← Back to team overview

ffc team mailing list archive

Re: tabulate_tensor error

 

Kristian Oelgaard wrote:
Quoting Jake Ostien <tostien@xxxxxxxxx>:

Hi,

I am facing a problem when trying to compile a DOLFIN solver using code output from ffc. I am using quadrature since the sheer size of the form overwhelms my machine if I don't, and the resulting output header file currently sits at ~7.1M, which takes about 30 minutes to compile.

Running make yields the following error:

`pkg-config --variable=compiler dolfin` `pkg-config --cflags dolfin` -c main.cpp DGSGPa.h: In member function ‘virtual void UFC_DGSGPaBilinearForm_cell_integral_0::tabulate_tensor(double*, const double* const*, const ufc::cell&) const’:
DGSGPa.h:86551: error: ‘Jinv_03’ was not declared in this scope
DGSGPa.h:86551: error: ‘Jinv_13’ was not declared in this scope
DGSGPa.h:86551: error: ‘Jinv_23’ was not declared in this scope
DGSGPa.h: In member function ‘virtual void UFC_DGSGPaBilinearForm_interior_facet_integral_0::tabulate_tensor(double*, const double* const*, const ufc::cell&, const ufc::cell&, unsigned int, unsigned int) const’:
DGSGPa.h:88266: error: ‘Jinv0_03’ was not declared in this scope
DGSGPa.h:88266: error: ‘Jinv0_13’ was not declared in this scope
DGSGPa.h:88266: error: ‘Jinv0_23’ was not declared in this scope
DGSGPa.h:88272: error: ‘Jinv0_03’ was not declared in this scope
.
.
.

which repeats thousands of times for every 'Jinv*' term that ends in a '3'. I am not that familiar with this part of the code (yet), but I plan on looking at it now. If anyone has any clues of where to look for this error, or may know what is going on, I'd be happy to hear about it.

Looks like there's a bug when generating the names for some of the Jacobian
terms. In 3D the last two indices should only run from 0 - 2. I'll take a look
at it.
This was not a bug in FFC. I had introduced the error in a local definition of the curl operator (on a 2nd order tensor) that took derivatives over the indices 1, 2, 3 (instead of 0, 1, 2). The mysterious '3's came from that bad definition. (Thanks Kristian)

Jake
Kristian


Thanks,

Jake
_______________________________________________
FFC-dev mailing list
FFC-dev@xxxxxxxxxx
http://www.fenics.org/mailman/listinfo/ffc-dev








References