Approximate Reciprocal Square Root with Single - and Half-Precision Floats

  • Matheus M. Susin UNICAMP
  • Lucas Wanner UNICAMP

Resumo


In this work, we compared the precision, speed, and power consumption of the reciprocal square root of a single-precision floating point number, using different approximation techniques. We also devised an equivalent approximation for half-precision floating point numbers, and evaluated its performance across the whole range of positive non-zero 16-bit floating point values.

Referências

IEEE 754-2008 (2008). IEEE Standard for Floating-Point Arithmetic. IEEE Std 754-2008.

Kushner, D. (2002). The wizardry of id. IEEE Spectrum, pages 42–47.

McEniry, C. (2007). The mathematics behind the fast inverse square root function code.

Rau, C. (accessed on 2017-11-16). IEEE 754-based half-precision floating point library. https://half.sourceforge.net/.
Publicado
13/04/2018
SUSIN, Matheus M.; WANNER, Lucas. Approximate Reciprocal Square Root with Single - and Half-Precision Floats. In: ESCOLA REGIONAL DE ALTO DESEMPENHO DE SÃO PAULO (ERAD-SP), 9. , 2018, São José dos Campos. Anais [...]. Porto Alegre: Sociedade Brasileira de Computação, 2018 . p. 49-52. DOI: https://doi.org/10.5753/eradsp.2018.13600.