mod(x,y) with x < 0 computed incorrectly in algorithms

[casella]

Please run this simple test case (also attached)

model TestMod
  Integer j, j_mod;
  Integer k, k_mod;
algorithm
  when initial() then
    j := -5;
    j_mod := mod(j,3);
  end when;
equation
    k = -5;
    k_mod = mod(k,3);
end TestMod;

k_mod is correcly computed in the equation section as 1, while j_mod is incorrectly computed in the algorithm section as -2. This is a bit weird, as the same Modelica function is used in both cases.

@lochel, could you have a quick look if this is an easy fix?

[sjoelund.se]

This is a problem with the generated code for integer mod(). It is missing a floor() call. It is slightly problematic that Ceval and Simplify do not simply call mod(), and instead have their own definition without this problem so I will fix the generated code and these modules.

[lochel]

The difference is when the mod function is evaluated. It works correctly if it is evaluated during compile time. The runtime implementation is wrong. I fixed it in the generated c code but now I have to check where this code is actually generated.

[lochel]

It seems Martin was faster...

[sjoelund.se]

​https://github.com/OpenModelica/OMCompiler/pull/2432 :)

[sjoelund.se]

The bootstrapped compiler depends on generating the wrong code for integer modulus, so it's not that easy to fix this.

[sjoelund.se]

So the specification says to use:

return x-floor(x/y)*y;

Which works fine for doubles. But for integers, this fails if the integer is so large that it does not fit in a double precision float.

The old code had:

return x%y;

Which returns negative values in some cases. I think the correct code would then be:

tmp = x%y;
return tmp < 0 ? tmp+y : tmp;

[lochel]

Replying to sjoelund.se:

Which returns negative values in some cases. I think the correct code would then be:

tmp = x%y;
return tmp < 0 ? tmp+y : tmp;

That should work just well.