1

I didn't expect a result of 1's:

ILArray<int> test = ILMath.ones<int>(20, 20);
test = test / 2;

I thought the result would be 0's. If ILNumerics uses the Intel IPP library then I believe 1's is the result the IPP library would return for integer division. If this is by design then I guess my best option would be to use something like this to get integer division results:

test = toint32(tosingle(test) / 2.0f);

Any comments? Questions? RTFM pointers? Eric

Haymo Kutschbach
  • 3,322
  • 1
  • 17
  • 25
user3316386
  • 29
  • 1

1 Answers1

0

ILNumerics sticks to the same integer division convention as Matlab (and apparently IPP?) does: it does not truncate the fractional part of the result but rounds it (away from zero at midpoint) and saturates it at the datatype boundaries.

Personally, I am not very happy with that - it grew out of early compatibility aspirations. While the saturation is good and conform to the intuition, the rounding obviously conflicts with the way .NET (and all major general purpose languages) handle integer division by simply truncating the fractional part. You can vote for a change in behavior here: http://ilnumerics.net/mantis/view.php?id=188

In order to get the 'C/.NET-Convention' results, you can use the ILMath.apply function as a replacement: 

ILArray<int> a = 1, b = 2; 
ILArray<int> c = a / b; //gives 1
c = ILMath.apply((_a,_b) => _a / _b, a, b);  // gives 0
Haymo Kutschbach
  • 3,322
  • 1
  • 17
  • 25