Sample a shader in 3D space

fwilleke80

Hi & thanks for your reply.

@zipit said in Sample a shader in 3D space:

I am not quite sure if I understand your question correctly and also am unclear on what you would consider "not working" in the context of sampling a volumetric shader.

I get only one value, regardless of the sample coordinates.

However, what stood out to me in your code, is that you only do sample in a slice of the volume, basically treat the shader like a 2D-Shader, which might have side effects as your sampled shaders might simply only return the zero vector in that slice (here comes up the question of what is not working).
 sampleCd.p = Vector((Float)x * gridWidthI, (Float)y * gridHeightI, 0.0);

z is 0.0 because I'm sampling at positions on a plane. ChannelData::p is, according to the SDK docs, only in UVW space, anyway (that's why I multiply the X and Y grid coordinate with the inverse X and Y size, so my actual sample values are in the range of 0.0 ... 1.0). And sampling those positions works. For 3D shaders, as far as I always understood it, the sampleCd.vd->p is the one that counts, as that is actually a position in 3D space.

def main():
"""
"""
shader = c4d.BaseList2D(c4d.Xwood)
irs = c4d.modules.render.InitRenderStruct(doc)
shader.InitRender(irs)
for i in range(10):
    n = i / 9.
    cd = c4d.modules.render.ChannelData()
    cd.p = c4d.Vector(0, 0, n)
    print shader.Sample(cd)    
shader.FreeRender()
if name=='main':
main()
Vector(0.187, 0.077, 0.03)
Vector(0.171, 0.07, 0.028)
Vector(0.091, 0.038, 0.015)
Vector(0.163, 0.067, 0.027)
Vector(0.255, 0.105, 0.041)
Vector(0.275, 0.113, 0.044)
Vector(0.314, 0.13, 0.05)
Vector(0.249, 0.103, 0.04)
Vector(0.095, 0.039, 0.016)
Vector(0.318, 0.131, 0.05)

Hmm, that looks meaningful in deed. I still wonder why I get nothing out of the Wood shader and others.

ferdinand

Hm,

I might be wrong about this, but I always thought that VolumeData.p is a sample position in object space and ChannelData.p is the sample position in uvw space. The latter would lie in the half interval uv(w) coordinates are normally defined in, while the former would lie in an arbitrary interval, depending on the bounding box of the sampled geometry (i.e. 0 <= c <= 1 is not guaranteed to be met for the components of VolumeData.p). So a geometry-agnostic shader should not care about VolumeData.p.

Also note that I did deliberately only sample on the z-axis in my example and still get a varying output. If your premise would be true, I should get the same vector for all ten calls.

Cheers,
zipit

m_adam

Hi sorry for not answering you earlier, I just wanted to inform you I reached the development team about it especially since you use multi-threading, and sampling a shader with Multiple thread is different (for example in my case your code does not work even on a 2D gradient).
I have a property but which doesn't support all shaders (especially Layer shader).

So I will keep you in touch when I have more news.
Cheers,
Maxime.

m_adam

Sorry for the delay, this took me a bit more time than expected.

I would say @zipit is completely right here, but since you are using multi threads you have to attach a FakeVolumeData for each thread since some shader really needs and output their result according to this VolumeData.

Here a complete example.

// class that hold per thread sampling data
class ShaderSampler
{
	public:

	BaseContainer _rdc;
	BaseObject* _camera = nullptr;

	ChannelData _channelData;
	::Ray _ray;
	VolumeData* _volumeData = nullptr;

	ShaderSampler(){};
	~ShaderSampler() 
	{
		if (_camera)
			BaseObject::Free(_camera);

		if (_volumeData)
			VolumeData::Free(_volumeData);
	};

	void SetSamplingContext(const Vector& uv, const Vector& delta)
	{
		VolumeData& volumeData = *_volumeData;

		const Vector d = Vector(delta.x, delta.y, delta.z);
		_channelData.d = d;
		volumeData.delta = d;

		const Vector p = Vector(uv.x, uv.y, uv.z);

		_channelData.p = p;
		volumeData.uvw = p;
		volumeData.p = p;
		volumeData.pp[0] = p;
		volumeData.pp[1] = p;
		volumeData.pp[2] = p;

		_ray.p = p;
		_ray.pp[0] = p;
		_ray.pp[1] = p;
		_ray.pp[2] = p;
	}

};

static maxon::Result<void> SampleMaterialBis(BaseDocument* doc)
{
	iferr_scope;

	//Get the BaseShader linked in the color shader of the first  material
	BaseShader* shader;
	MAXON_SCOPE
	{
		// Get the first material and check its a Cinema 4D default material
		BaseMaterial* firstMat = doc->GetFirstMaterial();
		CheckState(firstMat != nullptr, "Failed to get the first material");
		CheckState(firstMat->IsInstanceOf(Mmaterial), "material is not a default c4d material");

		// look for the material BaseContainer and the BaseList2D associated to the color param
		BaseContainer* matBC = firstMat->GetDataInstance();
		CheckState(matBC != nullptr, "Failed to get the BaseContainer");

		BaseList2D* matColorBL = matBC->GetLink(MATERIAL_COLOR_SHADER, doc);
		CheckState(matColorBL != nullptr, "Failed to get the Color Shader");

		// cast to BaseShader
		shader = (BaseShader*)matColorBL;
	}

	// Allocate and init the InitRenderStruct
	// This is important to init with valid FakeVolumeData since some shaders needs them.
	InitRenderStruct irs;
	VolumeData* volumeData = VolumeData::Alloc();
	BaseObject* camera = BaseObject::Alloc(Ocamera);
	CheckState(volumeData != nullptr, "Failed to initialize volume data");
	CheckState(camera != nullptr, "Failed to initialize camera");
	MAXON_SCOPE
	{
		BaseContainer renderData;
		const Int threadIndex = maxon::JobRef::GetCurrentWorkerThreadIndex();
		renderData.SetInt32(RDATA_VDFAKE_CURRENTTHREAD, threadIndex);
		renderData.SetInt32(RDATA_VDFAKE_THREADCOUNT, (Int32)GeGetCurrentThreadCount());

		const Bool attached = C4DOS.Sh->AttachVolumeDataFake(volumeData, camera, renderData, nullptr);
		CheckState(attached == true, "Failed to attach volume data");

		irs.vd = volumeData;
		irs.flags = INITRENDERFLAG::TEXTURES;
		irs.time = doc->GetTime();
		irs.docpath = doc->GetDocumentPath();
		irs.thread = GeGetCurrentThread();

		COLORSPACETRANSFORMATION transform = COLORSPACETRANSFORMATION::NONE;
		// check if linear workflow is enabled
		if (irs.linear_workflow)
			transform = COLORSPACETRANSFORMATION::LINEAR_TO_SRGB;

		const INITRENDERRESULT initResult = shader->InitRender(irs);
		CheckState(initResult == INITRENDERRESULT::OK, "Shader init failed");
	}

	// Allocate a list of ShaderSampler to store VolumeData and ChannelData per thread
	maxon::BaseArray<ShaderSampler*> samplers;
	MAXON_SCOPE
	{
		for (Int32 i = 0; i < GeGetCurrentThreadCount(); i++)
		{
			ShaderSampler* sampler = NewObjClear(ShaderSampler);

			const Int threadIndex = maxon::JobRef::GetCurrentWorkerThreadIndex();
			sampler->_rdc.SetInt32(RDATA_VDFAKE_CURRENTTHREAD, threadIndex);
			sampler->_rdc.SetInt32(RDATA_VDFAKE_THREADCOUNT, GeGetCurrentThreadCount());

			sampler->_camera = BaseObject::Alloc(Ocamera);
			CheckState(sampler->_camera != nullptr, "Failed to initialize camera");

			sampler->_volumeData = VolumeData::Alloc();
			CheckState(sampler->_volumeData != nullptr, "Failed to initialize volume data");

			const Bool attached = C4DOS.Sh->AttachVolumeDataFake(sampler->_volumeData, sampler->_camera, sampler->_rdc, nullptr);
			CheckState(attached == true, "Failed to attach volume data");

			samplers.Append(sampler) iferr_return;
		}
	}

	// Worker lambda to build the grid
	Int gridSize = 64;

	// Create a BaseArray of matrices and colors (to store the results of our sampling, and visualize them in an instanceObject)
	maxon::BaseArray<Matrix>	matrices;
	maxon::BaseArray<maxon::Color64> colors;

	// * 3 because we store the data for x,y,z axis in one big array
	matrices.Resize(gridSize * gridSize * gridSize) iferr_return;
	colors.Resize(gridSize * gridSize * gridSize) iferr_return;

	auto worker = [&samplers, &shader, &matrices, &colors, &gridSize](UInt y)
	{
		// Retrieve the current sampler data for the current thread
		const Int threadIndex = maxon::JobRef::GetCurrentWorkerThreadIndex();
		ShaderSampler* sampler = samplers[threadIndex];
		if (sampler == nullptr)
			return;

		for (Int32 z = 0; z < gridSize; ++z)
		{
			for (Int32 x = 0; x < gridSize; ++x)
			{
				// Map 3D array to a 1D array
				Int32 arrayId = (z * gridSize * gridSize) + (y * gridSize) + x;
				matrices[arrayId] = MatrixMove(Vector(x, y, z));

				// Define where we want to sample
				Vector uvw = Vector(
					double(x) / double(gridSize),
					double(y) / double(gridSize),
					double(z) / double(gridSize)
					);

