time-to-botec

Benchmark sampling in different programming languages
Log | Files | Refs | README

ndarray.js (4571B)

      1 /**
      2 * @license Apache-2.0
      3 *
      4 * Copyright (c) 2020 The Stdlib Authors.
      5 *
      6 * Licensed under the Apache License, Version 2.0 (the "License");
      7 * you may not use this file except in compliance with the License.
      8 * You may obtain a copy of the License at
      9 *
     10 *    http://www.apache.org/licenses/LICENSE-2.0
     11 *
     12 * Unless required by applicable law or agreed to in writing, software
     13 * distributed under the License is distributed on an "AS IS" BASIS,
     14 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
     15 * See the License for the specific language governing permissions and
     16 * limitations under the License.
     17 */
     18 
     19 /* eslint-disable max-len */
     20 
     21 'use strict';
     22 
     23 // MODULES //
     24 
     25 var float64ToFloat32 = require( '@stdlib/number/float64/base/to-float32' );
     26 var floor = require( '@stdlib/math/base/special/floor' );
     27 
     28 
     29 // VARIABLES //
     30 
     31 // Blocksize for pairwise summation (NOTE: decreasing the blocksize decreases rounding error as more pairs are summed, but also decreases performance. Because the inner loop is unrolled eight times, the blocksize is effectively `16`.):
     32 var BLOCKSIZE = 128;
     33 
     34 
     35 // MAIN //
     36 
     37 /**
     38 * Adds a constant to each single-precision floating-point strided array element and computes the sum using pairwise summation.
     39 *
     40 * ## Method
     41 *
     42 * -   This implementation uses pairwise summation, which accrues rounding error `O(log2 N)` instead of `O(N)`. The recursion depth is also `O(log2 N)`.
     43 *
     44 * ## References
     45 *
     46 * -   Higham, Nicholas J. 1993. "The Accuracy of Floating Point Summation." _SIAM Journal on Scientific Computing_ 14 (4): 783–99. doi:[10.1137/0914050](https://doi.org/10.1137/0914050).
     47 *
     48 * @param {PositiveInteger} N - number of indexed elements
     49 * @param {number} alpha - constant
     50 * @param {Float32Array} x - input array
     51 * @param {integer} stride - stride length
     52 * @param {NonNegativeInteger} offset - starting index
     53 * @returns {number} sum
     54 *
     55 * @example
     56 * var Float32Array = require( '@stdlib/array/float32' );
     57 * var floor = require( '@stdlib/math/base/special/floor' );
     58 *
     59 * var x = new Float32Array( [ 2.0, 1.0, 2.0, -2.0, -2.0, 2.0, 3.0, 4.0 ] );
     60 * var N = floor( x.length / 2 );
     61 *
     62 * var v = sapxsumpw( N, 5.0, x, 2, 1 );
     63 * // returns 25.0
     64 */
     65 function sapxsumpw( N, alpha, x, stride, offset ) {
     66 	var ix;
     67 	var s0;
     68 	var s1;
     69 	var s2;
     70 	var s3;
     71 	var s4;
     72 	var s5;
     73 	var s6;
     74 	var s7;
     75 	var M;
     76 	var s;
     77 	var n;
     78 	var i;
     79 
     80 	if ( N <= 0 ) {
     81 		return 0.0;
     82 	}
     83 	if ( N === 1 || stride === 0 ) {
     84 		return float64ToFloat32( alpha + x[ offset ] );
     85 	}
     86 	ix = offset;
     87 	if ( N < 8 ) {
     88 		// Use simple summation...
     89 		s = 0.0;
     90 		for ( i = 0; i < N; i++ ) {
     91 			s = float64ToFloat32( s + float64ToFloat32( alpha + x[ ix ] ) );
     92 			ix += stride;
     93 		}
     94 		return s;
     95 	}
     96 	if ( N <= BLOCKSIZE ) {
     97 		// Sum a block with 8 accumulators (by loop unrolling, we lower the effective blocksize to 16)...
     98 		s0 = float64ToFloat32( alpha + x[ ix ] );
     99 		s1 = float64ToFloat32( alpha + x[ ix+stride ] );
    100 		s2 = float64ToFloat32( alpha + x[ ix+(2*stride) ] );
    101 		s3 = float64ToFloat32( alpha + x[ ix+(3*stride) ] );
    102 		s4 = float64ToFloat32( alpha + x[ ix+(4*stride) ] );
    103 		s5 = float64ToFloat32( alpha + x[ ix+(5*stride) ] );
    104 		s6 = float64ToFloat32( alpha + x[ ix+(6*stride) ] );
    105 		s7 = float64ToFloat32( alpha + x[ ix+(7*stride) ] );
    106 		ix += 8 * stride;
    107 
    108 		M = N % 8;
    109 		for ( i = 8; i < N-M; i += 8 ) {
    110 			s0 = float64ToFloat32( s0 + float64ToFloat32( alpha + x[ ix ] ) );
    111 			s1 = float64ToFloat32( s1 + float64ToFloat32( alpha + x[ ix+stride ] ) );
    112 			s2 = float64ToFloat32( s2 + float64ToFloat32( alpha + x[ ix+(2*stride) ] ) );
    113 			s3 = float64ToFloat32( s3 + float64ToFloat32( alpha + x[ ix+(3*stride) ] ) );
    114 			s4 = float64ToFloat32( s4 + float64ToFloat32( alpha + x[ ix+(4*stride) ] ) );
    115 			s5 = float64ToFloat32( s5 + float64ToFloat32( alpha + x[ ix+(5*stride) ] ) );
    116 			s6 = float64ToFloat32( s6 + float64ToFloat32( alpha + x[ ix+(6*stride) ] ) );
    117 			s7 = float64ToFloat32( s7 + float64ToFloat32( alpha + x[ ix+(7*stride) ] ) );
    118 			ix += 8 * stride;
    119 		}
    120 		// Pairwise sum the accumulators:
    121 		s = float64ToFloat32( float64ToFloat32( float64ToFloat32(s0+s1) + float64ToFloat32(s2+s3) ) + float64ToFloat32( float64ToFloat32(s4+s5) + float64ToFloat32(s6+s7) ) );
    122 
    123 		// Clean-up loop...
    124 		for ( i; i < N; i++ ) {
    125 			s = float64ToFloat32( s + float64ToFloat32( alpha + x[ ix ] ) );
    126 			ix += stride;
    127 		}
    128 		return s;
    129 	}
    130 	// Recurse by dividing by two, but avoiding non-multiples of unroll factor...
    131 	n = floor( N/2 );
    132 	n -= n % 8;
    133 	return float64ToFloat32( sapxsumpw( n, alpha, x, stride, ix ) + sapxsumpw( N-n, alpha, x, stride, ix+(n*stride) ) );
    134 }
    135 
    136 
    137 // EXPORTS //
    138 
    139 module.exports = sapxsumpw;