Hi,
thanks for the feedback, I agree with most of what you said. Just a few
comments:
> I think the main thing is that you really don't want branches inside the
> inner loop. You can handle the cases where a component is unused
> by building a mask before entering the loop and just or'ing with that
> mask inside the loop.
Good point. Not sure whether that eliminates the need for both the
'if(DestFormat.bits[x])' AND 'if(SrcFormat.bits[x])', but I can merge
them into one 'if' at least.
> As for the shifts,
> in the large majority of cases you will only need two shifts. The only
> time you need more than that is when the destination component is more
> than two times the size of the source component, so I don't think you
> want to bother with that in the common function.
>
I don't think there's much overhead by doing the for loop since even in
the cases that need to shifts there's at least one if(SrcFormat.bits <
DestFormat.bits) necessary. The code inside the for loop will just add
one assignment and an additional if. Perhaps lateron this could really
be split into a seperate function, but I don't think it's worth to add
another hundred lines of code for this small optimization :/
Apart from that, I'll be changing my code accordingly to your suggestions.
Best regards,
Tony
