Bump ACLSHMEM to v1.5.0.rc1

Pick up the aclshmem_finalize spelling fix and update Ascend tests/tutorials to use the corrected APIs.

Author: janssen-v

3rdparty/shmem

@@ -50,7 +50,7 @@ def run_test_distributed(rank, world_size):
     y_ref = torch_mype(x0, rank, 3)
     _shmemx_get_my_pe[3, 1, 1](x0)
     assert torch.equal(x0.cpu(), y_ref.cpu())
-    _ = ash.aclshmem_finialize()
+    _ = ash.aclshmem_finalize()
 
 
 @pytest.mark.dist

python/triton_dist/test/ascend/test_my_pe.py

@@ -50,7 +50,7 @@ def run_test_distributed(rank, world_size):
     y_ref = torch_num_ranks(x0, world_size, 3)
     _shmemx_get_num_ranks[3, 1, 1](x0)
     assert torch.equal(x0.cpu(), y_ref.cpu())
-    _ = ash.aclshmem_finialize()
+    _ = ash.aclshmem_finalize()
 
 
 @pytest.mark.dist

python/triton_dist/test/ascend/test_num_ranks.py

@@ -72,7 +72,7 @@ def run_test_distributed(rank, world_size):
 
     # Cleanup
     ash.aclshmem_free_tensor(remote_mem)
-    _ = ash.aclshmem_finialize()
+    _ = ash.aclshmem_finalize()
 
 
 @pytest.mark.dist

python/triton_dist/test/ascend/test_symm_at.py

@@ -92,7 +92,7 @@ def run_test_distributed(rank, world_size):
     # Cleanup
     ash.aclshmem_free_tensor(data_mem)
     ash.aclshmem_free_tensor(signal_mem)
-    _ = ash.aclshmem_finialize()
+    _ = ash.aclshmem_finalize()
 
 
 @pytest.mark.dist

python/triton_dist/test/ascend/test_wait_notify.py

@@ -333,7 +333,7 @@ def run_test_distributed():
             raise AssertionError(error_msg)
 
     ash.aclshmem_free_tensor(peer_mem)
-    _ = ash.aclshmem_finialize()
+    _ = ash.aclshmem_finalize()
 
 
 if __name__ == "__main__":

tutorials/ascend/01-ascend-allgather-gemm.py

@@ -347,7 +347,7 @@ def run_test_distributed():
             raise AssertionError(error_msg)
 
     ash.aclshmem_free_tensor(peer_mem)
-    _ = ash.aclshmem_finialize()
+    _ = ash.aclshmem_finalize()
 
 
 if __name__ == "__main__":

tutorials/ascend/02-ascend-gemm-reduce-scatter.py

@@ -360,7 +360,7 @@ def run_test_distributed():
             raise AssertionError(error_msg)
 
     ash.aclshmem_free_tensor(peer_mem)
-    _ = ash.aclshmem_finialize()
+    _ = ash.aclshmem_finalize()
 
 
 if __name__ == "__main__":