Skip to content

Instantly share code, notes, and snippets.

@phillipberndt
Forked from IainNZ/gist:9b5f1eb1bcf923ed02d9
Last active August 29, 2015 14:05
Show Gist options
  • Save phillipberndt/7dc0aed7eb855f900f0d to your computer and use it in GitHub Desktop.
Save phillipberndt/7dc0aed7eb855f900f0d to your computer and use it in GitHub Desktop.
function magic!(M::Array{Int}, n::Int)
# Fast magic square generation
#
#
if n % 2 == 1
# Odd-order magic square
# Algorithm: http://en.wikipedia.org/wiki/Magic_square#Method_for_constructing_a_magic_square_of_odd_order
#
ndiv2 = div(n, 2);
for I = 1:n
intermediate1 = (I-1+ndiv2) % n
intermediate2 = (2I - 2) % n
@simd for J = 1:n
# Using these instead of modulo operations gives a huge speed improvement
# by Simon Kornblith, https://gist.github.com/simonster/6195af68c6df33ca965d
x = ifelse(J + intermediate1 < n, J + intermediate1, J + intermediate1 - n)
y = ifelse(J + intermediate2 < n, J + intermediate2, J + intermediate2 - n)
@inbounds M[J, I] = n*x + y + 1
end
end
elseif n % 4 == 0
# Doubly-even order magic square
# Algorithm: http://en.wikipedia.org/wiki/Magic_square#A_method_of_constructing_a_magic_square_of_doubly_even_order
# http://hal.archives-ouvertes.fr/docs/00/94/51/29/PDF/magicsquarev2.pdf
# by Iain Dunning, https://gist.github.com/IainNZ/9b5f1eb1bcf923ed02d9
for I = 1:n
In = (I-1)*n
@simd for J = 1:n
@inbounds M[I,J] = In+J
end
end
for outer_col = 1:4:n, outer_row = 1:4:n
# Sides
for i in (1,2), j in (0,3)
@inbounds M[outer_row+i,outer_col+j] = n^2 - M[outer_row+i,outer_col+j]
end
# Top and bottom
for i in (0,3), j in (1,2)
@inbounds M[outer_row+i,outer_col+j] = n^2 - M[outer_row+i,outer_col+j]
end
end
else
# Even order magic square
# Translated from https://pypi.python.org/pypi/magic_square/0.2
# and devectorized by suggestion of the Julia-users group
# by Phillip Berndt
m = n >> 1
k = m >> 1
b = m*m
b3 = 3b
magic!(M, m)
for J = 1:m
Jm = J+m
@simd for I = 1:m
@inbounds M[I+m, J] = M[I, J]
@inbounds M[I, Jm] = M[I, J]
@inbounds M[I+m, Jm] = M[I, J]
end
end
for J = 1:n
@simd for I = 1:n
if I <= k
if J <= m
@inbounds M[I, J] += b3
end
elseif I <= m
if 1+m <= J
@inbounds M[I, J] += b3
end
elseif I <= n-k + 1
if J <= m
@inbounds M[I, J] += b + b
else
@inbounds M[I, J] += b
end
else
if J <= m
@inbounds M[I, J] += b
else
@inbounds M[I, J] += b + b
end
end
end
end
M[1+k, 1+k] += b3
M[1, 1+k] -= b3
M[1, 1+m+k] += b3
M[1+k, 1+m+k] -= b3
end
end
# Speed test:
@time begin
for i = 3:1:1000
M = Array(Int, i, i)
magic!(M, i);
end
end
// Direct translation of the current version.
// Not nice code, sorry ;) Only for testing..
#include <time.h>
#include <stdio.h>
struct mat {
int n;
int *mem;
mat(int n) {
this->n = n;
this->mem = new int[n*n];
}
~mat() {
delete this->mem;
}
inline int& operator()(const int row, const int col) {
return this->mem[this->n * (col - 1) + row - 1];
}
};
void iain_magic(mat &M, int n) {
if(n % 2 == 1) {
int ndiv2 = n / 2;
for(int I = 1; I<=n; I++) {
int intermediate1 = (I-1+ndiv2 % n);
int intermediate2 = (2*I - 2) % n;
for(int J = 1; J<=n; J++) {
int x = (J + intermediate1 < n? J + intermediate1: J + intermediate1 - n);
int y = (J + intermediate2 < n? J + intermediate2: J + intermediate2 - n);
M(I, J) = n*x + y + 1;
// M(I, J) = n*((J + intermediate1)%n) + ((J + intermediate2)%n) + 1;
}
}
}
else if(n % 4 == 0) {
for(int I = 1; I <= n; I++) {
int In = (I-1)*n;
for(int J = 1; J<=n; J++) {
M(I,J) = In+J;
}
}
for(int outer_col = 1; outer_col <= n; outer_col += 4) {
for(int outer_row = 1; outer_row <= n; outer_row += 4) {
for(int i= 1; i<= 2; i++) {
for(int j=0; j<= 3; j++) {
M(outer_row+i,outer_col+j) = n*n - M(outer_row+i,outer_col+j);
}
}
for(int i= 0; i<= 3; i++) {
for(int j=1; j<= 2; j++) {
M(outer_row+i,outer_col+j) = n*n - M(outer_row+i,outer_col+j);
}
}
}
}
}
else {
int m = n >> 1;
int k = m >> 1;
int b = m*m;
int b3 = 3*b;
iain_magic(M, m);
for(int J = 1; J<=m; J++) {
int Jm = J+m;
for(int I = 1; I <=m ; I++) {
M(I+m, J) = M(I, J);
M(I, Jm) = M(I, J);
M(I+m, Jm) = M(I, J);
}
}
for(int J=1; J<=n; J++) {
for(int I=1; I<=n; I++) {
if(I <= k) {
if(J <= m)
M(I, J) += b3;
else if(I <= m)
if(1+m <= J) M(I, J) += b3;
} else if(I <= n-k + 1) {
if(J <= m)
M(I, J) += b + b;
else
M(I, J) += b;
} else {
if(J <= m)
M(I, J) += b;
else
M(I, J) += b + b;
}
}
}
M(1+k, 1+k) += b3;
M(1, 1+k) -= b3;
M(1, 1+m+k) += b3;
M(1+k, 1+m+k) -= b3;
}
}
int main(int argc, char *argv[]) {
clock_t start = clock();
for(int i=3; i<1000; i++) {
mat A(i);
iain_magic(A, i);
}
printf("Needed %f s\n", (clock() - start) * 1. / CLOCKS_PER_SEC);
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment