dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

I think it will improve run-time for high order. Just look at our
timings in the FFC paper (Table V). The speedup flattens out for high
order where it should increase. For q = 8, we generate something like
25000 lines of code...

But I agree, level 3 BLAS will be even better.

/Anders

On Tue, Sep 27, 2005 at 09:46:09AM -0500, Robert C. Kirby wrote:
> Level 2 blas might not improve run-time (should improve compile time),
> depending on how many zeros FFC finds.  Calling N level 2 BLAS operations is
> *much* slower than one level 3 BLAS with N columns.
> 
> But it's a start.
> 
> 
>      We should definitely do this. I'll probably do level 2 BLAS first,
>      since then I only have to modify the output from FFC to use BLAS
>      calls. This will be an immediate benefit for Johan Hoffman with 3D
>      Navier-Stokes since compiling the code generated by FFC (with gcc) is
>      currently the bottle-neck. Hopefully, I can implement this by the end
>      of this week.
> 
>      /Anders
> 
>      _______________________________________________
>      DOLFIN-dev mailing list
>      _D_O_L_F_I_N_-_d_e_v_@_f_e_n_i_c_s_._o_r_g
>      _h_t_t_p_:_/_/_w_w_w_._f_e_n_i_c_s_._o_r_g_/_c_g_i_-_b_i_n_/_m_a_i_l_m_a_n_/_l_i_s_t_i_n_f_o_/_d_o_l_f_i_n_-_d_e_v
> 
> 
> Rob Kirby
> 
> "Mathematical software should be mathematical."
> 
> 
> 

> _______________________________________________
> DOLFIN-dev mailing list
> DOLFIN-dev@xxxxxxxxxx
> http://www.fenics.org/cgi-bin/mailman/listinfo/dolfin-dev


-- 
Anders Logg
Research Assistant Professor
Toyota Technological Institute at Chicago
http://www.tti-c.org/logg/