FPBench includes a suite of benchmarks, drawn
from published floating-point research.
Benchmark sources
Rosa
37
Herbie
28
Salsa
25
FPTaylor
21
Features used
Arithmetic
111
Temporaries
57
Comparison
33
Loops
28
Exponents
16
Trigonometry
15
Conditionals
10
Domains
Textbooks
28
Mathematics
24
Controls
10
Science
10
(unknown)
39
These benchmarks have been contributed by multiple research
groups. These contributions are listed here in chronological order.
Each listed paper has all its evaluation examples incorporated into
FPBench.
For each benchmark, FPBench records its name (preferably as used
in a paper), descriptive information such as citations, and finally
semantic metadata such as the precision used and preconditions.
The CPU group at the University of Utah
Features used
Arithmetic
21
Temporaries
9
Exponents
3
Trigonometry
2
Domains
(unknown)
21
The Center for Parallel
Computing at Utah group's FPTaylor project contributed 21
benchmarks, many short textbook examples and mathematical
computations.
Rigorous Estimation of Floating-Point Round-off Errors with Symbolic Taylor Expansions
Alexey Solovyev, Charlie Jacobsen, Zvonimir Rakamaric, Ganesh Gopalakrishnan, FM'15
The AVA group at the Max Plank Institute for Software Systems
Features used
Arithmetic
37
Temporaries
23
Comparison
8
Conditionals
5
Loops
3
Exponents
1
Trigonometry
1
Domains
Science
10
Mathematics
5
Controls
4
(unknown)
18
The Automated Verification and Approximation group's Rosa project
contributed 37 benchmarks, mostly drawn from scientific
computations and control theory.
Sound Compilation of the Reals
Eva Darulova, Viktor Kuncak, POPL'14