In vanilla xGPU this is not necessary. I believe it is needed in the dp4a branch though, and there it's done as a pre-processing step in GPU memory.
On 16 June 2017 at 12:48, Gerald Harp <gh...@seti.org> wrote: > Hi David, or whomever > > We have found that to make best use of vector operations in the Xeon Phi > machines that we have to de-interleave the complex data. That is a single > long array of length 2N > > [re1, im1, re2, im2, ....,reN, imN] > > gets de-interleaved into two arrays > > [re1, re2, ....,reN] > > [im1, im2,..., imN] > > before the data are loaded into the vector registers. Is this done in > XGPU, or is it not necessary? > > If de-interleaving is done, then where is it done? In the GPU? Or in > memory prior to GPU? > > Thanks > > Gerry > > -- > Gerald (Gerry) R. Harp, PhD > Director, SETI Research / Senior Scientist > SETI Institute > 189 Bernardo, Ste. 200 > Mountain View, CA 94043 > 650-960-4576 > > -- > You received this message because you are subscribed to the Google Groups " > casper@lists.berkeley.edu" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to casper+unsubscr...@lists.berkeley.edu. > To post to this group, send email to casper@lists.berkeley.edu. > -- You received this message because you are subscribed to the Google Groups "casper@lists.berkeley.edu" group. To unsubscribe from this group and stop receiving emails from it, send an email to casper+unsubscr...@lists.berkeley.edu. To post to this group, send email to casper@lists.berkeley.edu.