[Mlir-commits] [mlir] [mlir][mesh] Handling changed halo region sizes during spmdization (PR #114238)

Frank Schlimbach llvmlistbot at llvm.org
Mon Nov 4 02:06:43 PST 2024


================
@@ -240,7 +243,7 @@ def Mesh_ShardingOp : Mesh_Op<"sharding", [
     // The tensor is sharded on its second dimension along axis 0 of @mesh1d_4
     // and it has pre-defined shard sizes. The shards of the devices will have
     // the following shapes: [4x2, 4x3, 4x4, 4x5]
-    %sharding4 = mesh.sharding @mesh1d_4 split_axes = [[] split_axes = [0]] sharded_dims_sizes = [2, 3, 4, 5]
+    %sharding4 = mesh.sharding @mesh1d_4 split_axes = [[] split_axes = [0]] sharded_dims_offsets = [2, 5, 9, 14]
----------------
fschlimb wrote:

Probably a cut&paste error. I wonder why the tests pass.

https://github.com/llvm/llvm-project/pull/114238


More information about the Mlir-commits mailing list