Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: FFTW 3.3.8: Advanced distributed-transpose interface Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82:
Chris@82:

Chris@82: Next: , Previous: , Up: FFTW MPI Transposes   [Contents][Index]

Chris@82:
Chris@82:
Chris@82: Chris@82:

6.7.2 Advanced distributed-transpose interface

Chris@82: Chris@82:

The above routines are for a transpose of a matrix of numbers (of type Chris@82: double), using FFTW’s default block sizes. More generally, one Chris@82: can perform transposes of tuples of numbers, with Chris@82: user-specified block sizes for the input and output: Chris@82:

Chris@82:
Chris@82: 
fftw_plan fftw_mpi_plan_many_transpose
Chris@82:                 (ptrdiff_t n0, ptrdiff_t n1, ptrdiff_t howmany,
Chris@82:                  ptrdiff_t block0, ptrdiff_t block1,
Chris@82:                  double *in, double *out, MPI_Comm comm, unsigned flags);
Chris@82:
Chris@82: Chris@82: Chris@82:

In this case, one is transposing an n0 by n1 matrix of Chris@82: howmany-tuples (e.g. howmany = 2 for complex numbers). Chris@82: The input is distributed along the n0 dimension with block size Chris@82: block0, and the n1 by n0 output is distributed Chris@82: along the n1 dimension with block size block1. If Chris@82: FFTW_MPI_DEFAULT_BLOCK (0) is passed for a block size then FFTW Chris@82: uses its default block size. To get the local size of the data on Chris@82: each process, you should then call fftw_mpi_local_size_many_transposed. Chris@82: Chris@82: Chris@82:

Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: