What happened?
when using advanced indexing on a distributed array, where the list for indexing only contains one element, it will return an empty array, even if it should in theory work. When adding another element to the index array it returns the expected two values.
When not distributed it return the expected behaviour.
Code snippet triggering the error
import heat as ht
arr = ht.arange(25, split=0)
rank = arr.comm.rank
print(rank, arr[4])
if rank == 0: # ignore rank 1 as those indices are only on rank 0
print(rank, arr[4].larray)
print(rank, arr[[4]].larray)
print(rank, arr[(4,)].larray)
print(rank, arr[[3, 4]].larray)
arr.comm.Barrier()Error message or erroneous outcome
0 DNDarray(4, dtype=ht.int32, device=cpu:0, split=None)
1
0 tensor(4, dtype=torch.int32)
0 tensor([], dtype=torch.int32)
0 tensor(4, dtype=torch.int32)
0 tensor([3, 4], dtype=torch.int32)
Version
features/1789
Python version
3.11
PyTorch version
2.5.1+cu124
MPI version
mpirun (Open MPI) 4.1.6
What happened?
when using advanced indexing on a distributed array, where the list for indexing only contains one element, it will return an empty array, even if it should in theory work. When adding another element to the index array it returns the expected two values.
When not distributed it return the expected behaviour.
Code snippet triggering the error
Error message or erroneous outcome
Version
features/1789
Python version
3.11
PyTorch version
2.5.1+cu124
MPI version
mpirun (Open MPI) 4.1.6