[Mlir-commits] [mlir] [mlir][sparse] fuse concat and extract_slice op if possible. (PR #89825)

[Aart Bik]

================
@@ -209,6 +209,86 @@ static void concatSizesFromInputs(OpBuilder &builder,
 
 namespace {
 
+/// TODO: move it to tensor dialect instead.
+///
+/// Fold `tensor.concat` and `tensor.extract_slice`
+///
+/// %concat = tensor.concat dim(2) %t0, %t1
----------------
aartbik wrote:

+1 on that, just a CHECK test in  test/Dialect/SparseTensor/ with a simple before after pattern


https://github.com/llvm/llvm-project/pull/89825