dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Tue, Oct 06, 2009 at 10:27:01PM +0200, Johan Hake wrote:
> On Tuesday 06 October 2009 22:18:04 Anders Logg wrote:
> > On Tue, Oct 06, 2009 at 09:13:45PM +0100, Garth N. Wells wrote:
> > > Anders Logg wrote:
> > > > On Tue, Oct 06, 2009 at 10:05:24PM +0200, Johan Hake wrote:
> > > >> On Tuesday 06 October 2009 21:58:29 Andy Ray Terrel wrote:
> > > >>> On Tue, Oct 6, 2009 at 2:43 PM, Anders Logg <logg@xxxxxxxxx> wrote:
> > > >>>> On Tue, Oct 06, 2009 at 02:28:40PM -0500, Andy Ray Terrel wrote:
> > > >>>>> Sorry for being dense but computing the SHA1 hash for what
> > > >>>>> essentially is a string compare seems a bit overkill.
> > > >>>>
> > > >>>> I want to avoid a call to strcmp in a piece of code that needs to be
> > > >>>> fast. What I do (it seems to work now) is to compute the 20 byte
> > > >>>> SHA-1 hash from the ~40 byte finite element signature which is
> > > >>>> typically something like
> > > >>>>
> > > >>>>  FiniteElement('Lagrange', 'triangle', 1)
> > > >>>>
> > > >>>> I then cast the SHA-1 hash to an 8 byte void* and use that for
> > > >>>> comparison. This may not be very safe but this is not exactly
> > > >>>> cryptography. The number of possible hashes is the same as the
> > > >>>> address space so it should be fine.
> > > >>>
> > > >>> Yes I guess if you have a large number of comparisons it might be
> > > >>> faster.  I forget there are a lot of signature checks for the
> > > >>> function space stuff.
> > > >>>
> > > >>> Computing the hash is O(n) just like strcmp, it really depends on the
> > > >>> constants.  I would just do the timing before I said anything.
> > > >>
> > > >> Would it be an idea to extend ufc::finite_element with a
> > > >> hash_signature()? Then you can compute the hash using hashlib in
> > > >> python during code creation.
> > > >>
> > > >> Johan
> > > >
> > > > Sounds like a good idea. Only it requires changes to DOLFIN, UFC, FFC
> > > > and SyFi...
> > > >
> > > > Are there any objections to this?
> > >
> > > I'm not sure that this is a good idea. It complicates UFC.
>
> Complicates? One function returning a string? But I agree with Anders that it
> requires a lot to change and that it therefore is a good thing to think about
> it.

I tried Boost now, which simplifies the implementation a great deal
since boost can work directly with std::string instead of unsigned
char*.

I agree it's not a big complication to UFC but its a complication
whenever we need to rework the design of the generated code.

--
Anders


> > It's either that or using boost. I'm looking at both options.
> >
> > Calling boost is simpler since it requires the smallest change. If we
> > decide to go with that option now, we can keep this in mind for the
> > next update of UFC.
>
> Sounds good.
>
> Johan
>

Attachment: signature.asc
Description: Digital signature