Checking for NaN's in script

[ashwin.sudhir]

How do we check if a float is a NaN, denormal etc. in UnigineScript?

A quick search of the docs didn't throw up anything...

 

-a.

[ulf.schroeter]

Don't think there are any functions for checking invalid float/double values. Use proper pre-condition checks to avoid such conditions (not sure if Unigine would throw runtime exceptions in case of invalid results)

[ashwin.sudhir]

IMO any floating point library should have these, it's much easier to put an assert( !isnan( f ) ) than sift through a changing codebase and make sure all cases are handled well.

[ulf.schroeter]

I would always prefer safe coding techniques because asserts on NaN / denormal are just indications of a deeper algorithmic problem. But of course this is purely a matter of taste

[ashwin.sudhir]

No, I completely agree, the point of the assert is that we *realize* that there's a deeper algorithmic problem and fix it. The assert is of course, never the solution :(

[wickedshell]

Does UnigineScript set any signaling NaN's or are they all quiet?

[frustum]

You can use the following function to detect "wrong" fpu numbers:

int isnan(int v) {
 return (v != v);
}

 

This is an example:

void test(float f) {
 printf("%s: %d\n",typeinfo(f),isnan(f));
}

test(1.0f);
test(1.0);
test(vec3_one);
test(dvec3_one);

test(1.0f / 0.0f);
test(0.0f / 0.0f);
test(1.0 / 0.0);
test(0.0 / 0.0);
test(vec3_one / 0.0f);
test(vec3_zero / 0.0f);
test(dvec3_one / 0.0);
test(dvec3_zero / 0.0);

The result under win32 is:

float: 1: 0
double: 1: 0
vec3: 1 1 1: 0
dvec3: 1 1 1: 0
float: 1.#INF: 1
float: -1.#IND: 1
double: 1.#INF: 1
double: -1.#IND: 1
vec3: 1.#INF 1.#INF 1.#INF: 1
vec3: -1.#IND -1.#IND -1.#IND: 1
dvec3: 1.#INF 1.#INF 1.#INF: 1
dvec3: -1.#IND -1.#IND -1.#IND: 1