CFD/hypre
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Rewrote for speed and use with sparse matrices

Browse Source
This commit is contained in:
falgout 1996-12-16 23:47:00 +00:00
parent 7d76223062
commit dc45137eac
2 changed files with 58 additions and 70 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

64
seq_linear_solvers/amg/tools/manalyze.m
View File

@ -2,42 +2,36 @@ function [nv,ia,ja,a] = manalyze(A);
%-----------------------------------------
%function [nv,ia,ja,a] = manalyze(A);
% Analyzes matrix A, yields ysmp format
%
% A must be a matrix in "full" form
%-----------------------------------------
[nv, cols]=size(A);
ia=zeros(nv+1,1);
ip = zeros(nv,1);
iu = ip;
iv = ia;
u = rand(nv,1);
f = ip;
disp('analyzing matrix')
iaend=1;
a=[];
ja=[];
for k=1:nv,
if rem(k,100) == 0, disp(['Row ' num2str(k)]);end
ia(k)=iaend;
nzind = find(A(k,:)~=0);
numnz = length(nzind);
iaend = iaend+numnz;
ctr = find(nzind==k);
lft = nzind(1:ctr-1);
rowind = [nzind(ctr:numnz) lft];
ja = [ja rowind];
a = [a A(k,rowind)];
% determine row/column indices of nonzero entries
[ja, I, a] = find(A');
nv = size(A, 1);
na = size(a, 1);
% compute ia and put diagonal entry first on each row
ia = zeros(nv+1, 1);
ia(1) = 1;
for i = 1 : nv
k = ia(i);
while (k <= na)
if (I(k) ~= i)
break;
end
% swap diagonal entry with first entry
if (ja(k) == i)
tmp = a(ia(i));
a(ia(i)) = a(k);
a(k) = tmp;
tmp = ja(ia(i));
ja(ia(i)) = ja(k);
ja(k) = tmp;
end
k = k + 1;
end
ia(i+1) = k;
end
ia(nv+1)=iaend;
ja=ja';
a=a';
nu = 1; % there is but one function
np = nv; % number of points equals number vars
iu = iu+1; % (all variables are unknown number 1)
ip = [1:nv]'; % each variable is at it's own point
iv = [1:nv+1]'; % each variable is at it's own point

64
seq_ls/amg/tools/manalyze.m
View File

@ -2,42 +2,36 @@ function [nv,ia,ja,a] = manalyze(A);
%-----------------------------------------
%function [nv,ia,ja,a] = manalyze(A);
% Analyzes matrix A, yields ysmp format
%
% A must be a matrix in "full" form
%-----------------------------------------
[nv, cols]=size(A);
ia=zeros(nv+1,1);
ip = zeros(nv,1);
iu = ip;
iv = ia;
u = rand(nv,1);
f = ip;
disp('analyzing matrix')
iaend=1;
a=[];
ja=[];
for k=1:nv,
if rem(k,100) == 0, disp(['Row ' num2str(k)]);end
ia(k)=iaend;
nzind = find(A(k,:)~=0);
numnz = length(nzind);
iaend = iaend+numnz;
ctr = find(nzind==k);
lft = nzind(1:ctr-1);
rowind = [nzind(ctr:numnz) lft];
ja = [ja rowind];
a = [a A(k,rowind)];
% determine row/column indices of nonzero entries
[ja, I, a] = find(A');
nv = size(A, 1);
na = size(a, 1);
% compute ia and put diagonal entry first on each row
ia = zeros(nv+1, 1);
ia(1) = 1;
for i = 1 : nv
k = ia(i);
while (k <= na)
if (I(k) ~= i)
break;
end
% swap diagonal entry with first entry
if (ja(k) == i)
tmp = a(ia(i));
a(ia(i)) = a(k);
a(k) = tmp;
tmp = ja(ia(i));
ja(ia(i)) = ja(k);
ja(k) = tmp;
end
k = k + 1;
end
ia(i+1) = k;
end
ia(nv+1)=iaend;
ja=ja';
a=a';
nu = 1; % there is but one function
np = nv; % number of points equals number vars
iu = iu+1; % (all variables are unknown number 1)
ip = [1:nv]'; % each variable is at it's own point
iv = [1:nv+1]'; % each variable is at it's own point