We have already talked about how to create materials, assign them parameters and bind them to a Renderable component. In this section we'll show an advanced way to assign material parameters, how to use materials for rendering directly (without using Renderable component) and how to create your own shaders without the use of BSL.
Previously we have shown to how to set Material parameters by calling methods like @bs::Material::setTexture, @bs::Material::setFloat, @bs::Material::setColor, @bs::Material::setVec4 and similar.
As an alternative you can also set materials through material parameter handles. Once a material handle is retrieved it allows you to set material parameters much more efficiently than by calling the methods above directly.
To retrieve the handles call any of the following methods, depending on material parameter type:
Handles provide set() and get() methods that can be used for writing and reading the parameter values.
HMaterial material = ...;
MaterialParamMat4 myMatParam;
MaterialParamTexture myTextureParam;
material->getParamMat4("vertProjMatrix", myMatParam);
material->getParamTexture("mainTexture", myTextureParam);
Matrix4 viewProjMat = ...;
SPtr<Texture> someTexture = ...;
myVectorParam.set(viewProjMat);
myTextureParam.set(someTexture);
Material handles are very similar as GpuParams handles we talked about earlier. There are three major differences:
So far when we wanted to create a shader we would create a BSL file which would then be imported, creating a @bs::Shader. But you can also create shaders manually by explicitly providing HLSL/GLSL code for GpuProgram%s and non-programmable states. Most of the things outlined in this section are performed by BSL compiler internally when a Shader is imported.
Each shader definition contains two things:
To summarize, the relationship between materials, shaders, techniques and passes is:
Optionally, the shader definition can also contain a set of @bs::SubShader objects. As explained previously sub-shaders are a set of techniques that are used to override specific renderer behaviour.
A Pass can be created by filling out a @bs::PASS_DESC descriptor and passing it to @bs::Pass::create method. PASS_DESC is fairly simple and it expects a set of GPU program and non-programmable state descriptors.
GPU_PROGRAM_DESC vertexProgDesc = ...;
GPU_PROGRAM_DESC fragmentProgDesc = ...;
BLEND_STATE_DESC blendStateDesc = ...;
// Create a pass with a vertex and fragment program, and a non-default blend state
PASS_DESC desc;
desc.vertexProgram = vertexProgDesc;
desc.fragmentProg = fragmentProgDesc;
desc.blendState = blendStateDesc;
SPtr<Pass> pass = Pass::create(desc);
The descriptors for GPU programs and non-programmable states are filled out as described in the low-level rendering API manual.
Now that we know how to create a pass, we can use one or multiple passes to initialize a Technique. A technique is just a container for one or multiple passes.
To create a technique call @bs::Technique::create and provide it with:
For example:
SPtr<Pass> pass = ...;
// Create a technique that uses HLSL and has a single pass
SPtr<Technique> technique = Technique::create("HLSL", { pass });
Now that we have a technique we can create the shader by calling @bs::Shader::create, which expects the name of the shader and a @bs::TSHADER_DESC structure as input.
@bs::TSHADER_DESC contains a list of techniques to initialize the shader with, as well as a set of optional parameters to control how the shader rendering works:
@bs::TSHADER_DESC::separablePasses - An optimization hint to the renderer that can improve performance when turned on. Only relevant if the shader has techniques with multiple passes. When true the renderer will not necessarily execute passes right after another, but might render other objects in-between passes. This can reduce state switching as multiple objects can be rendered with a single pass, but is only relevant for algorithms that can handle such a process (most can't).
SPtr<Technique> technique = ...;
SHADER_DESC desc;
desc.queueSortType = QueueSortType::None;
desc.queuePriority = 0;
desc.separablePasses = false;
desc.techniques = { technique };
SPtr<Shader> shader = Shader::create("MyShader", desc);
Shader parameters allow you to change values of parameters in GPU programs through the Material interface. They are similar to GPU program parameters described earlier, but they set the values on all GPU programs on all passes in the active technique, instead of doing so only on a single GPU program. Additionally they also support renderer semantics (see below).
To create the parameter interface you must populate the SHADER_DESC structure by calling one of the @bs::TSHADER_DESC::addParameter overloads.
Parameters come in two variants:
For each parameter you must specify:
Type of the GPU variable, as described above.
// Extended example from above
SPtr<Technique> technique = ...;
SHADER_DESC desc;
desc.queueSortType = QueueSortType::None;
desc.queuePriority = 0;
desc.separablePasses = false;
desc.techniques = { technique };
// Add a 4x4 transform matrix data parameter
desc.addParameter(SHADER_DATA_PARAM_DESC("WorldTfrm", "WorldTfrm", GPDT_MATRIX_4X4));
// Add a texture parameter
desc.addParameter(SHADER_OBJECT_PARAM_DESC("AlbedoTex", "AlbedoTex", GPOT_TEXTURE2D));
SPtr<Shader> shader = Shader::create("MyShader", desc);
When adding parameters you can also specify two additional properties we didn't touch on in the previous section: renderer semantic and default value.
