If stream keyword argument has type dpctl.SyclQueue, then submit a barrier into self.sycl_queue, and than submit another barrier into stream with dependency on the event returned by submission of the barrier into self.sycl_queue:

   ev = self.sycl_queue.submit_barrier()
   stream.submit_barrier(dependent_events=[ev])

All other types of stream are ignored, in particular integer values of stream.

• Have you provided a meaningful PR description?
• Have you added a test, reproducer or referred to an issue with a reproducer?
• Have you tested your changes locally for CPU and GPU devices?
• Have you made sure that new changes do not introduce compiler warnings?
• If this PR is a work in progress, are you filing the PR as a draft?

--

With this PR, MPI transfer of usm_ndarray allocated using level-zero backend becomes possible:

from mpi4py import MPI
import dpctl.tensor as dpt
import numpy as np

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

# automatic MPI datatype discovery
if rank == 0:
    data = dpt.arange(100, dtype="float64")
    comm.Send(data, dest=1, tag=13)
elif rank == 1:
    data = dpt.empty(100, dtype="float64")
    comm.Recv(data, source=0, tag=13)
    X = dpt.asnumpy(data)
    print("From rank=1, received: ", X)

The example run using mpi4py built using Intel(R) MPI from oneAPI basekit:

$ I_MPI_OFFLOAD=2 mpirun -n 2 python mpi_exchange.py
From rank=1, received:  [ 0.  1.  2.  3.  4.  5.  6.  7.  8.  9. 10. 11. 12. 13. 14. 15. 16. 17.
 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35.
 36. 37. 38. 39. 40. 41. 42. 43. 44. 45. 46. 47. 48. 49. 50. 51. 52. 53.
 54. 55. 56. 57. 58. 59. 60. 61. 62. 63. 64. 65. 66. 67. 68. 69. 70. 71.
 72. 73. 74. 75. 76. 77. 78. 79. 80. 81. 82. 83. 84. 85. 86. 87. 88. 89.
 90. 91. 92. 93. 94. 95. 96. 97. 98. 99.]