CMake changes for Visualiser

[Robadob]

Closes #69

These changes enable VISUALISATION flag to be set at CMake time.

This flag pulls in the visualiser repo and marks it as a dependency of FGPU2, configures include dirs and copies SDL.dll on compilation success.

Have tested it with the root CMake project, and added it/tested with example ones too.

This should probably remain draft until the visualiser is ready, and an example has an optional vis.

Two possible remaining tasks:

• Visualiser depends on GLM, FGPU2 will depend on GLM, this could mean the repo is cloned twice independently (and due to size its a fairly slow clone).
• Due to the nature of CMake options, if VISUALISATION is set ON all examples/tests will be setup to have visualiser enabled, even if they don't require it. Probably not a big deal but notable.

[Robadob]

Now have it working with a simple fixed example, I haven't commit these changes due to the large amount of dependency hell I need to cleanup (to save polluting the vis repo).

Brief video demo here: https://youtu.be/-W7fkSgg3IM (poor quality because youtube has downscaled from 1440p to 1080p)

Jobs Todo:

• Removal of non core C/C++ from interface (this will reduce the need to share includes, it's kind of recommended behaviour for libraries)

• Bring the visualiser code upto FGPU2 standards
  • Lint
  • Warnings (vs2015+, also higher warning level)
  • Address the remaining GLM warnings (made a cpp wrapper for the cudaMemcpy call, so glm wasn't in include hierarchy).
• Improve CMake
  • remove include sharing (dependent on interface thingy above)
  • handle freetype (download/build libraries and copy dlls)
  • handle glew (download/build libraries and copy dlls)
  • Allow windows users to provide their own libs, rather than force download
  • Replace glm git clone with .zip download? (should speed up initial configure, i think)

• Handling resources (e.g. models, shader files) [Bundle into exe and extract at runtime if missing, CMake tool for this]
• Replace use of assert/fprintf throughout where necessary
• Make CMake gen separate targets for vis and no vis (update readme to explain this)

Mostly the dependency stuff tied to CMake I want to fix before putting these changes on git.

[ptheywood]

I'm happy to test / fix this on linux once it's worth doing. The SDL installation wiki suggests using the distribution provided copy rather than building your own copy. This should just be looking for SDL2, and if not found but requested building it at that point in time.

As an aside we might want to think about offline builds. I.e. what if you have a local copy of gtest but due to lack of internet the download fails? This is probably worthy of an isse?

[Robadob]

I'm happy to test / fix this on linux once it's worth doing.

I'd wait until the todo list is completed, some of that cleanup should simplify any issues you'll need to address.

This should just be looking for SDL2, and if not found but requested building it at that point in time.

SDL already works like that, I haven't looked into freetype or glew yet.

As an aside we might want to think about offline builds. ...

That's not too impactful, I've already done that for the FGPU2_vis include, so I can dev it not in a weird nested folder.

[Robadob]

It should now be possible to build the visualisation on windows via the main FGPU2 cmake. Running might be a challenge due to the resource paths for shader/model, which are currently hardcoded to paths on my machine.

[Robadob]

Should start build a list of configuration options to add support for: (feel free to edit this comment with suggestions)

P.S. Re the Agent/AgentState, i'm a fan of agent states reverting to agent defaults for any values not set through the state, even if the agent setting is updated after the agent state is created. However this is programatically icky to implement (e.g. an additional bool flag per var, as far as i can tell).

Model/Window Level:

• Window title
• Window dimensions
• Clear color
• Display FPS (Render FPS is not linked to model speed)
• Text color (don't want it to clash with clear color)

• Environment scale
  • Camera start pos
  • Camera target pos
  • Camera move speed
  • Near/far clip
  • It would make sense to have a debug output to screen to help in deciding these values

• Some flag to instead make it 2D/orthographic
• Font for HUD elements?
• An option to force the simulation to execute slower (by blocking the mutex for n-seconds per pass)
• Display step number on HUD (optional like FPS)

Agent/AgentState Level:

• Model file
• Model scale
• Textured models
• User defined model colour schemes
• 2 frame animated models
• TBD other Shader customisation stuff

Misc:

• Log specific env properties to screen
• Add static models (set position, scale, rotation, shader options)
• Add static line art
• Pause sim via keypress

[Robadob]

Started working on agent customisation, model path and scale (1d or 3d) can be chosen. Need to also allow this to be customised on per agent-state basis.

Scale isn't really worth using until env scale options are in place (shader is still hardcoded to multiply agent locs by some constant atm).

[Robadob]

From today's meeting Paul wants a minimal functional version, so that I can move onto submodels.

I think the key things remaining for this are:

• Environment scale
• Replace use of assert/fprintf
• Display step numberf on HUD

Should all be quite trivial to complete
Probably also the Linux/threading bug Pete is chasing (single iteration deallocates agents before they're copied to buffer).

[Robadob]

I'm happy for this to be merged now. Maybe a case for squashing the final few commits. and updating messages on others to clarify visualisation.

Note: Need to update CMakeLists to point to visualisation repo master (once Pete has given Linux a check and the branch has been merged. (Done by Pete in final commit)