Optimization Tips c# Unity

optimization-tips-snippets.txt

access internal classes, by making your assemble definition named as:
Unity.InternalAPIEditorBridgeDev.001
Unity.InternalAPIEditorBridgeDev.002..

Get full internal access without reflection to any Unity package. Create an assembly reference file and point it to the assembly you want internal access to, now add your scripts into the same folder. The scripts will be compiled as part of the target assembly!
https://twitter.com/KarlJamesJones/status/1523992385071501313


textmeshpro not working, missing files, reinstall doesnt help
- check if some package has TMPRO included, remove that first, then install full TMPRO

point light cookie
- must have alpha (or from grayscale) enabled
- need to adjust intensity once! to update light cookie

manually maximize window
https://forum.unity.com/threads/windowed-mode-goes-under-windows-taskbar.1161893/#post-7451738

TRS translation, rotation, scale
https://forum.unity.com/threads/changing-local-mesh-vertices-into-world-space-coordinates.1160849/#post-7444358

"If you are encoding bytes yourself, try using sbyte[] instead of byte[] (store the same values, for example cast sbyte* to byte* in unsafe block).
The performance problem comes from the fact that byte in C# is unsigned, but in Java it is signed. "
https://forum.unity.com/threads/performance-issues-to-send-byte-from-c-to-java-all-scenarios-tested-are-there-others.991742/#post-6440735

Understand mobile devices are always Vsynced. That means they always wait for the screen refresh to happen before rendering the next frame. The above example may be taking just longer than 16.6ms to render, or the target frame rate is set to 30, it’ll wait until 33.3ms have passed before starting to render again. If it takes just longer than 33.3ms it’ll wait until 66.6ms have passed.
https://forum.unity.com/threads/mobile-graphics-optimization-unplayable-game.987153/#post-6411705

if slow builds: disable mesh compression, happens on every build?
https://www.reddit.com/r/Unity3D/comments/e9gz6h/my_own_unity_report_profiler_mobile_how_i/

burst intrinsics
https://www.youtube.com/watch?v=BpwvXkoFcp8
https://docs.unity3d.com/Packages/[email protected]/manual/index.html#unityburstintrinsics

calculate FPS
1000/targetFPS = time per frame (ms)

internal timer (double):
InputState.currentTime


string concat it faster
https://forum.unity.com/threads/concatenating-strings-speed-test-which-is-faster.924905/

profiler tips, vsync is forced in mobile hw
https://www.youtube.com/watch?v=uXRURWwabF4

"This is why a cheaper alternative to a super high memory GPU would be to use two GPUs, one for the editor and one dedicated to light baking."
https://forum.unity.com/threads/progressive-gpu-lightmapper-preview.561103/page-12#post-5866954

gpu baking
- prioritize view option : not recommend using it for long bakes (high sample count). It has some overhead in the scheduling we do internally and make the overall bake longer
- antialiasing sample count: if you are not using hard shadows on any of your lights I would also recommend reducing the antialisaing sample count for GI more aggressively. In fact, 1spp should be fine in terms of quality if you don't have any hard shadows or if your texel resolution is high enough. Reducing it will reduce in-flight baking memory and thus memory bandwidth on your GPU
https://forum.unity.com/threads/gpu-lightmapper-hardly-uses-gpu.870109/#post-5846482

big structs cause string memcopy!
https://forum.unity.com/threads/string-memcpy-in-profiler.119166/

ar foundation has fast device camera access
https://youtu.be/1a_ik-uXWo4?t=457

faster c# array access
https://twitter.com/SergioPedri/status/1242844697208541189

We have two modes for particle rendering:

    dead particles get culled in vertex shader. This is the behavior you're seeing. This is optimal when particles have few vertices and when the alive count is close to the capacity. This is the default mode.
    Indirect mode, dead particles are culled in compute shader. This is less sensitive to alive/capacity ratio but adds some overhead. Use this when your alive particles count / capacity ratio is low (and particularly if you use mesh particles with many vertices). This mode is forced when GPU sorting is used.

