After some test in which very simple operation (subtraction of images) were done, the results were that considering `int32' data type was more time consumption than using `float32'.

For example:

astarithmetic im3.fits im4.fits float32 float32 -

finished in 0.111564. While

astarithmetic im3.fits im4.fits int32 int32 -

finished in 0.120998.

Note that images are different names but the size is the same.
That difference in time makes no sense and this overhead along big projects will cause a lot of extra time in processing data.