problem implementing central finite difference on multicore computer?

3 次查看（过去 30 天）
Per
Per 2012-8-20
Hi,
I am trying to implement the central finite difference scheme for use on a multicore computer. The computer has 12 cores.
Currently I have a code that takes several days to run and any help in improving the performance would be helpful.
My finite difference function looks like this:
%nx,ny,nz = number of grid points in x,y,z direction
%dx = grid spacing
%v = velocity model
%ricker = ricker wavelet
%nt = number of time samples
%dt = number of time samples
%xs,ys,zs = source position
% receivers = receiver positions in y direction
function u = fd3d(nx,ny,nz,dx,v,ricker,nt,dt,xs,ys,zs,receivers)
p0 = zeros(nx,ny,nz); p1 = p0; p2 = p0;
cns = 6*(dt*v).^2;
u = zeros(length(receivers),nt);
alpha=dt/dx;
for it = 1:nt
p2 = 2*p1 - p0 + cns.*del2(p1,dx);
p2(xs,ys,zs) = p2(xs,ys,zs) + ricker(it);
%ABC BOUNDARY CONDITIONS
p2(1,:,:) = p1(1,:,:) + alpha.*v(1,:,:).*(p1(2,:,:)-p1(1,:,:));
p2(end,:,:) = p1(end,:,:) + alpha.*v(end,:,:).*(p1(end-1,:,:)-p1(end,:,:));
p2(:,1,:) = p1(:,1,:) + alpha.*v(:,1,:).*(p1(:,2,:)-p1(:,1,:));
p2(:,end,:) = p1(:,end,:) + alpha.*v(:,end,:).*(p1(:,end-1,:)-p1(:,end,:));
p2(:,:,1) = p1(:,:,1) + alpha.*v(:,:,1).*(p1(:,:,2)-p1(:,:,1));
p2(:,:,end) = p1(:,:,end) + alpha.*v(:,:,end).*(p1(:,:,end-1)-p1(:,:,end));
a = p2(xs,receivers,zs); extract seismogram from current timestep
u(:,it) = a'; %store seismogram
p0 = p1; p1 = p2; %update pressure fields
end
The function that takes the longest time to compute is the builtin del2 function. I was hoping if anyone had any suggestion on a function that can be used instead of del2 that works together with the parallel computation toolbox.

请先登录，再回答此问题。

回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Graphics Performance 的更多信息

标签

产品

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by