Question

I’m trying to get rid of floats in my functions.

I have one function that operates on a 16 bit integer value which is an upscaled 8 bit value. Then a downscaled 8 bit is sent to the output.

I’m sure I’m not explaining it well. Something like this:

int8 spot_value = 21;  //arbitrary. just need a starting point
int16 running_value;

running_value = spot_value << 8;  //multiply by 256 which is 5376
running_value += 154;  //my upscaled value is now 5530
spot_value = running_value >> 8;  //downscale again

if we were working with floats my downscaled value would be 21.6 which I could round easily to 22 and convert to an int8. But as-is it will truncate to 21 which I don’t want.

Is there a way to “round” the integer up if it would have been appropriate to do so had it been a float but without converting anything to a float (even temporarily)?

It is probably a lot simpler than I’m making it out to be.

Answer 1

You could add 1/2 to the value. In your scaled system that’s 128.

int8 spot_value = 21;  //arbitrary. just need a starting point
int16 running_value;

running_value = spot_value << 8;  //multiply by 256 which is 5376
running_value += 154;  //my upscaled value is now 5530
spot_value = running_value + 128;  //add an additional 1/2 for rounding
spot_value = spot_value >> 8;  //downscale again

or…

scale_shift = 8;
scale = 1 << scale_shift
round = scale >> 1;

int8 spot_value = 21;  //arbitrary. just need a starting point
int16 running_value;

running_value = spot_value << scale_shift;  //multiply by 256 which is 5376
running_value += 154;  //my upscaled value is now 5530
spot_value = (running_value + round) >> scale_shift;  //downscale again

or a bit trick based on Dav’s answer…

running_value = spot_value << scale_shift;  //multiply by 256 which is 5376
running_value += 154;  //my upscaled value is now 5530
spot_value = (running_value >> 8) + ((running_value >> 7) & 1)

.