Chris@42: Next: An improved replacement for MPI_Alltoall, Previous: Basic distributed-transpose interface, Up: FFTW MPI Transposes [Contents][Index]
Chris@42:The above routines are for a transpose of a matrix of numbers (of type
Chris@42: double), using FFTW’s default block sizes. More generally, one
Chris@42: can perform transposes of tuples of numbers, with
Chris@42: user-specified block sizes for the input and output:
Chris@42:
fftw_plan fftw_mpi_plan_many_transpose Chris@42: (ptrdiff_t n0, ptrdiff_t n1, ptrdiff_t howmany, Chris@42: ptrdiff_t block0, ptrdiff_t block1, Chris@42: double *in, double *out, MPI_Comm comm, unsigned flags); Chris@42:
In this case, one is transposing an n0 by n1 matrix of
Chris@42: howmany-tuples (e.g. howmany = 2 for complex numbers).
Chris@42: The input is distributed along the n0 dimension with block size
Chris@42: block0, and the n1 by n0 output is distributed
Chris@42: along the n1 dimension with block size block1. If
Chris@42: FFTW_MPI_DEFAULT_BLOCK (0) is passed for a block size then FFTW
Chris@42: uses its default block size. To get the local size of the data on
Chris@42: each process, you should then call fftw_mpi_local_size_many_transposed.
Chris@42:
Chris@42:
Chris@42: