Skip to content

Module: SVO Voxelizer

Jump to bottom
ninlilizi edited this page Apr 23, 2020 · 16 revisions

Features

  • Multi-threaded.
  • Asynchronous GPU operations.
  • Progressive tree construction with priority-based triage.
  • Progressive mip filtering, organised into waves for efficiency.
  • Performant: Less than 4% GPU utilization on an RTX2080S.
  • Auto-CPU usage scaling. Won't get in the way of your game logic.
  • Easy to integrate, instantiable object.

Usage

Instantiation

// Instantiate SVO Tree
SVO = ScriptableObject.CreateInstance<NKLI.Nigiri.SVO.Tree>();
SVO.Create(attached_camera, max_depth, max_nodes);

// Instantiate voxelizer
voxelizer = new NKLI.Nigiri.SVO.Voxelizer(SVO, emmissive_intensity, shadow_strength, occlusion_gain, gi_area_world_size);

Destruction

// Dispose voxelizer
voxelizer.Dispose();

// Destroy SVO
NKLI.Nigiri.Helpers.DestroyScriptableObject(ref SVO);

Voxelize scene

// Update voxelization parameters
voxelizer.UpdateParameters(emmissive_intensity, shadow_strength, occlusion_gain);

// Voxelize scene
voxelizer.VoxelizeScene(_sample_count, position_texture, lighting_texture, mask_buffer);

// Split nodes
voxelizer.SplitNodes();

// Mipmap nodes
voxelizer.MipmapNodes();

Variables

attached_camera - Main camera
max_depth - Default max depth of the tree
max_nodes - Maximum nodes to store [Calculate VRAM as: (max_nodes * 16Bytes)]
emmissive_intensity - Emissive lighting intensity
shadow_strength - Strength of shadows
occlusion_gain - Ambiant occlusion gain
gi_area_world_size - World space axis length of GI area

Properties (Read-Only)

NKLI.Nigiri.SVO.Tree.MaxDepth - Default, max tree depth
NKLI.Nigiri.SVO.Tree.RAM_Usage - Estimation of allocated RAM
NKLI.Nigiri.SVO.Tree.VRAM_Usage - Estimation of allocated VRAM
NKLI.Nigiri.SVO.Tree.Runtime_Thread_Split - Execution time of node split worker thread
NKLI.Nigiri.SVO.Tree.Runtime_Thread_Mipmap - Execution time of node mipmap worker thread
NKLI.Nigiri.SVO.Tree.Buffer_SVO_ByteLength - Byte length of SVO buffer
NKLI.Nigiri.SVO.Tree.SplitQueueMaxLength - Current size of split queue buffer
NKLI.Nigiri.SVO.Tree.MipmapQueueMaxLength - Current size of the mipmap queue buffer
NKLI.Nigiri.SVO.Tree.SplitQueueSparseCount - Number of node splits waiting to be processed
NKLI.Nigiri.SVO.Tree.MipmapQueueSparseCount - Number of mipmaps waiting to be processed
NKLI.Nigiri.SVO.Tree.AbleToSplit - Are there nodes waiting to be processed?
NKLI.Nigiri.SVO.Tree.AbleToMipmap - Are there mipmaps waiting to be processed?
NKLI.Nigiri.SVO.Voxelizer.SVO_Tree - Bound SVO Tree
NKLI.Nigiri.SVO.Voxelizer.Emissive_Intensity - Emissive lighting intensity
NKLI.Nigiri.SVO.Voxelizer.Shadow_Strength - Strength of shadows
NKLI.Nigiri.SVO.Voxelizer.Occlusion_Gain - Ambiant occlusion gain
NKLI.Nigiri.SVO.Voxelizer.GI_Area_Size - World space axis length of GI area

Properties (Read-Write)

NKLI.Nigiri.SVO.Voxelizer.Debug_Filtering - Enable debug visualisation of mipmap filtering

API

NKLI.Nigiri.SVO.Voxelizer.UpdateParameters(emissiveIntensity, shadowStrength, occlusionGain)

  • Updates voxelisation parameters

NKLI.Nigiri.SVO.Voxelizer.VoxelizeScene(sampleCount, positionTexture, lightingTexture, maskBuffer)

  • Voxelise scene using the input render textures and mask buffer

NKLI.Nigiri.SVO.Voxelizer.SplitNodes()

  • Processes the node split queue

NKLI.Nigiri.SVO.Voxelizer.MipmapNodes()

  • Processes the node mipmap queue

NKLI.Nigiri.SVO.Tree.SetSplitQueue(_splitQueue)

  • Submit custom split queue
    _splitQueue - Byte array of uint32 offset values. Position zero denotes number of items
    (Size of array must equal NKLI.Nigiri.SVO.Tree.SplitQueueMaxLength)

NKLI.Nigiri.SVO.Tree.SetMipMapQueue(_mipmapQueue)

  • Submit custom mipmap queue
    _mipmapQueue - Byte array of uint32 offset values. Position zero denotes number of items
    (Size of array must equal NKLI.Nigiri.SVO.Tree.MipmapQueueMaxLength)

NKLI.Nigiri.SVO.Tree.ResumeMipmapWorker()

  • Mipmap worker thread automatically suspends when it's finished processing a queue to
    allow for it's output to be processed by the GPU. Calling this function will signal
    the thread to look for a new queue to repeat the cycle.
    (This is normally handled automatically if using the attached voxelization function)

NKLI.Nigiri.SVO.Tree.SuspendWorkers()

  • Suspend all worker threads
    This is useful if you want to freeze the SVO for a period to save GPU time
    NOTE: Calling NKLI.Nigiri.SVO.Voxelizer.VoxelizeScene() will unfreeze workers automatically.

NKLI.Nigiri.SVO.Tree.ResumeWorkers()

  • Resume all worker threads
    This is useful if you want to freeze the SVO for a period to save GPU time
    NOTE: Calling NKLI.Nigiri.SVO.Voxelizer.VoxelizeScene() will unfreeze workers automatically.

Resources

Required files

C#

Scripts/NKLI.Nigiri.SVO.Tree.cs
Scripts/NKLI.Nigiri.SVO.Helpers.cs
Scripts/NKLI.Nigiri.SVO.SVONode.cs
Scripts/NKLI.Nigiri.SVO.TraversalNode.cs
Scripts/NKLI.Nigiri.SVO.Voxelizer.cs
Scripts/NKLI.Nigiri.SVO.TestUnitHooks.cs
Scripts/NKLI.Nigiri.Helpers.cs
Scripts/NKLI.Nigiri.Tools.MainThreadDispatcher.cs

Shaders

Resources/NKLI_Nigiri_SVOVoxelizer.compute
Resources/NKLI_Nigiri_SVOVoxelizer_Functions.cginc
Resources/NKLI_Nigiri_SVOSplitter.compute
Resources/NKLI_Nigiri_SVOMipmapper.compute
Resources/NKLI_Nigiri_SVONode.cginc
Resources/NKLI_Nigiri_SVOTraversalNode.cginc
Resources/NKLI_Nigiri_MortonOrder2D.cginc

Clone this wiki locally