Alien Gl Renderer

Cloth Simulation Video here!

The cloth simulation source code is []

This project is a sandbox for the for me to improve and learn numerous graphics concepts:

I post my progress here: https://x.com/isacsac2017

Below is a running list of finished, in progress, and cut features, in addition to my ramblings.

Finished features

• Loading glTF Models
• Loading glTF Materials
• GPU Frustum Culling + standalone models
• Debug Frustums on cameras
• Multiple Cameras
• Post-processing
• Instancing
• Indirect Draws using glMultiDrawElementsIndirect
• Multiple Cameras
• ImGui interface
• Tracy Profiler
• Scene loading
• Cloth Simulation

(Fixed) Bugs:

• Materials not loading:

  • a result of unable to create a "buffer of textures"
  • fix: use a texture atlas

• Instancing not working with MultiDrawIndirect?

  • This was caused when I used a struct to contain both the visibleInstIndices and the visibleInstCount.
  • Using sizeof the struct yielded 32 bytes only. sizeof did not account for the fact that there was a vector inside of itself.
  • I just used a single vector like before to hold the count as well (since it's a uint as well).
  • I may change this solution if there is other data (that has a different type) to be accounted for.

• cirno's eyes disappear sometimes... figure out why

  • This was also happening with the cactuar
  • Fragments get discarded because a comparison in the fragment shader keeps firing
    • this is because the alpha cuttoff was greater than the alpha
    • this was caused by the model loading a material out of the range of its material vector
      • out of range exception wasn't thrown because [] operator has undefined behavior for indices out of range
    • so the alpha cutoff was garbage sometimes and 0 most of the time.
    • Fixed by making sure I assume that there is a default material in every model.

• Indirect Draw Debugging

  • First symptom: only one model is drawn
    • each instance is probably using the same model matrix, this was because I was using gl_DrawID and DrawElementsIndirect, so glDrawID was always 0
  • Second symptom: meshes with multiple primitives are not drawn correctly.
    • check the draw calls in renderdoc.
      • seems that the baseVertex and firstIndex members in the IndirectDrawCommand struct s were wrong (I flipped them on the CPU side)
    • even after flipping them something was wrong...
      • after checking the values in the node primitives properties struct in renderdoc, it seems that the struct's alignment was wrong. this is because I used alignas(16) in the struct and the struct only had 2 4 byte members
      • remember, the std430 layout has more flexible rules on struct alignment than std140

Capstone TODOs:

• Implement an L system
• Add concepts in animation
  • Mass spring dampers
  • Particle System
  • Splines
  • Inverse kinematics
• Cloth Simulation
  • Try your best ot make it real time
• Particle Systems
• Animation

TODO

• Abstract organize the logic better

• A renderpass system

• lighting

• baked lighting + instancing... damn im gonna regret this aint i

• Get rid of include clutter

• shadows

• reflections

• animation

• Integrating OpenUSD

• multidraws/indirect draws/culling that actually reduces compute

  • a draw a model instance with 1 draw call
    • not possible thats not what multidraw means
    • instead batch the draw calls (every N is a new instance)
    • we can only really do it this way since if i were to have instance count be greater than one in the draw command, i would need to move around amtrices into a new buffer (inefficient...)
    • the draw command probably needs the index of the model though
  • Steps:
    • 1. Modify the renderer to use indirect draws, DrawElementsIndirect.
      • the glviewer example code has a way to set this up per mesh.
      • very intuitive code, every primitive begins with a draw call
      • a mesh contains a buffer of primitives
      • DrawElementsIndirect is called on the location in the buffer (takes the 20 bytes from that pointer)
    • 2. Aggregate all the primitive's vertices and indices into 1 giant buffer
      • make sure i can render indirectly like that
    • 3. Use MultiDrawElementsIndirect instead, this time to draw everything in that buffer
    • 4. Modify the compute shader so that it modifies the draw out cmd buffer
    • 5. Modify the shaders so that it uses materials and textures (using an index to query them from ssbos);

• deferred shading

• add doxygen

• Craig Reynold's Boids

• Realistic Ocean Waves with help from GPU Gems and this video

• Improve the waves. I made it so that the light can be seen underwater, but I'm not exactly sure if it's correct.

• A new boid behavior? Not sure what I should do.

• A debug interface. I should play around with compute shaders so that I can see the normals of the objects I render. Not sure how to implement the window though.

• Sounds (they do improve the experience...)

Bugs:

• fix noninstanced model shader (im lazy so i dwanna right now)

CUT:

• Make sure to delete all buffers after exiting // eeeeeeehhh? There's no point if memory is not in demand in the lifetime. The program will release all its resources upon exit anyway! My excuse for allowing memory leaks lol!
• Use git submodules? Debatable Don't wanna mess with this, in production code, using public modules is tied to legality issues,

But yea, I hope I can continue playing around with this!