Skip to content

Latest commit

 

History

History
314 lines (200 loc) · 9.34 KB

README.md

File metadata and controls

314 lines (200 loc) · 9.34 KB

dasumpw

Calculate the sum of absolute values (L1 norm) of double-precision floating-point strided array elements using pairwise summation.

The L1 norm is defined as

$$\|\mathbf{x}\|_1 = \sum_{i=0}^{n-1} \vert x_i \vert$$

Usage

var dasumpw = require( '@stdlib/blas/ext/base/wasm/dasumpw' );

dasumpw.main( N, x, strideX )

Computes the sum of absolute values (L1 norm) of double-precision floating-point strided array elements using pairwise summation.

var Float64Array = require( '@stdlib/array/float64' );

var x = new Float64Array( [ 1.0, -2.0, 2.0 ] );

var sum = dasumpw.main( x.length, x, 1 );
// returns 5.0

The function has the following parameters:

  • N: number of indexed elements.
  • x: input Float64Array.
  • strideX: stride length for x.

The N and stride parameters determine which elements in the strided array are accessed at runtime. For example, to access every other element in x,

var Float64Array = require( '@stdlib/array/float64' );

var x = new Float64Array( [ 1.0, 2.0, 2.0, -7.0, -2.0, 3.0, 4.0, 2.0 ] );

var sum = dasumpw.main( 4, x, 2 );
// returns 9.0

Note that indexing is relative to the first index. To introduce an offset, use typed array views.

var Float64Array = require( '@stdlib/array/float64' );

var x0 = new Float64Array( [ 2.0, 1.0, 2.0, -2.0, -2.0, 2.0, 3.0, 4.0 ] );
var x1 = new Float64Array( x0.buffer, x0.BYTES_PER_ELEMENT*1 ); // start at 2nd element

var sum = dasumpw.main( 4, x1, 2 );
// returns 9.0

dasumpw.ndarray( N, x, strideX, offsetX )

Computes the sum of absolute values (L1 norm) of double-precision floating-point strided array elements using pairwise summation and alternative indexing semantics.

var Float64Array = require( '@stdlib/array/float64' );

var x = new Float64Array( [ 1.0, -2.0, 2.0 ] );

var sum = dasumpw.ndarray( x.length, x, 1, 0 );
// returns 5.0

The function has the following additional parameters:

  • offsetX: starting index for x.

While typed array views mandate a view offset based on the underlying buffer, the offset parameter supports indexing semantics based on a starting index. For example, to access every other element starting from the second element:

var Float64Array = require( '@stdlib/array/float64' );

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

var v = dasumpw.ndarray( 4, x, 2, 1 );
// returns 9.0

Module

dasumpw.Module( memory )

Returns a new WebAssembly module wrapper instance which uses the provided WebAssembly memory instance as its underlying memory.

var Memory = require( '@stdlib/wasm/memory' );

// Create a new memory instance with an initial size of 10 pages (640KiB) and a maximum size of 100 pages (6.4MiB):
var mem = new Memory({
    'initial': 10,
    'maximum': 100
});

// Create a BLAS routine:
var mod = new dasumpw.Module( mem );
// returns <Module>

// Initialize the routine:
mod.initializeSync();

dasumpw.Module.prototype.main( N, xp, sx )

Computes the sum of absolute values (L1 norm) of double-precision floating-point strided array elements using pairwise summation.

var Memory = require( '@stdlib/wasm/memory' );
var oneTo = require( '@stdlib/array/one-to' );
var zeros = require( '@stdlib/array/zeros' );

// Create a new memory instance with an initial size of 10 pages (640KiB) and a maximum size of 100 pages (6.4MiB):
var mem = new Memory({
    'initial': 10,
    'maximum': 100
});

// Create a BLAS routine:
var mod = new dasumpw.Module( mem );
// returns <Module>

// Initialize the routine:
mod.initializeSync();

// Define a vector data type:
var dtype = 'float64';

// Specify a vector length:
var N = 3;

// Define a pointer (i.e., byte offset) for storing the input vector:
var xptr = 0;

// Write vector values to module memory:
mod.write( xptr, oneTo( N, dtype ) );

// Perform computation:
var sum = mod.main( N, xptr, 1 );
// returns 6.0

The function has the following parameters:

  • N: number of indexed elements.
  • xp: input Float64Array pointer (i.e., byte offset).
  • sx: stride length for x.

dasumpw.Module.prototype.ndarray( N, xp, sx, ox )

Computes the sum of absolute values (L1 norm) of double-precision floating-point strided array elements using pairwise summation and alternative indexing semantics.

var Memory = require( '@stdlib/wasm/memory' );
var oneTo = require( '@stdlib/array/one-to' );
var zeros = require( '@stdlib/array/zeros' );

// Create a new memory instance with an initial size of 10 pages (640KiB) and a maximum size of 100 pages (6.4MiB):
var mem = new Memory({
    'initial': 10,
    'maximum': 100
});

// Create a BLAS routine:
var mod = new dasumpw.Module( mem );
// returns <Module>

// Initialize the routine:
mod.initializeSync();

// Define a vector data type:
var dtype = 'float64';

// Specify a vector length:
var N = 3;

// Define a pointer (i.e., byte offset) for storing the input vector:
var xptr = 0;

// Write vector values to module memory:
mod.write( xptr, oneTo( N, dtype ) );

// Perform computation:
var sum = mod.ndarray( N, xptr, 1, 0 );
// returns 6.0

The function has the following additional parameters:

  • ox: starting index for x.

Notes

  • If N <= 0, both main and ndarray methods return 0.0.
  • This package implements routines using WebAssembly. When provided arrays which are not allocated on a dasumpw module memory instance, data must be explicitly copied to module memory prior to computation. Data movement may entail a performance cost, and, thus, if you are using arrays external to module memory, you should prefer using @stdlib/blas/ext/base/dasumpw. However, if working with arrays which are allocated and explicitly managed on module memory, you can achieve better performance when compared to the pure JavaScript implementations found in @stdlib/blas/ext/base/dasumpw. Beware that such performance gains may come at the cost of additional complexity when having to perform manual memory management. Choosing between implementations depends heavily on the particular needs and constraints of your application, with no one choice universally better than the other.

Examples

var discreteUniform = require( '@stdlib/random/array/discrete-uniform' );
var dasumpw = require( '@stdlib/blas/ext/base/wasm/dasumpw' );

var opts = {
    'dtype': 'float64'
};
var x = discreteUniform( 10, 0, 100, opts );
console.log( x );

var sum = dasumpw.ndarray( x.length, x, 1, 0 );
console.log( sum );