Automatic parallel

General Discussion
1

Hello, everyone! Is there some tricks on FreeFEM such that I write codes in serial, but it works in parallel.

2

There is very little to be done to get a sequential script working in parallel using the PETSc interface.
Here is an example for the Poisson equation.

mesh Th = square(100, 100);
fespace Vh(Th, P1);
varf vPb(u, v) = int2d(Th)(dx(u)*dx(v) + dy(u)*dy(v)) + int2d(Th)(v) + on(1, u = 0);
matrix A;
A = vPb(Vh, Vh);
real[int] rhs = vPb(0, Vh);
Vh sol;
sol[] = A^-1 * rhs;
plot(sol, cmm = "Sequential solution");

And in parallel.

load "PETSc"
include "macro_ddm.idp"
mesh Th = square(100, 100);
fespace Vh(Th, P1);
varf vPb(u, v) = int2d(Th)(dx(u)*dx(v) + dy(u)*dy(v)) + int2d(Th)(v) + on(1, u = 0);
Mat A;
createMat(Th, A, P1);
real[int] rhs = vPb(0, Vh);
A = vPb(Vh, Vh);
Vh sol;
sol[] = A^-1 * rhs;
plotD(Th, sol, cmm = "Parallel solution");

If you want to understand this better, you could have a look here and there.

3

Thanks for your reply. How about multiple variables such as Stokes equations, and do we have to use a matrix?

4

It works the same, see FreeFem-sources/stokes-2d-PETSc.edp at develop · FreeFem/FreeFem-sources · GitHub. By “do we have to use a matrix?”, are you asking if it works with the keyword solve instead of varf? If that is the case, then no, it does not work with solve and you have to assemble varf instead. Note that any solve can be converted into a varf.

5

Thanks for your kindness!