Complex number exponential function

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello everyone,

I am new to CUDA programming.

I need to write a kernel function which has exponential operation on complex numbers. What is the right way to do this?

If A is float number, in kernel, I can use exp(A) or expf(A). But how to do on complex number?

If I do :
cuComplex a;
B = expf(a);

I will have error:
error: no suitable conversion function from “cuComplex” to “float” exists

Can anyone give me any advice? Thanks

2

cuComplex.h only provides complex types and basic complex arithmetic, intended to serve the needs of CUBLAS and other libraries. The complex exponential function can be computed component-wise in straightforward manner:

exp(z) = exp(x + iy) = exp(x) * exp (iy) = exp(x) * (cos(y) + i sin(y))

This then leads to code like this (untested):

__device__ __forceinline__ cuComplex my_cexpf (cuComplex z)

{

    cuComplex res;

    float t = expf (z.x);

    sincosf (z.y, &res.y, &res.x);

    res.x *= t;

    res.y *= t;

    return res;

}
3

Thank you.