Quadruple precision

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

Hi all,

I am having some issues with an algorithm that requires more than double precision for some computations.

The problem is that the math includes cos(z) and sin(z), where z is a complex number with a small real and a large imaginary part. This results in some very large numbers (Numbers of order ±10e40).

Is it possible to get higher than double precision in CUDA Fortran for complex number math?

Thanks, Jan

Edit: I should mention that I tried various ways outside CUDA to get the algorithm to produce correct results. The only successful way was gfortran with REAL(KIND=16).

2

Is it possible to get higher than double precision in CUDA Fortran for complex number math?

Sorry, no.

  • Mat