Expected speed gains

[mrquincle]

Hi,

Although this is older code, perhaps you've some valuable feedback anyway. If I run your code the CPU version is faster than the GPU version. Is that to be expected for an array of a million values?

-- Analysis --

Sorting algorithm: CUDA Merge Sort
Array type: Random
Array size: 1000000
Time elapsed: 4.821247 s


-- Analysis --

Sorting algorithm: CUDA Quick Sort
Array type: Random
Array size: 1000000
Time elapsed: 0.708224 s


-- Analysis --

Sorting algorithm: Merge
Array type: Random
Array size: 1000000
Number of comparisons: 19951424
Number of swaps: 39902848
Time elapsed: 0.263655 s


-- Analysis --

Sorting algorithm: Quick
Array type: Random
Array size: 1000000
Number of comparisons: 16357222
Number of swaps: 4991962
Time elapsed: 0.205382 s

I've created a python notebook to show: https://colab.research.google.com/drive/1ehWJ30GWAcw6Vf3vREFxqlnMd3UdXNYh (should be visible without Google account, if you run it yourself you need a Google account though).

[icaroharry]

Hi @mrquincle. I couldn't find the results that we got when we did this experiments (it was a long time ago, but I should have kept the data 😞).

I'd recommend you to test with larger arrays. I remember we concluded that it doesn't worth to run in the GPU unless you have a considerably large array. But we did find some improvements when running with a very large array (don't remember the actual length, though).