BUG: Double runs as single?

Discussion created by Atmapuri on Sep 9, 2011
Latest reply on Sep 9, 2011 by dimkar


I have a kernel like this:

__kernel void  vzPowx_dc(__global const double *aSrc, const int aSrcIdx,
                              const double bRe, const double bIm,
                      __global double *Dst, const int DstIdx,
                                            const int Length)
        dobule Re, Im, XRe, XIm, ExpX;
        size_t i = get_global_id(0);

        XRe = aSrc[aSrcIdx + i];

              Re = log(fabs(XRe));
              Im = atan2(0, XRe);

               XRe = Re * bRe - Im * bIm;
               XIm = Im * bRe + Re * bIm;

               ExpX = exp(XRe);
               Im = sincos(XIm, &Re);

               Re *= ExpX;
               Im *= ExpX;

              Dst[2*(DstIdx + i)] = Re;

              Dst[2*(DstIdx + i)+1] = Im;

The code running via AMD Open CL using CPU as Target runs with only 4 numerals of precision (single precision typical) instead of 8.