top | item 39517573

Herbgrind analyzes binaries to find inaccurate floating point expressions

70 points| bshanks | 2 years ago |herbgrind.ucsd.edu

13 comments

order

aSanchezStern|2 years ago

Wow, didn't expect this tool to be on Hacker News six years after publication! Herbgrind author here, ask me any questions you like! Also, if you're interested in this stuff, the Herbie project (which I also worked on) for numerical program synthesis is also really helpful for writing numerical code, and has had a lot of development over the years.

zokier|2 years ago

What do you think programming languages/libraries/tools could do to make floats less scary for the common people? I feel there is lot of superstition around floats, some of it more well deserved than others, but it leads people to approach fp with great distrust.

Of course herbie is already good step here, but it is still somewhat niche.

phkahler|2 years ago

Is it feasible to run this on something large like Solvespace[1] (CAD) which is ~5MB executable? Or would we just get an insanely long list of issues?

[1] https://solvespace.com/index.pl

There are hundreds of numerical algorithms in there, and we have some bugs that might be related to this kind of implementation error.

99112000|2 years ago

I've tried out HerbGrind but it often gets stuck and reports an error E_STONED.

touisteur|2 years ago

Thinking of all the people (some even actual friends) worked so long on numerical stability on scalar floating point operations, nearing a time when model-checking and proof tool are actually usable by the really-motivated developer ; thinking of them witnessing the arrival of GPUs and Tensor Cores, mixed-precision-everywhere and the tools now needing to scale so far it seems like starting from scratch...