https://bugs.kde.org/show_bug.cgi?id=214950
--- Comment #17 from Bruno Lathuilière <bruno.lathuili...@edf.fr> ---
As a side product of the valgrind tool verrou, we have a similar functionality
(activated by default) for all emulated rounding mode. So if the rounding-mode
nearest is selected (by default), the behavior of verrou and exp-floattrap is
similar.
The specialized tool exp-floattrap has several advantages :
- exp-floattrap deals with more IOP (by example Iop_Recip*, Iop_RSqst*,
Iop_Neg* ... ).
- I think (I did not check) exp-floattrap should be faster than verrou
as the NaN detection is done with IR. In verrou it is done inside the dirty
call to a function which replace the floating point operation.
- exp-floattrap should be more portable than verrou.
There one point where verrou is maybe more user friendly (concerning only the
nan/inf detection). Indeed with option --libm=instrumented (not yet the
default), the dynamic libm call are intercepted and so the nan/inf detection
inside a libm call are ignored : there is only one detection on the result.
To conclude both tools are complementary but there is overlapping, so it is
good to keep in mind. As I just discovered this tool, I think it could be a
good idea to add a link in this page :
https://valgrind.org/downloads/variants.html
--
You are receiving this mail because:
You are watching all bug changes.
