@stdlib/strided-dispatch

Create a strided array function interface which performs multiple dispatch.

Usage no npm install needed!

<script type="module">
  import stdlibStridedDispatch from 'https://cdn.skypack.dev/@stdlib/strided-dispatch';
</script>

README

Dispatch

NPM version Build Status Coverage Status

Create a strided array function interface which performs multiple dispatch.

Installation

npm install @stdlib/strided-dispatch

Usage

var dispatch = require( '@stdlib/strided-dispatch' );

dispatch( fcns, types, data, nargs, nin, nout )

Returns a strided array function interface which performs multiple dispatch.

var unary = require( '@stdlib/strided-base-unary' );
var Float64Array = require( '@stdlib/array-float64' );
var Float32Array = require( '@stdlib/array-float32' );

function foo( x ) {
    return x * 10.0;
}

function bar( x ) {
    return x * 5.0;
}

// Define a list of strided array functions for applying a unary callback:
var fcns = [
    unary,
    unary
];

// Define a one-dimensional list of input and output array types:
var types = [
    'float64', 'float64', // input, output
    'float32', 'float32'  // input, output
];

// Define a list of callbacks which should be applied based on the provided array types:
var data = [
    foo,
    bar
];

// Define the total number of input arguments:
var nargs = 7; // N + input_array_dtype + input_array + input_array_stride + output_array_dtype + output_array + output_array_stride

// Define the number of input strided arrays:
var nin = 1;

// Define the number of output strided arrays:
var nout = 1;

// Create a strided array function interface:
var strided = dispatch( fcns, types, data, nargs, nin, nout );

// ...

var x = new Float64Array( [ 1.0, 2.0, 3.0 ] );
var y = new Float64Array( x.length );

strided( x.length, 'float64', x, 1, 'float64', y, 1 );
// y => <Float64Array>[ 10.0, 20.0, 30.0 ]

x = new Float32Array( [ 1.0, 2.0, 3.0 ] );
y = new Float32Array( x.length );

strided( x.length, 'float32', x, 1, 'float32', y, 1 );
// y => <Float32Array>[ 5.0, 10.0, 15.0 ]

The function accepts the following arguments:

  • fcns: list of strided array functions.
  • types: one-dimensional list of strided array argument data types. The length of types must be the number of strided array functions multiplied by nin+nout. If fcns is a function, rather than a list, the number of strided array functions is computed as types.length / (nin+nout).
  • data: strided array function data (e.g., callbacks). If a list, the length of data must equal the number of strided array functions. If null, a returned strided array function interface does not provide a data argument to an invoked strided array function.
  • nargs: total number of strided array function interface arguments (including data types, strides, and offsets).
  • nin: number of input strided arrays.
  • nout: number of output strided arrays.

Notes

  • Without offsets, a returned strided array function interface has the following signature:

    f( N, dtypeX, x, strideX, dtypeY, y, strideY, ... )
    

    where

    • N: number of indexed elements.
    • dtypeX: data type for x.
    • x: strided array.
    • strideX: index increment for x.
    • dtypeY: data type for y.
    • y: strided array.
    • strideY: index increment for y.
    • ...: additional strided arrays and associated data types and strides.
  • The number of strided array function interface parameters is derived from nargs, the number of input strided arrays is derived from nin, and the number of output strided arrays is derived from nout.

  • Without offsets, the number of parameters must obey the following relation:

    nargs = 3*(nout+nin) + 1
    
  • With offsets, the number of parameters must obey the following relation:

    nargs = 4*(nout+nin) + 1
    
  • With offsets, a returned strided array function interface has the following signature:

    f( N, dtypeX, x, strideX, offsetX, dtypeY, y, strideY, offsetY, ... )
    

    where

    • N: number of indexed elements.
    • dtypeX: data type for x.
    • x: strided array.
    • strideX: index increment for x.
    • offsetX: starting index for x.
    • dtypeY: data type for y.
    • y: strided array.
    • strideY: index increment for y.
    • offsetY: starting index for y.
    • ...: additional strided arrays and associated data types, strides, and offsets.

    The choice of which strided array function interface to return depends on the use case. The former is suitable for typed array views; while the latter affords alternative indexing semantics more suitable for n-dimensional arrays (ndarrays).

  • Without offsets, a strided array function (i.e., a value provided for the fcns argument) should have the following signature:

    f( arrays, shape, strides[, data] )
    

    where

    • arrays: array containing strided input and output arrays.
    • shape: array containing a single element, the number of indexed elements.
    • strides: array containing the stride lengths for the strided input and output arrays.
    • data: strided array function data (e.g., a callback).
  • With offsets, a strided array function should have the following signature:

    f( arrays, shape, strides, offsets[, data] )
    

    where

    • offsets: array containing the starting indices (i.e., index offsets) for the strided input and output arrays.
  • For convenience, a single strided array function may be provided which will be invoked whenever the strided array argument data types match a sequence of types in types. Providing a single strided array function is particularly convenient for the case where, regardless of array data types, traversing arrays remains the same, but the strided array function data differs (e.g., callbacks which differ based on the array data types). For example, the following

    var unary = require( '@stdlib/strided-base-unary' );
    
    function foo( x ) {
        return x * 10.0;
    }
    
    function bar( x ) {
        return x * 5.0;
    }
    
    var fcns = [
        unary,
        unary
    ];
    var types = [
        'float64', 'float64',
        'float32', 'float32'
    ];
    var data = [
        foo,
        bar
    ];
    
    var strided = dispatch( fcns, types, data, 7, 1, 1 );
    

    is equivalent to

    var unary = require( '@stdlib/strided-base-unary' );
    
    function foo( x ) {
        return x * 10.0;
    }
    
    function bar( x ) {
        return x * 5.0;
    }
    
    var types = [
        'float64', 'float64',
        'float32', 'float32'
    ];
    var data = [
        foo,
        bar
    ];
    
    var strided = dispatch( unary, types, data, 7, 1, 1 );
    

Examples

var unary = require( '@stdlib/strided-base-unary' ).ndarray;
var abs = require( '@stdlib/math-base-special-abs' );
var Float64Array = require( '@stdlib/array-float64' );
var dispatch = require( '@stdlib/strided-dispatch' );

var types = [ 'float64', 'float64' ];

var data = [
    abs
];

var strided = dispatch( unary, types, data, 9, 1, 1 );

var x = new Float64Array( [ -1.0, -2.0, -3.0, -4.0, -5.0 ] );
var y = new Float64Array( [ 0.0, 0.0, 0.0, 0.0, 0.0 ] );

strided( 3, 'float64', x, 1, 2, 'float64', y, 1, 2 );
console.log( y );
// => <Float64Array>[ 0.0, 0.0, 3.0, 4.0, 5.0 ]

Notice

This package is part of stdlib, a standard library for JavaScript and Node.js, with an emphasis on numerical and scientific computing. The library provides a collection of robust, high performance libraries for mathematics, statistics, streams, utilities, and more.

For more information on the project, filing bug reports and feature requests, and guidance on how to develop stdlib, see the main project repository.

Community

Chat


License

See LICENSE.

Copyright

Copyright © 2016-2022. The Stdlib Authors.