ERROR: No algorithm ‘BroadcastWorkspace’ version ‘1’, skipping directive

Description¶

BroadcastWorkspace distributes a workspace from one MPI process (the broadcaster) to all other processes in the MPI job. Only the broadcaster rank needs to provide an input workspace; all other ranks will receive an identical copy as their output. The algorithm is optimized for large workspaces through chunked broadcasting, which processes the data in manageable pieces to balance memory usage and performance. When spectra share identical X data, the algorithm broadcasts the X array only once and shares it across all spectra in the output workspace, reducing both transmission time and memory footprint. This is a collective MPI operation, meaning all processes in the communicator must call the algorithm simultaneously or the job will hang.

Usage¶

Example - Simple Broadcast

The following code should be saved as broadcast.py

from mantid.simpleapi import CreateWorkspace, BroadcastWorkspace
from mpi4py import MPI

comm = MPI.COMM_WORLD
rank = comm.Get_rank()
size = comm.Get_size()

# Only rank 0 creates the workspace to broadcast
input_ws = None
if rank == 0:
    print(f"Rank {rank}: Creating workspace to broadcast")
    x = [0.0, 1.0, 2.0, 3.0, 4.0]
    y = [1.0, 2.0, 3.0, 4.0,
        5.0, 6.0, 7.0, 8.0,
        9.0, 10.0, 11.0, 12.0]
    e = [0.1, 0.1, 0.1, 0.1,
        0.2, 0.2, 0.2, 0.2,
        0.3, 0.3, 0.3, 0.3]

    input_ws = CreateWorkspace(
        DataX=x,
        DataY=y,
        DataE=e,
        NSpec=3,
        OutputWorkspace="input_ws"
    )

    print(f"Rank {rank}: Created workspace with {input_ws.getNumberHistograms()} spectra")

# All ranks participate in the broadcast
# Only rank 0 needs to provide InputWorkspace
if rank == 0:
    output_ws = BroadcastWorkspace(
        InputWorkspace=input_ws,
        BroadcasterRank=0,
        OutputWorkspace="output_ws"
    )
else:
    # Other ranks don't provide InputWorkspace
    output_ws = BroadcastWorkspace(
        BroadcasterRank=0,
        OutputWorkspace="output_ws"
    )

# All ranks now have the workspace
print(f"Rank {rank}: Received workspace with {output_ws.getNumberHistograms()} spectra")
if rank != 0:
    print(f"Rank {rank}: First spectrum Y values: {output_ws.readY(0)}")
    print(f"Rank {rank}: Last spectrum Y values: {output_ws.readY(2)}")

To run:

mpiexec -n 4 broadcast.py

Output:

Note

Due to nature of MPI, the output may be recieved out of order

Rank 0: Creating workspace to broadcast
Rank 0: Created workspace with 3 spectra
Rank 0: Received workspace with 3 spectra
Rank 1: Received workspace with 3 spectra
Rank 1: First spectrum Y values: [1. 2. 3. 4.]
Rank 1: Last spectrum Y values: [ 9. 10. 11. 12.]
Rank 2: Received workspace with 3 spectra
Rank 2: First spectrum Y values: [1. 2. 3. 4.]
Rank 2: Last spectrum Y values: [ 9. 10. 11. 12.]
Rank 3: Received workspace with 3 spectra
Rank 3: First spectrum Y values: [1. 2. 3. 4.]
Rank 3: Last spectrum Y values: [ 9. 10. 11. 12.]

ERROR: No algorithm ‘BroadcastWorkspace’ version ‘1’, skipping directive