Renderer semantic allows you to give the parameter a unique tag that can be recognized by the active renderer. The renderer can then use these semantics to automatically assign values to them while rendering. For example the "WVP" semantic might notify the renderer to populate this parameter with the world-view-projection matrix. This way the user is not responsible for setting such parameters manually. The actual semantics supported depend on the active renderer. If provided and renderer doesn't support a semantic, it will be ignored. We'll talk more on how to access semantics in the renderer manual.
The parameter default value allows you to provide a value that will be used for initializing the parameter when a Material is initially constructed. For data parameters the default value is a provided as a raw block of memory, and for object parameters it can be a reference to a Texture or a SamplerState.
// An extended example from above with semantics and default values:
SPtr<Technique> technique = ...;
SHADER_DESC desc;
desc.queueSortType = QueueSortType::None;
desc.queuePriority = 0;
desc.separablePasses = false;
desc.techniques = { technique };
// Add a 4x4 transform matrix data parameter with a "W" semantic and identity matrix as default
desc.addParameter(SHADER_DATA_PARAM_DESC("WorldTfrm", "WorldTfrm", GPDT_MATRIX_4X4, "W"), &Matrix4::Identity);
// Add a texture parameter with an "Albedo" semantic and a white texture as default
desc.addParameter(SHADER_OBJECT_PARAM_DESC("AlbedoTex", "AlbedoTex", GPOT_TEXTURE2D, "Albedo"), Texture::White);
HShader shader = Shader::create("MyShader", desc);
In an earlier manual we have shown how to render using a Material by attaching it to a Renderable component and letting the renderer do the rest. You can however render using the material completely manually, using the low-level rendering API.
Material is a CoreObject which means it also has a core-thread interface accessible through @bs::Material::getCore(). The interface is the same as the non-core interface we have described so far.
Material cannot be bound directly to the low level rendering API. You must instead manually retrieve a pipeline state for one of its passes.
You can retrieve a specific pass from a material by calling @bs::Material::getPass(). The method expects an index of a technique and an index of a pass. To get the number of supported techniques call @bs::Material::getNumTechniques(), and to get the number of passes for a specific technique call @bs::Material::getNumPasses() with a specific technique index.
Once you have a Pass you can retrieve from it either a GraphicsPipelineState or a ComputePipelineState by calling @bs::Pass::getGraphicsPipelineState() and @bs::Pass::getComputePipelineState(), respectively. Those can then be bound for rendering as shown in the low level rendering API manual.
SPtr<Material> material = ...;
int passIdx = 0;
int techniqueIdx = 0;
SPtr<Pass> pass = material->getPass(passIdx, techniqueIdx);
RenderAPI& rapi = RenderAPI::instance();
rapi.setGraphicsPipeline(pass->getGraphicsPipelineState());
Alternatively you can use the helper methods @bs::ct::RendererUtility::setPass or @bs::ct::RendererUtility::setComputePass.
In order to bind material parameters we need to somehow get access to a GpuParams object from the material. This is done through an intermediate class @bs::GpuParamsSet, created by a call to @bs::Material::createParamsSet(), which as a parameter takes a technique index.
GpuParams for a specific pass can then be retrieved by calling @bs::GpuParamsSet::getGpuParams() with the pass index. They can then be bound as described in the low level render API manual.
SPtr<Material> material = ...;
int passIdx = 0;
int techniqueIdx = 0;
SPtr<GpuParamsSet> paramsSet = material->createParamsSet(techniqueIdx);
RenderAPI& rapi = RenderAPI::instance();
rapi.setGpuParams(paramsSet->getGpuParams(passIdx));
Note that creation of a GpuParamsSet object is expensive, and the intent is that it will be created once (or just a few times) per material. GpuParamsSet contains a completely separate storage from the Material it was created from, therefore whenever material parameters are updated you must transfer its contents into GpuParams by calling @bs::GpuParamsSet::update().
SPtr<GpuParamsSet> paramsSet = material->createParamsSet(techniqueIdx);
// ...update some parameters on the material...
// Transfer the updated data
paramsSet->update(material);
In case your material contains any animated properties like animation curves, color gradients or color textures, you can also provide a time parameter to GpuParamsSet::update(), which determines the point at which to sample animated properties.
// Sample animated properties at 0.5 seconds into the animation
paramsSet->update(material, 0.5f);
Once both the material's pipeline state and parameters are bound, you can then proceed to render as normally, as described in the low-level rendering manual.