Vector Dot Product (dotprod)

API Keywords: vector dot product dotprod dotprod_rrrf dotprod_crcf dotprod_cccf

This module provides interfaces for computing a vector dot product between two equally-sized vectors. Dot products are commonly used in digital signal processing for communications, particularly in filtering and matrix operations. Given two vectors of equal length\(\vec{x} = \left[x(0),x(1),\ldots,x(N-1)\right]^T\) and\(\vec{v} = \left[v(0),v(1),\ldots,v(N-1)\right]^T\) , the vector dot product between them is computed as

$$ \vec{x} \cdot \vec{v} = \vec{x}^T \vec{v} = \sum_{k=0}^{N-1}{ x(k) v(k) } $$

A number of other liquid modules rely on dotprod , such as filtering and equalization.

Specific machine architectures

The vector dot product has a complexity of \(\mathcal{O}(N)\) multiply-and-accumulate operations. Because of its prevalence in multimedia applications, a considerable amount of research has been put into computing the vector dot product as efficiently as possible. Software-defined radio is no exception as basic profiling will likely demonstrate that a considerable portion of the processor is spent computing it. Certain machine architectures have specific instructions for computing vector dot products, particularly those which use a single instruction for multiple data (SIMD) such as MMX, SSE, AltiVec, etc.


There are effectively two ways to use the dotprod module. In the first and most general case, a vector dot product is computed on two input vectors \(\vec{x}\) and \(\vec{v}\) whose values are not known a priori . In the second case, a dotprod object is created around vector \(\vec{v}\) which does not change (or rarely changes) throughout its life cycle. This is the more convenient method for filtering objects which don't usually have time-dependent coefficients. Listed below is a simple interface example to the dotprod module object:

#include <liquid/liquid.h>

int main() {
    // create input arrays
    float x[] = { 1.0f,  2.0f,  3.0f,  4.0f,  5.0f};
    float v[] = { 0.1f, -0.2f,  1.0f, -0.2f,  0.1f};
    float y;

    // run the basic vector dot product, store in 'y'

    // create dotprod object and execute, store in 'y'
    dotprod_rrrf q = dotprod_rrrf_create(v,5);

In both cases the dotprod can be easily integrated with the window object ([ref:section-buffer-window] ) for managing input data and alignment. There are three types of dot product objects and are listed in[ref:tab-dotprod-objects] .

Table [tab-dotprod-objects]. Data types for dotprod family of objects

Precision Input/Output Coefficients Interface
float real real dotprod_rrrf
float complex complex dotprod_cccf
float complex real dotprod_crcf

Listed below is a brief description of the dotprod object interfaces. While the types are described using the dotprod_rrrf object, the same holds true for all other types.

  • dotprod_rrrf_run(h,x,n,y) executes a vector dot product between two vectors \(\vec{h}\) and\(\vec{x}\) , each of length \(n\) and stores the result in the output\(y\) . This is not a structured method and does not require creating a dotprod object, however does not take advantage of SIMD instructions if available. Rather than speed, its intent is to provide a simple interface to demonstrate functional correctness.
  • dotprod_rrrf_create(v,n) creates a dotprod object with coefficients \(\vec{v}\) of length\(n\) .
  • dotprod_rrrf_recreate(q,v,n) recreates a dotprod object with a new set of coefficients\(\vec{v}\) with a (possibly) different length \(n\) .
  • dotprod_rrrf_destroy(q) destroys a dotprod object, freeing all internally-allocated memory.
  • dotprod_rrrf_print(q) prints the object internals to the screen.
  • dotprod_rrrf_execute(q,x,y) executes a dot product with an input vector \(\vec{x}\) and stores the result in \(y\) .