Firstly, I write image blurring code with non-shared memory for all pictures.
nvcc test.cu output.exe 1.ppm 1_out.ppm
For rgb I added for block “ for (int color = 0; color < channels; color++)”. Then I run image blurring code with shared memory for all pictures.. When we use shared memory, we see a significant reduction in code execution time. This is because we use the same data over and over, and if we recall it in shared memory, it will be faster to retrieve data than global memory.
Also I used __syncthreads() twice in this code.The firstly __syncthreads() is enures that threads have finished loading input data into N_ds[] before any of them can move foward. Second __syncthreads() is ensure that all threads finished N_ds []elements in shared memory before any of them, move to the next iteration and load the elements in the next tiles. Otherwise threads would load elemnts too early and corrupt the input data for other threads.
Lastly I used shared memoy and constant memory for all pictures.
The fastest running code was the code I used in the constant memory and the shared memory. Because constant memory is cached to constant cache and it can be accessed by each thread, so all threads needs to acces mask elements in the same order.
Then , I added constant variable for mask.
picture non-shared memory execution time(ms) shared memory execution time(ms) shared memory and constant memory execution time(ms) 1.ppm 1.663328 0.907202 0.78048 2.ppm 6.501664 3.744288 3.244064 3.ppm 14.506112 8.383712 7.408128 4.ppm 25.38336 15.124896 13.163232 5.ppm 40.081791 23.758017 20.58544 6.ppm 56.329407 33.941921 29.522272 7.ppm 75.391396 45.061089 40.249729 8.ppm 98.765121 59.157726 52.655006 9.ppm 125.083809 74.433823 67.103455 10.ppm 153.996735 94.170845 82.63591
The larger the image size, the more noticeable the time difference. Because more and more threads are being used. So the thread that uses global memory or shared memory or constant memory is growing.