ffc team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Fri, Oct 14, 2011 at 01:30:27PM +0200, Marie E. Rognes wrote:
> On 10/14/2011 11:37 AM, Kristian Ølgaard wrote:
> >On 14 October 2011 10:51, Anders Logg<logg@xxxxxxxxx>  wrote:
> >>The FFC tests are failing on my laptop. I might have screwed something
> >>up locally but it might also be related to updating to Ubuntu 11.10
> >>last night which has GCC 4.6.1-9ubuntu3.
> >>
> >>Here's the error I get:
> >>
> >>Verifying element 2 of 143:<CG2 on a<interval cell in R1>>
> >>GCC compilation failed.
> >>
> >>
> >>element:<CG2 on a<interval cell in R1>>,
> >>In file included from evaluate_basis.cpp:4:0:
> >>test.h: In member function 'virtual void
> >>test_finite_element_0::evaluate_basis(unsigned int, double*, const
> >>double*, const ufc::cell&) const':
> >>test.h:128:14: error: variable 'tmp0' set but not used
> >>[-Werror=unused-but-set-variable]
> >>test.h:165:14: error: variable 'tmp0' set but not used
> >>[-Werror=unused-but-set-variable]
> >>test.h:202:14: error: variable 'tmp0' set but not used
> >>[-Werror=unused-but-set-variable]
> >>
> >>Anyone else seeing this?
> >Yes, I also get this with the same version of GCC. The problem is code like:
> >
> >       double tmp0 = 0.0;
> >       double tmp1 = 0.0;
> >       double tmp2 = 0.0;
> >       for (unsigned int r = 2; r<  5; r++)
> >       {
> >         tmp0 = 2.0*r*r*(2.0*r-2.0);
> >         tmp1 = 2.0*r*(2.0*r-1.0)*(2.0*r-2.0)/(2.0*r*r*(2.0*r-2.0));
> >         tmp2 = 4.0*r*(r-1.0)*(r-1.0)/(2.0*r*r*(2.0*r-2.0));
> >         basisvalues[r] = (X*basisvalues[r - 1]*tmp1 - basisvalues[r - 2]*tmp2);
> >       }// end loop over 'r'
> >
> >As you see, tmp0 is not used to compute the entry in basisvalues[].
> >However, a value is being assigned to it in
> >
> >tmp0 = 2.0*r*r*(2.0*r-2.0);
> >
> >maybe this is why it is not being removed by the remove_unused()
> >function in cpp.py?
>
> Sounds like an argument for unrolling the loops for 1D (as for 2D
> and 3D) as well. Can fix.

Thanks! The other option would be to dig into remove_unused and it's
not a place I want to go.

--
Anders