[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [csmith-dev] Floating-point support in csmith
Shafiul, one thing you can do is look for a possible wrong code bug in a
floating-point program reduce it using C-Reduce. I'm not sure anyone
has done this lately. The tiny resulting test case will show you right
away at least one problem that needs to be solved (assuming it does not
reveal a compiler bug).
What's our current story for UB detection of FP programs: Frama-C in
interpreter mode as usual?
On 11/6/15 11:03 PM, Shafiul Azam wrote:
Thank you All very much for responses! To summarize, csmith with --float
can be used to detect crash bugs for the moment, and should NOT be used
for wrong code bugs. Please correct me if I got it wrong.
On Wed, Nov 4, 2015 at 3:23 PM, John Regehr <email@example.com
- Floating-point support was introduced in version 2.2.0
the release note . Floating point support must be turned on
--float command-line option. This note acknowledges work of Dr.
Donaldson for the initial implementation.
Yes, please check with Ally, he and his people have been doing
this. I think basically all of us are interested but it's been a
matter of making time to work on it.
- I conducted a run of csmith 2.2.0 with --float argument on
10.04 (gcc 4.4.3). Number of crash-programs is significantly
Running csmith without --float found 3 crash-bugs, while running
--float found more than 300 crash-bugs within the (almost) same
Nice. There are a few things going on here. As Yang observes,
there's some number <300 of bugs that are getting triggered multiple
times. To find the true set of bugs you can start to report them,
one at a time.
Another thing that is going on is that GCC has built up a huge
resistance to Csmith. We've beaten it up for years now. We have
not beaten on its FP code so of course there's more low-hanging fruit.
I am interested to know current support-level for floating
limitations and any future works you plan regarding float-point
I'll also appreciate any source which explains current support
floating-point in csmith.
Hopefully Ally will chime in. You should the source code too,
there's not that much that deals with FP, as far as I know.