Floating-point Bugs in Embedded GNU C Library
Abstract
We report serious bugs in floating-point computations for evaluating elementary functions in the Embedded GNU C Library. For instance, the sine function can return values larger than 1053 in certain rounding modes. Further investigation also exposed faulty implementations in the most recent version of the library, which seemingly fixed some bugs, but only by discarding user-specified rounding-mode requirements.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 19, 2013
- Accession Number
- ADA600185
Entities
People
- Edmund M. Clarke
- Sicun Gao
- Soonho Kong
Organizations
- Carnegie Mellon University