cutil_math.h bugs? operator / bug in cutil_math.h

Accelerated Computing CUDA CUDA Programming and Performance
1

The following code can be found in cutil_math.h in the SDK:

inline __host__ __device__ float2 operator/(float s, float2 a)

{

    float inv = 1.0f / s;

    return a * inv;

}

which means that

float2 foo = 1.0f / make_float2(2.0f, 3.0f);

will result in (2.0f, 3.0f), whereas I would expect (1.0f / 2.0f, 1.0f / 3.0f).

I think that the code in cutil_math.h should be replaced by:

inline __host__ __device__ float2 operator/(float s, float2 a)

{

    return make_float2(s / a.x, s / a.y);

}

The same is true for float3 and float4 divisions.

Any thoughts?

2

this indeed seems to be suspicious. seems like arguments are in wrong places. had it been (float2 a, float s) the definition would be then more intuitive :-)

this is why in Java language there are no operators overloading ;-) only explicit method calls, this is extra work to type it’s name or to go and see the method definition, but there is no possibility to misinterpret operator meaning then!