Published April 12, 2024 | Version v3
Software Open

Numerical Fuzz: A Type System for Rounding Error Analysis



This is the artifact for NumFuzz ("Numerical Fuzz"), a prototype implementation of the type system and floating-point error analysis tool described in the paper Numerical Fuzz: A Type System for Rounding Error Analysis.
This artifact supports the following claim made in the Evaluation section (Section 6.2) of the paper: **compared to state-of-the-art tools that soundly and automatically bound floating-point errors, NumFuzz provides practically useful---and usually better---relative error bounds in at least an order of magnitude less time**.
This artifact supports this claim by automatically generating floating-point error bounds using NumFuzz, FPTaylor, and Gappa for the 17 benchmark problems listed in Table 1 of Section 6.2, and by reporting the timing for each tool on each benchmark.
We can't guarantee that this artifact will, on every machine, produce the exact timing values reported in Table 1 of Section 6.2 for each of the tools on all of the benchmarks. However, this artifact should support the claim that NumFuzz generates floating-point error bounds at least an order of magnitude faster than FPTaylor and Gappa on most benchmarks.
This artifact can also be used to reproduce the NumFuzz results reported in Tables 2 & 3 of Section 6.2.
See the README for details on how to use this artifact. 


Files (3.3 MB)

Name Size Download all
3.3 MB Download
19.7 kB Preview Download

Additional details




Programming language