Hi,
I'm currently developing a kernel where I'm computing the average of an array this way:
float16 values;
float16 validCount;
float16 average = values / validCount;
My concern is when some elements of validCount are at 0, which is a valid usecase for me. I did some testing and the C7X doesn't seem to have a problem dividing by zero, it just results in a +inf value and __get_FSR(float16)::DIV0 gets set.
My questions are:
1- Is there a performance drop for dividing by zero?
2- Does __get_FSR(float16)::DIV0 ever get reset? I can't seem to get the bit to go back to zero.
3- Is it harmful that __get_FSR(float16)::DIV0 is always set?
Thanks,
Fred
