Efficiently Swapping Columns in a Matrix
조회 수: 13 (최근 30일)
이전 댓글 표시
Is there a more efficient/faster way to switch out two columns in a matrix than calling:
A(:,[i j]) = A(:,[j i]);
If anyone knew of faster implementation (possibly even using MEX), I would greatly appreciate them sharing.
댓글 수: 3
James Tursa
2016년 6월 23일
편집: James Tursa
2016년 6월 23일
This could easily be done in a mex routine in-place (I can post the code if you want). I don't know how much faster it would run. The caveat is that if A is shared with another variable then you would have unwanted side effects. Do you know if A is shared with any other variable?
채택된 답변
James Tursa
2016년 6월 24일
편집: James Tursa
2016년 6월 28일
Here is the mex code to swap columns in-place. Note that it is up to the user to make sure the input matrix is not shared with another variable, since every shared variable would have their columns swapped in that case.
/* swapcolumns.c Swaps columns of X in-place.
* Relies on user to make sure X is not shared with another variable
* since there is no checking for this in the code.
* Syntax: swapcolumns(X,i,j)
* X = a matrix (any standard class ... no objects)
* i,j = columns to swap
* Programmer: James Tursa
* Date: June 24, 2016
*/
#include "mex.h"
void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
{
unsigned char b;
unsigned char *data, *target, *source;
size_t M, N, i, j, k, c, bytes;
if( nrhs == 0 ) {
mexPrintf("swapcolumns -> Swaps columns of X in-place.\n");
mexPrintf("Relies on user to make sure X is not shared with another variable\n");
mexPrintf("since there is no checking for this in the code.\n");
mexPrintf("Syntax: swapcolumns(X,i,j)\n");
mexPrintf(" X = a matrix (any standard class ... no objects)\n");
mexPrintf(" i,j = columns to swap\n");
return;
}
if( nrhs != 3 ) {
mexErrMsgTxt("Need exactly 3 inputs");
}
if( nlhs > 0 ) {
mexErrMsgTxt("Too many outputs\n");
}
if( !(mxIsNumeric(prhs[0]) || mxIsChar(prhs[0]) || mxIsLogical(prhs[0]) ||
mxIsCell(prhs[0])) ) {
mexErrMsgTxt("1st argument needs to be standard class");
}
M = mxGetM(prhs[0]);
N = mxGetN(prhs[0]);
if( M == 0 || N == 0 ) {
return;
}
i = (size_t) mxGetScalar(prhs[1]);
j = (size_t) mxGetScalar(prhs[2]);
if( i > N || j > N ) {
mexErrMsgTxt("Column index(es) are too large");
}
if( i == 0 || j == 0 ) {
mexErrMsgTxt("Column index(es) cannot be 0");
}
data = mxGetData(prhs[0]);
bytes = M * mxGetElementSize(prhs[0]);
c = 2;
while( c-- ) {
target = data + (i-1) * bytes;
source = data + (j-1) * bytes;
for( k=0; k<bytes; k++ ) {
b = *target;
*target++ = *source;
*source++ = b;
}
if( mxIsNumeric(prhs[0]) && mxIsComplex(prhs[0]) ) {
data = mxGetImagData(prhs[0]);
} else {
break;
}
}
}
추가 답변 (3개)
the cyclist
2016년 6월 23일
Are you certain you actually have to "physically" swap the columns every time? Could you instead simply keep track of the swaps, and then index into the resulting matrix, or do one massive multi-column swap at the end? Just a thought.
댓글 수: 2
Philip Borghesani
2016년 6월 23일
편집: Philip Borghesani
2016년 6월 23일
Does your data need to be in a matrix? Swaping two elements of a cell array will be much faster for large element sizes and will be similar to an array list or general list.
%swaptime.m
sz=1e8;
m=ones(sz,5);
for col=1:5
c{col}=ones(sz,1);
end
tic
m(:,[2,5])=m(:,[5,2]);
toc
tic
c([2,5])=c([5,2]);
toc
>> swaptime
Elapsed time is 1.295767 seconds.
Elapsed time is 0.000899 seconds.
댓글 수: 2
Roger Stafford
2016년 6월 25일
It is possible that matlab accomplishes A(:,[i j]) = A(:,[j i]); by the compiler equivalent of
t1 = A(:,i);
t2 = A(:,j);
A(:,i) = t2;
A(:,j) = t1;
which requires 4*n transfers. If you wrote a mex file that does the equivalent of
t = A(:,i);
A(:,i) = A(:,j);
A(:,j) = t;
it would take only 3*n transfers. I cannot think of any more significant gains that you might make on this kind of swapping. Swapping inherently involves a transfer to temporary storage, transfer of another into the previous location, and finally transfer from temporary storage into the second location.
댓글 수: 0
참고 항목
제품
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!