time-to-botec

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

nansumpw.js (4070B)

      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 'use strict';
     20 
     21 // MODULES //
     22 
     23 var floor = require( '@stdlib/math/base/special/floor' );
     24 
     25 
     26 // VARIABLES //
     27 
     28 // 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`.):
     29 var BLOCKSIZE = 128;
     30 
     31 
     32 // MAIN //
     33 
     34 /**
     35 * Computes the sum of a strided array elements, ignoring `NaN` values and using pairwise summation.
     36 *
     37 * ## Method
     38 *
     39 * -   This implementation uses pairwise summation, which accrues rounding error `O(log2 N)` instead of `O(N)`. The recursion depth is also `O(log2 N)`.
     40 *
     41 * ## References
     42 *
     43 * -   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).
     44 *
     45 * @private
     46 * @param {PositiveInteger} N - number of indexed elements
     47 * @param {NumericArray} out - two-element output array whose first element is the accumulated sum and whose second element is the accumulated number of summed values
     48 * @param {NumericArray} x - input array
     49 * @param {integer} stride - stride length
     50 * @param {NonNegativeInteger} offset - starting index
     51 * @returns {NumericArray} output array
     52 *
     53 * @example
     54 * var floor = require( '@stdlib/math/base/special/floor' );
     55 *
     56 * var x = [ 2.0, 1.0, 2.0, -2.0, -2.0, 2.0, 3.0, 4.0, NaN, NaN ];
     57 * var N = floor( x.length / 2 );
     58 *
     59 * var out = [ 0.0, 0 ];
     60 * var v = nansumpw( N, out, x, 2, 1 );
     61 * // returns [ 5.0, 4 ]
     62 */
     63 function nansumpw( N, out, x, stride, offset ) {
     64 	var ix;
     65 	var s0;
     66 	var s1;
     67 	var s2;
     68 	var s3;
     69 	var s4;
     70 	var s5;
     71 	var s6;
     72 	var s7;
     73 	var M;
     74 	var s;
     75 	var n;
     76 	var v;
     77 	var i;
     78 
     79 	ix = offset;
     80 	if ( N < 8 ) {
     81 		// Use simple summation...
     82 		s = 0.0;
     83 		n = 0;
     84 		for ( i = 0; i < N; i++ ) {
     85 			v = x[ ix ];
     86 			if ( v === v ) {
     87 				s += v;
     88 				n += 1;
     89 			}
     90 			ix += stride;
     91 		}
     92 		out[ 0 ] += s;
     93 		out[ 1 ] += n;
     94 		return out;
     95 	}
     96 	if ( N <= BLOCKSIZE ) {
     97 		// Sum a block with 8 accumulators (by loop unrolling, we lower the effective blocksize to 16)...
     98 		s0 = 0.0;
     99 		s1 = 0.0;
    100 		s2 = 0.0;
    101 		s3 = 0.0;
    102 		s4 = 0.0;
    103 		s5 = 0.0;
    104 		s6 = 0.0;
    105 		s7 = 0.0;
    106 		n = 0;
    107 
    108 		M = N % 8;
    109 		for ( i = 0; i < N-M; i += 8 ) {
    110 			v = x[ ix ];
    111 			if ( v === v ) {
    112 				s0 += v;
    113 				n += 1;
    114 			}
    115 			ix += stride;
    116 			v = x[ ix ];
    117 			if ( v === v ) {
    118 				s1 += v;
    119 				n += 1;
    120 			}
    121 			ix += stride;
    122 			v = x[ ix ];
    123 			if ( v === v ) {
    124 				s2 += v;
    125 				n += 1;
    126 			}
    127 			ix += stride;
    128 			v = x[ ix ];
    129 			if ( v === v ) {
    130 				s3 += v;
    131 				n += 1;
    132 			}
    133 			ix += stride;
    134 			v = x[ ix ];
    135 			if ( v === v ) {
    136 				s4 += v;
    137 				n += 1;
    138 			}
    139 			ix += stride;
    140 			v = x[ ix ];
    141 			if ( v === v ) {
    142 				s5 += v;
    143 				n += 1;
    144 			}
    145 			ix += stride;
    146 			v = x[ ix ];
    147 			if ( v === v ) {
    148 				s6 += v;
    149 				n += 1;
    150 			}
    151 			ix += stride;
    152 			v = x[ ix ];
    153 			if ( v === v ) {
    154 				s7 += v;
    155 				n += 1;
    156 			}
    157 			ix += stride;
    158 		}
    159 		// Pairwise sum the accumulators:
    160 		s = ((s0+s1) + (s2+s3)) + ((s4+s5) + (s6+s7));
    161 
    162 		// Clean-up loop...
    163 		for ( i; i < N; i++ ) {
    164 			v = x[ ix ];
    165 			if ( v === v ) {
    166 				s += v;
    167 				n += 1;
    168 			}
    169 			ix += stride;
    170 		}
    171 		out[ 0 ] += s;
    172 		out[ 1 ] += n;
    173 		return out;
    174 	}
    175 	// Recurse by dividing by two, but avoiding non-multiples of unroll factor...
    176 	n = floor( N/2 );
    177 	n -= n % 8;
    178 	return nansumpw( n, out, x, stride, ix ) + nansumpw( N-n, out, x, stride, ix+(n*stride) ); // eslint-disable-line max-len
    179 }
    180 
    181 
    182 // EXPORTS //
    183 
    184 module.exports = nansumpw;