I've heard that there's a performance penalty with using both floats and doubles. Obviously mixing the two in a calculation would require a conversion. However, is there a penalty if you do some float calculations, then do some double calculations, without the floats ever touching the doubles?
>>1 However, is there a penalty if you do some float calculations, then do some double calculations, without the floats ever touching the doubles?
It's possible that breaking things up could limit opportunities for vectorization, but it's hard to say for sure without seeing the actual code and knowing what your target is.