Address review comments by @mtsokol.

Author: hameerabbasi

sparse/mlir_backend/_ops.py

@@ -181,6 +181,9 @@ def add(x1: Array, x2: Array, /) -> Array:
 
 
 def asformat(x: Array, /, format: StorageFormat) -> Array:
+    if x.format == format:
+        return x
+
     out_tensor_type = format._get_mlir_type(shape=x.shape)
     ret_storage = format._get_ctypes_type(owns_memory=True)()
 

sparse/mlir_backend/levels.py

@@ -231,7 +231,8 @@ def _determine_format(*formats: StorageFormat, dtype: DType, union: bool, out_nd
 
     1. Counts the sparse levels for `union=True`, and dense ones for `union=False`.
     2. Gets the max number of counted levels for each format.
-    3. Constructs a format with the same number of counted levels.
+    3. Constructs a format with rank of `out_ndim` (max rank of inputs is taken if it's `None`).
+       If `union=False` counted levels is the number of sparse levels, otherwise dense.
        Sparse levels are replaced with `LevelFormat.Compressed`.
 
     Returns
@@ -243,7 +244,7 @@ def _determine_format(*formats: StorageFormat, dtype: DType, union: bool, out_nd
         if out_ndim is None:
             out_ndim = 0
         return get_storage_format(
-            levels=(Level(LevelFormat.Dense),) * out_ndim,
+            levels=(Level(LevelFormat.Dense if union else LevelFormat.Compressed),) * out_ndim,
             order="C",
             pos_width=64,
             crd_width=64,

sparse/mlir_backend/tests/test_simple.py

@@ -416,9 +416,12 @@ def test_reshape_dense(dtype):
         np.testing.assert_equal(actual_np, expected)
 
 
-@pytest.mark.skip(reason="Segfault")
-@pytest.mark.parametrize("src_fmt", ["csr", "csc"])
-@pytest.mark.parametrize("dst_fmt", ["csr", "csc"])
+@pytest.mark.parametrize(
+    "src_fmt", ["csr", "csc", pytest.param("coo", marks=pytest.mark.skip(reason="TODO: Report MLIR issue"))]
+)
+@pytest.mark.parametrize(
+    "dst_fmt", ["csr", "csc", pytest.param("coo", marks=pytest.mark.skip(reason="TODO: Report MLIR issue"))]
+)
 def test_asformat(rng, src_fmt, dst_fmt):
     SHAPE = (100, 50)
     DENSITY = 0.5