A quick recap of the primitive creation step, which consists of the following:
Details on why all these steps are required can be found in Basic Concepts. The fact that is important for us now is that a primitive descriptor created at step 2 fully defines the operation that the corresponding primitive will execute. Once the primitive descriptor is created, it cannot be changed.
The parameters passed to create a primitive descriptor specify the problem. An engine specifies where the primitive will be executed. An operation descriptor specifies the basics: the operation kind; the propagation kind; the source, destination, and other tensors; the strides (if applicable); and so on.
Attributes specify some extra properties of the primitive. The attributes were designed to be extensible, hence they are an opaque structure. Users must create them before use and must set required specifics using the corresponding setters. The attributes are copied during primitive descriptor creation, so users can change or destroy attributes right after that.
If not modified, attributes can stay empty, which is equivalent to the default attributes. For that purpose, in the C API users can pass NULL
as an attribute to the dnnl_primitive_desc_create function. In the C++ API, primitive descriptors' constructors have empty attributes as default parameters, so, unless they are required, users can simply omit them.
Below are the skeletons of using attributes with the C and C++ APIs. Error handling is omitted to simplify reading.
As mentioned above, the attributes enable extending or changing the default primitive behavior. Currently the following attributes are supported. The detailed explanation is provided in the corresponding sections.
Because the attributes are created separately from the corresponding primitive descriptor, consistency checks are delayed. Users can successfully set attributes in whatever configuration they want. However, when they try to create a primitive descriptor with the attributes they set, it might happen that there is no primitive implementation that supports such a configuration. In this case, the library will return dnnl_unimplemented in the case of the C API or throw a corresponding dnnl::error exception in the case of the C++ API. Unfortunately, the library does not currently provide any hints about what exactly is going wrong in this case. The corresponding section of the documentation simply documents the primitives' capabilities.