efficient parallel algorithm modifying large arrays of integers?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi all

I’m trying to come up with an efficient algorithm to do the parallel equivalent of the following serial algorithm (in pseudocode/C++):

[codebox]valFrom = Change integers with this value

valTo = Change integers that have “valFrom” value to this value

numToChange = number of integers with value “valFrom” to change

arraylen = length of the array

copy array of integers from device to host array named “array”

int count = 0;

for (int i=0; i<arraylen && count < numToChange; ++i)

{

if (array[i] == valFrom)

{

    array[i] = valTo

    count++;

}

}

copy “array” host array back to device[/codebox]

Does anyone have any recommendations for turning this into a data parallel algorithm to run on the device? I was thinking something like a modified parallel reduction or perhaps sorting the integers into a new array and then assigning “numToChange” threads to the section of the array containing the relevant ints. But was just wondering if anyone has any better ideas?

Thanks!

nodlams