Because the machine has 4 cores, the first obvious attempt was to make 4 tasks and launch them. Due to work load imbalance though (specifically, tasks near the middle of the image were larger), it is actually more advantageous to create far more than 4 tasks to be dynamically assigned to the 4 cores, or statically assign tasks with more balanced loads per core, such as outermost to innermost chunks, or divide the image differently such as making the tasks near the middle of the image cover fewer rows.
Because the machine has 4 cores, the first obvious attempt was to make 4 tasks and launch them. Due to work load imbalance though (specifically, tasks near the middle of the image were larger), it is actually more advantageous to create far more than 4 tasks to be dynamically assigned to the 4 cores, or statically assign tasks with more balanced loads per core, such as outermost to innermost chunks, or divide the image differently such as making the tasks near the middle of the image cover fewer rows.
This comment was marked helpful 0 times.