Parameter Inference

Module parameters are often inferrable from the types and latencies of the module's ports. Most Parameter inference happens via exact match. So for instance, if a type float[3] is provided to an input of of type T, then T infers to type float[3].

module IAmInferrable#(T) {
    input T my_input
    //...
}

module Use {
    float[3] x = [0.1, 0.2, 0.3]
    IAmInferrable inf // Infers to T: type float[3]
    inf.my_input = x
}

Similarly, values used within the declaration of types can also be inferred. For instance, in the following case, the parameter WIDTH can be inferred from the size of the incoming boolean array.

module InferMyArraySize#(int WIDTH) {
    input bool[WIDTH] bits
    //...
}

module Use {
    InferMyArraySize inf // Infers to WIDTH: 32
    inf.bits = 32'hDEADBEEF
}

Integer Inference

While other parameters can only be inferred by exact match, integer parameters actually can be inferred more flexibly due to the integer subtyping relationship, with inference candidates providing minimum and maximum values rather than requiring an exact match.

For example, let's take this IntegerEquals module:

module IntegerEquals #(int FROM, int TO) {
    action IntegerEquals :
        int#(FROM: FROM, TO: TO) a,
        int#(FROM: FROM, TO: TO) b ->
        bool result

    result = a == b
}

module UseIntegerEquals {
    int#(FROM: 3, TO: 9) x
    int#(FROM: -2, TO: 5) y

    bool are_eq = IntegerEquals(x, y)
    // infers to IntegerEquals#(FROM: -2, TO: 9)
}

On IntegerEquals::FROM, there are two maximum constraints: One for FROM <= 3 for a, and one for FROM <= -2 for b. Therefore FROM infers to -2. Likewise IntegerEquals::TO has two minimum constraints: TO >= 9 for a and TO >= 5 for b, inferring TO to 9.

Latency Inference

Latency Inference is a kind of integer inference. And as with inference of integer bounds, Latency Inference only applies a maximum latency constraint from an output to an input, which translates to a minimum or maximum bound on the latency inferrable parameter.

See Latency Inference.