SCALFFT_PRE_A2A_ALGO
Set the transpose algorithm for the pre-stage of 3D FFT computations.
Parameters
Parameter |
Value |
Description |
|---|---|---|
A2A_ALGO_AUTO_TUNING |
0 |
The library automatically selects the optimal transpose algorithm based on the number of processes and the size of the input matrix. |
A2A_ALGO_COLL_TASK_ALLTOALLV |
1 |
Task-based transpose algorithm using the MPI A2A interface. |
A2A_ALGO_P2P_TASK_LINEAR |
2 |
Task-based transpose algorithm using the linear A2A interface. |
A2A_ALGO_P2P_TASK_PAIRWISE |
3 |
Task-based transpose algorithm using the pairwise A2A interface. |
A2A_ALGO_P2P_PIPELINE_LINEAR |
4 |
Pipelined transpose algorithm. |
A2A_ALGO_P2P_LAMMPS |
5 |
LAMMPS native collective communication transpose algorithm. |
A2A_ALGO_P2P_PAIRWISE_WO_PACK |
6 |
Pack-free transpose algorithm using pairwise A2A. Currently, only slab decomposition is supported. |
A2A_ALGO_P2P_LINEAR_WO_PACK |
7 |
Pack-free transpose algorithm using linear A2A. Currently, only slab decomposition is supported. |
A2A_ALGO_P2P_PIPELINE_INTRA_NODE |
8 |
Single-node pipelined transpose algorithm utilizing only pack and unpack routines. |
Examples
The transpose algorithm for the 3D FFT pre-stage can be specified as the linear algorithm by setting the environment variable -x SCALFFT_PRE_A2A_ALGO=2.