PTX for Complex Numbers

Accelerated Computing CUDA CUDA Programming and Performance
1

I need to move complex numbers between registers, shared mem and global mem. It looks like that cuComplex cuComplex is a vector type like float2, but inline asm only accept scalar types. Should I try to convert complex float into double or something? How should I do that?

2

use 2 registers, one for the real part and one for the imaginary part.

There isn’t any support for complex arithmetic at the PTX level, so you will need to code all that yourself using basic supported ops.

1 Like
3

Thank you for your answer!