)
• codecheck passed (See contributing)
• All tests passed (See test coverage)
• While waiting for a review on your PR, please help review another open pull request to support the maintainers

Note to maintainers: Remember to use Squash-Merge

[codecov]

Codecov Report

Merging #388 (caacd56) into ign-rendering4 (a3679f4) will increase coverage by 0.00%.
The diff coverage is 100.00%.

❗ Current head caacd56 differs from pull request most recent head 9a9a8dc. Consider uploading reports for the commit 9a9a8dc to get more accurate results

@@               Coverage Diff               @@
##           ign-rendering4     #388   +/-   ##
===============================================
  Coverage           55.91%   55.91%           
===============================================
  Files                 147      147           
  Lines               13981    13983    +2     
===============================================
+ Hits                 7817     7819    +2     
  Misses               6164     6164           

Impacted Files	Coverage Δ
ogre2/src/Ogre2ParticleNoiseListener.cc	89.47% <100.00%> (+0.38%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a3679f4...9a9a8dc. Read the comment docs.

[darksylinc]

I knew there was something amiss with that line!

Your solution is a temporary workaround though. Once time becomes large enough, the fractional part becomes always 0 (and before that, it will always be 0 and 0.5, and before that...).

It's easier to mentally visualize it with half floating point format

A better solution would be to fix it from C++ side of things.

If you're using Ogre's auto param, IIRC there was an option to make it periodic.

Update: it's time_0_x 1
Update 2 ugh Ogre makes the same mistake of doing mod in floating point

[iche033]

A better solution would be to fix it from C++ side of things.

yeah you're right, once time becomes too large, the randomness goes away. I took the C++ approach and passed a random number to the shaders. 2690ca2