Performance differences in different MATLAB interface

[Moritz Jena]

Dear Getfem-users,

due to a change in our system architecture
from Windows 32Bit to 64Bit, our numerical calculation core has to be updated.
Currently we are working with the MATLAB
interface of GetFEM version 4.2. 

During tests with the available MATLAB
interface of GetFEM version 5.2 I noticed that it works much slower
than the interface of version 4.2.

The tests were performed with different
test programs of our numerical calculation core. 
The runtime difference can be determined
to be approximately four times higher.

Can anyone explain to me, using the
attached test program, where this difference could come from?
Are there differences in the implementation
of the functions in the different versions?

With best regards,

Moritz Jena        


Example test-program: Runtime difference:
v4.2 * 3.6 = v5.2 (v4.3: 12.35s vs. v5.2: 44.96s)

%%  ---------- Mesh ----------
L = 10;
m = gfMesh('cartesian', 0:.5:L, 0:.5:1,
0:.5:1);

%% ---------- FEM ----------
n = gf_mesh_get(m, 'dim');
k = 1; % Degree

% FEM u
f =  gf_fem(sprintf('FEM_QK(%d,%d)',n,k));
mfu = gf_mesh_fem(m,3); gf_mesh_fem_set(mfu,
'fem', f);

% FEM Von Mises
f2 = gf_fem(sprintf('FEM_QK_DISCONTINUOUS(%d,1)',
n));
mfvm = gf_mesh_fem(m,1);  gf_mesh_fem_set(mfvm,
'fem', f2);

%% ---------- IM ----------
k_i = 2*k;
INTEG_TYPE = sprintf('IM_GAUSS_PARALLELEPIPED(%d,%d)',n,k_i);
im = gf_integ(INTEG_TYPE); mim = gf_mesh_im(m,im);

%% ---------- region ----------
P = get(m,'pts');
boundary_left = gf_mesh_get(m, 'faces
from pid', find(abs(P(1,:))<1e-6));
boundary_right = gf_mesh_get(m, 'faces
from pid', find(abs(P(1,:) - L)<1e-6));
gf_mesh_set(m, 'region', 1, boundary_left);
gf_mesh_set(m, 'region', 2, boundary_right);

%% ---------- data ----------
E = 200000;
nu = 0.3;
% Lamé
lambda = nu*E/((1+nu)*(1-2*nu));
mu = E/(2*(1+nu));
%% ---------- model ----------
md = gf_model('real');

gf_model_set(md, 'add fem variable',
'u', mfu);
gf_model_set(md, 'add initialized data',
'lambda', lambda);
gf_model_set(md, 'add initialized data',
'mu', mu);

gf_model_set(md, 'add isotropic linearized
elasticity brick', ...
    mim, 'u', 'lambda', 'mu');

% homogenous Dirichlet on the left (region
1)
% gf_model_set(md, 'add Dirichlet condition
with penalization', ...
%       mim, 'u', 1e10,
1);
gf_model_set(md, 'add Dirichlet condition
with multipliers', ...
    mim, 'u', mfu, 1);

alt = 1; %alternative 1 Force, alternative
2 Displacement

% alternative 1: Force on the right
side (region 2)
if alt ==1
    F = [0 -100 0];
    gf_model_set(md, 'add
initialized data', 'Kraft', F);
    gf_model_set(md, 'add
source term brick', mim, 'u', 'Kraft', 2);
else
    % alternative 2: Displacement
on the right side (region 2)
    diri = [0 -1.5 0];
    gf_model_set(md, 'add
initialized data', 'dirichletdata', diri);
    gf_model_set(md, 'add
Dirichlet condition with penalization', ...
        mim, 'u',
1e10, 2);
    %      
gf_model_set(md, 'add Dirichlet condition with multipliers', ...
    %     mim, 'u',
mfu, 2, 'dirichletdata');
end

%% ---------- solving ----------
gf_model_get(md, 'solve');

u = gf_model_get(md, 'variable', 'u');
VM = gf_model_get(md, 'compute isotropic
linearized Von Mises or Tresca', 'u', 'lambda', 'mu', mfvm);

figure
gf_plot(mfvm,VM, 'deformation',u, 'deformation_mf',mfu,
'deformed_mesh', ...
    'on', 'cvlst', get(m,
'outer faces'), 'deformation_scale', 1, ...
    'disp_options', 'off');
colorbar;
xlabel('x'); ylabel('y'); zlabel('z');