在由Octave / Matlab的pararrayfun并行化的函数内部操作的存储和访问矩阵

Question

Pararrayfun is used this way

pkg load parallel;
# fun is the function to apply 
fun = @(x) x^2;
vector_x = 1:10000;
vector_y = pararrayfun(16, fun, vector_x);

I intend to apply it to a function with this form

fun = @(i) zz(1:100,i+1) = z.^2;

where z is a column vector, and zz is a 100xi matrix. I want to get the matrix zz out after the pararrayfun() but I get nothing back. I dont know if I am using the right code or the right argument as I have just started using octave a month ago. Any help would really be appreciated.

Here is my attempted "full" code:

pkg load parallel;
z = linspace(-1,10,100).';
fun = @(i) zz(1:100,i+1) = z.^2;
i = 0:9999;
pararrayfun(16, fun, i);

I tried to use global zz but I still get nothing.

Answer 1

The first answer is that the zz variable inside you fun is a local variable. It would have to be declared as global , but this is not possible inside an anonymous function (defined with @() ... ).

One might be tempted to try to define fun with

pkg load parallel;
z = linspace(-1,10,100).';
global zz
zz = zeros(100, 10000);
function fun1(idx, z)
  global zz
  zz(:, idx) = z.^2;
endfunction

fun = @(idx) fun1(idx, z)

idx = 1:10000;
pararrayfun(nproc, fun, idx);

But the zz is not modified. Probably because each thread has its own context.

This is a good thing, because otherwise modifying a global variable inside a pararrayfun call would lead to potential race conditions.

Here is a safer way:

pkg load parallel;
z = linspace(-1,10,100).';
fun = @(i) z.^2;
idx = 1:10000;
zz = pararrayfun(nproc, fun, idx);

As a side note, for this example,

zz = repmat(z, 1, 10000);

would be much faster.