https://forum.unity.com/threads/solved-mesh-output-particles-not-cleared-after-particle-is-dead.848650/#post-5601301

#DOTS #unitytips: The size of data really matters. Converting floats to halfs, ints to bytes etc in my structs made it A LOT faster, especially set/copy jobs. But excessive math operations in half will actually REDUCE perf. Conclusion: Use half for storage, floats for calculation
https://twitter.com/Nothke/status/1235183611214692353

faster editor compile times
https://forum.unity.com/threads/unity-assembly-definition-files-slower-not-faster-how-to-compile-faster.511199/#post-4509778

unity set audio rate lower than 8khz,
edit meta file!

tip, generated audioclip doesnt show audiodata in editor!

unitywebrequest dont use string url
https://forum.unity.com/threads/initial-unitywebrequest-performance-spike.821061/#post-5449386

Yes the mipmap streaming should mean only the lower mips are loaded at first. The 'Max Level Reduction' value specifies how many mips at not loaded at the start. Once it calculates the highest visible mip level for a texture it will then load that desired mip level, if its not already loaded (including all the mips below that level).

There are some requirements for a texture to be streamable which could effect the result.

    Are your textures Read/Write enabled. If they are then all mips are loaded for the CPU side copy.
    Crunch compressed textures are not mip streamable so will load in full.

https://forum.unity.com/threads/mipmap-streaming-asset-bundles-is-the-whole-texture-loaded-from-disk.821199/#post-5437719

sprite.create slow?
dont generate tight sprite, use rectangle, so texture doesnt need to be read
https://forum.unity.com/threads/any-way-to-speed-up-sprite-create.529525/#post-5427861

2D urp,
too many render texture switches
https://forum.unity.com/threads/urp-optimization.818568/#post-5424936

branch prediction and other code quality stuff
https://macton.smugmug.com/Other/2008-07-15-by-Eye-Fi/n-xmKDH/i-Dd6xtNh


comparing of atomics and nativearrays
https://forum.unity.com/threads/very-cool-stuff-in-com-unity-media-utilities.777740/#post-5177366

unity random uses xor-shift

threadsafe no alloc queue
https://forum.unity.com/threads/thread-safe-queue-with-no-allocations.308842/#post-2686958

https://forum.unity.com/threads/trygetlatestimage-performance.726833/#post-4854233
background thread to process pixels from ar texture

// slow to multiply or add vectors, new vector gets created
Vector3 a,p,c;
float, b
p = a*b+c;

// faster
p.x = a.x*b+c.x;
p.y = a.y*b+c.y;
p.z = a.z*b+c.z;


iterating mesh.vertices[] is much*1000 slower than putting verts in array first (vers = mesh.vertices) and iterating that.

note: slow down in one project, if enabled [x] instancing, in sprite shader! 30fps>15fps

getting batching in physics,
"move ALL objects first, before calling any physics stuff like raycast"
https://old.reddit.com/r/Unity3D/comments/a5d85a/unity_physics_best_practice_batching_transform/

https://docs.unity3d.com/uploads/ExpertGuides/Analyzing_your_game_performance_using_Event_Tracing_for_Windows.pdf

gc alloc check
https://docs.unity3d.com/2018.3/Documentation/ScriptReference/TestTools.Constraints.AllocatingGCMemoryConstraint.html


https://lemire.me/blog/2017/09/18/visiting-all-values-in-an-array-exactly-once-in-random-order/

// Vector2 vs Vector3 distance
https://forum.unity.com/threads/vector2-vs-vector3-distance.544206/
public static class VectorHelper
{
    public static float DistanceVector2(Vector2 a, Vector2 b)
    {
        float x = a.x - b.x;
        float y = a.y - b.y;
        return Mathf.Sqrt(x * x + y * y);
    }
}

// can set enum type to byte!
public enum Type : byte { Air, Filled };
https://forum.unity3d.com/threads/i-need-help-optimizing-a-procedurally-generated-chunk.487836/#post-3183778


Structs are value-types like ushorts and enums. There are no extra references. 

https://youtu.be/2A9hAhAQ7V4?t=19866