Creating a GridBuffer with PNanoVDB (DX12/HLSL)

[gaida-exe]

Hey 👋
currently, I am trying to set up a basic renderer with NanoVDB using PNanoVDB.h in a Framework using DX12 and HLSL.

I already created the grid with mGridHandle = nanovdb::createFogVolumeSphere<float, float, nanovdb::HostBuffer>(100.0f, nanovdb::Vec3f(-20, 0, 0), 1.0, 3.0, nanovdb::Vec3d(0), "sphere"); and try to load the grid = mGridHandle.grid<float>(); into a Buffer.
Also, I already saw that the #define pnanovdb_buf_t StructuredBuffer<uint> is a StructuredBuffer.

Now my question would be, how to set up the buffer on both sides and if I need to split the buffer into the node sections like shown in the Viewer with glsl. Or if I need a separate buffer, that contains the offsets and counts (Node::memUsage() * grid->tree().nodeCount<Node>()).
Is there somewhere a PNanoVDB example? I only found the Viewer and the examples that use CUDA.

[gaida-exe]

Well, let's answer my own question.
Yes, NanoVDB and PNanoVDB are fully compatible.
You need to use Nano on the CPU, if you plan on using shaders to compute. Just simply copy PNanoVDB.h to your shaders folder, include it in your shader and use a language define like #define PNANOVDB_HLSL
If you want to upload via something like DX12 you simply just need a StructuredBuffer without any type of splitting the NodeTypes (using CNanoVDB is different here aka the Viewer, where you need to bind the range).
The StructuredBuffer needs a few things from Nano:

const nanovdb::GridHandle<nanovdb::HostBuffer> grid = nanovdb::createFogVolumeSphere<float, float, nanovdb::HostBuffer>(100.0f, nanovdb::Vec3f(-20, 0, 0), 1.0, 3.0, nanovdb::Vec3d(0), "sphere");
const nanovdb::NanoGrid<float>* data = grid.grid<float>();
const uint64_t elementCount = grid.size() / sizeof(uint32_t);
const uint64_t byteSize = grid.size();
descriptor.structStride = sizeof(uint32_t);

Also, the buffer should be GPU_USAGE_ONLY and an equivilant of StructuredBuffer.

On the GPU you should be able to accept the Grid as the in the question mentioned #define pnanovdb_buf_t StructuredBuffer<uint>. Keep in mind that you should use the variable, so it reflects correctly.

pnanovdb_buf_t openVDBTestGrid : register(t1, UPDATE_FREQ_NONE);
reflect for a dummy load:

uint numStructs, stride;
openVDBTestGrid.GetDimensions(numStructs,stride);
//  I return the stride as the red of the color to use, so the variable is not optimized away
return float4( stride, 0.0, 0.0, 1.0); 

⚠️ take the following with a grain of salt

My first attempts on using Nano are limited, so I won't share any code on it, because it's not working right now ^^'
But what I tried was using the pnanovdb_hdda_zero_crossing and for it, you require different variables like an accessor and gridType.
Look these variables up, and you'll see that you can create them with init functions and so on.
Also take a look at variables that are marked as PNANOVDB_INOUT. These are most likely only there to put out variables, and you don't need to assign a value to them, since they are the result.

I don't know if this was helpful or redundant, but I would have liked to know this information to start with Nano.

I bet some things are wrong or poorly explained, so correct me and add to it.