				Vector delta = Vector(0.01, 0.01, 0.01);

				sampler->SetSamplingContext(uvw, delta);

				// sample
				const Vector sampledValue = shader->Sample(&sampler->_channelData);

				// Store the color result of the sampling operation
				colors[arrayId] = maxon::Color64(sampledValue.x, sampledValue.y, sampledValue.z);
			}
		}
	};

	// Execute worker
	maxon::ParallelFor::Dynamic(0, gridSize, worker, GeGetCurrentThreadCount());

	// Free samplers
	MAXON_SCOPE
	{
		for (Int32 i = 0; i < GeGetCurrentThreadCount(); i++)
		{
			ShaderSampler* sampler = samplers[i];
			if (sampler != nullptr)
			{
				DeleteObj(sampler);
				samplers[i] = nullptr;
			}
		}
	}

		// call the FreeRender to release allocated memory used for sampling
	shader->FreeRender();

	VolumeData::Free(volumeData);
	BaseObject::Free(camera);

	// Create a Multi-Instance object with the matrix and the color information to visualize them
	InstanceObject* const instance = InstanceObject::Alloc();
	CheckState(instance != nullptr, "Failed to allocate an Instance Object");
	MAXON_SCOPE
	{
		doc->InsertObject(instance, nullptr, nullptr);

		// Set to multi instance + to Points mode
		if (!instance->SetParameter(INSTANCEOBJECT_RENDERINSTANCE_MODE, INSTANCEOBJECT_RENDERINSTANCE_MODE_MULTIINSTANCE, DESCFLAGS_SET::NONE))
			return maxon::UnexpectedError(MAXON_SOURCE_LOCATION);

		if (!instance->SetParameter(INSTANCEOBJECT_DRAW_MODE, INSTANCEOBJECT_DRAW_MODE_POINTS, DESCFLAGS_SET::NONE))
			return maxon::UnexpectedError(MAXON_SOURCE_LOCATION);

		// Set data in the instance object
		instance->SetInstanceMatrices(matrices) iferr_return;
		instance->SetInstanceColors(colors) iferr_return;
	}

	EventAdd(EVENT::NONE);

	return maxon::OK;
}

Cheers,
Maxime.

fwilleke80

Thank you very much! I'll try that out ASAP!

Cheers,
Frank

fwilleke80

Almost perfect, thanks. It kinda only works half ok, I noticed.

The thing is, all shaders are always sampled in UV space. I tried adding the 3D position data for sampling, too, by setting volumeData.p to the global position of the point to sample, in SetSamplingContext(). but it is being ignored. For example, if the Shader is a Noise shader, I can set it to whatever space I want, the result is always the same.

With my old code, that worked. It just couldn't sample certain shaders (Layer Shader, Earth Shader, etc). Now, those shaders work, but sampling seems restricted to UV space. How can I sample those shaders in Texture Space, Object Space, and World Space (Camera, Screen, and the others can be neglected)?

Thanks in advance,
Frank

fwilleke80

Any news on this?

Cheers,
Frank

m_adam

Hi he completely get out of my mind sorry, I will work on your topic in the upcoming days.
However I'm not sure if this is possible since you will probably need to instantiate some VolumeData which is not possible, but I will have a look.

Cheers,
Maxime.

fwilleke80

No worries. I forgot about it, too, until I noticed the issue again recently.

I have sent an email to sdk_support(at)maxon(dot)net with some example code and more detailled info.
Let's continue the discussion there

Cheers,
Frank

m_adam

Hi again sorry for the long delay, pretty busy.

I looked at the noise code and it assume a cd->vd->tex to be defined unfortually allocating a dummy TexData is not working properly, I will need to reach development team to see if this is possible at all.

Cheers,
Maxime.

Paul Everett

you can sample a channel. it works perfectly for things like the layer shader.
if you really have to sample a shader specifically, what I do is make a clone of the shader per each context.

i.e. you have 16 cores, you make 16 clones, and insert them on the material and initing them, (being sure to clean up after)
Layer shader will then work as expected, off the bat, as long as you use the thread index to reference from your list of shader clones.
clones.shader_list[context.GetLocalThreadIndex()]

you also have to make sure you project your coordinate according to the texture tag. it wont do that for you.
p is always assumed to be a UV coordinate projected, as per the tex tag. you send p as a world coordinate with vd and if the shader needs that, that is where it would look. think of it from the perspective of a shader. where will it get a world position from? if you write a shader, and you need a world position for a hit, you look in vd for that. if vd is null, or the info has not been filled, how would it know where p is?

I found in most cases sampling a channel from a material rather than a specific shader, works best, and is less problematic.

best
Paul

fwilleke80

Thanks, Paul! I'll try that out

Cheers,
Frank