jakubowiczish / fortran-task1

My solution for Fortran assignment - task 1

My solution for Fortran assignment - task 1

Requirements

• ifort - Intel Fortran Compiler installed

Run project

In order to run this project, type following commands in the main project directory:

cmake CMakeLists.txt
make all
./mull

Conclusions

You can see below plots that are presenting the results of matrix multiplications - the time it took for every matrix depending on its size (N)

Kind = 4

• The advantage of matmul method - that is built in fortran - can be observed especially when kind = 4 and kind = 8. It's significantly faster, probably because of its optimalization and some tricks that are used in its implementation.

• The second fastest method is bettmull - it does much less operations than naivmull method and the difference is apparent.

• Dotmull method is slightly faster than naivmull, but the disparity is hardly noticeable for larger matrices.

Kind = 8

This case is quite similar to the previous one.

• Matmul method is still the fastest method, betmull somewhat slower.
• Results for naivmull and dotmull methods are almost the same.

Kind = 16

Major diffences can be noticed only when kind is equal to 16 and matrices are really large, although this plot does not present it in the most suitable way - I guess logarithmic scale would be better.

The operation of matrix multiplication itself can take significant of time.

There's a huge gap between all methods when the size of matrix is 1280 x 1280.

• It took almost 130 seconds for naive method to actually multiply matrices
• Dotmull method was the worst one - operation lasted over 200 seconds!
• Matmul method was best as usual - 78 seconds in comparison to over 200 is quite magnificent.
• Betmull method took second place, also as usual - 87 seconds is still pretty nice time.

I honestly believe that everything was done properly, although bear in mind the fact that the results may differ depending on what computer you run this project etc.