49
edits
Changes
→What was done
=== What was done ===
At first the power function used was switched out with __pow in the kernel as the traditional pow function is more heavy of a function.But the results were very small causing a different from around 10-30 milliseconds. Afterwards the kernel was upgraded to implement grid and strides. In doing so instead of doing all the calculations on one thread, many of them were able to be calculated on a separate row. This made it so that for every pixel with a row one thread would be responsible for the colour shift. Other implementations were made to transfer all the data necessary for calculations all at once from the beginning and then perform all calculations done by tatsy on the device side, but due to time constraints and the complication of the project we were unable to fully implement these changes. The code is however left (commented) in the included project download.
=== Optimized Kernel ===