|
From: Thomas R. <tr...@st...> - 2011-11-09 11:17:57
|
Thomas Rast wrote:
> Hi *,
>
> Due to us using valgrind in an exam environment[1] with CGAL[2], I am
> in the unfortunate position that I have to get FP rounding to work
> without restrictions.
>
> Since it's bad enough that we (or to be precise, my future successors)
> have to maintain a valgrind module, I would like to put in some work
> time to get this upstreamed. I am thus asking for your guidance on
> what approach you would want to be taken.
[...]
> From my end it looks as if the current idea is to put an IRExpr for
> the rounding mode field of every affected instruction, and then
> load/restore that IRExpr on demand without looking at it again. Is
> that correct?
No takers? :-(
I have attached new patches that add the extra rounding mode argument
to the Iops for FP add, multiply, etc. Please tell me if git patches
aren't okay, but AFAIK 'patch -p1' should process them correctly.
So far I only did amd64, notably because it's what I need and (apart
from x86) the only thing I can test.
Is that the way to go? Or would you rather see something different?
Assuming it is the right approach: am I supposed to ("blindly"!) hack
the same changes into all front- and backends, or are there
maintainers for each who can help with it?
Cheers,
--
Thomas Rast
trast@{inf,student}.ethz.ch
|