Fast exact summation using small and large superaccumulators

Radford M. Neal, Dept. of Statistical Sciences and Dept. of Computer Science, University of Toronto

I present two new methods for exactly summing a set of floating-point numbers, and then correctly rounding to the nearest floating-point number. Higher accuracy than simple summation (rounding after each addition) is important in many applications, such as finding the sample mean of data. Exact summation also guarantees identical results with parallel and serial implementations, since the exact sum is independent of order. The new methods use variations on the concept of a "superaccumulator" - a large fixed-point number that can exactly represent the sum of any reasonable number of floating-point values. One method uses a "small" superaccumulator with sixty-seven 64-bit chunks, each with 32-bit overlap with the next chunk, allowing carry propagation to be done infrequently. The small superaccumulator is used alone when summing a small number of terms. For big summations, a "large" superaccumulator is used as well. It consists of 4096 64-bit chunks, one for every possible combination of exponent bits and sign bit, plus counts of when each chunk needs to be transferred to the small superaccumulator. To add a term to the large superaccumulator, only a single chunk and its associated count need to be updated, which takes very few instructions if carefully implemented. On modern 64-bit processors, exactly summing a large array using this combination of large and small superaccumulators takes less than twice the time of simple, inexact, ordered summation, with a serial implementation. A parallel implementation using a small number of processor cores can be expected to perform exact summation of large arrays at a speed that reaches the limit imposed by memory bandwidth. Some common methods that attempt to improve accuracy without being exact may therefore be pointless, at least for large summations, since they are slower than computing the sum exactly.

Technical report, 20 May 2015, 22 pages: pdf.

Also available as arXiv:1505.05571.

The tests in the paper were done using these C programs.