Numba DimShuffle: special case for 0d input

This circumvents a bug when DimShuffle of a scalar shows up inside a Blockwise, as the outer indexing yields a float (as opposed to a numpy scalar) which has no `.shape` attribute.

Author: ricardoV94

pytensor/link/numba/dispatch/elemwise.py

@@ -466,6 +466,16 @@ def squeeze_to_0d(x):
 
         return squeeze_to_0d
 
+    elif op.input_ndim == 0:
+        # DimShuffle can only be an expand_dims or a no_op
+        # This branch uses asarray in case we get a scalar due to https://github.com/numba/numba/issues/10358
+        new_shape = shape_template
+        new_strides = strides_template
+
+        @numba_basic.numba_njit
+        def dimshuffle(x):
+            return as_strided(np.asarray(x), shape=new_shape, strides=new_strides)
+
     else:
 
         @numba_basic.numba_njit
@@ -490,7 +500,7 @@ def dimshuffle(x):
 
             return as_strided(x, shape=new_shape, strides=new_strides)
 
-    cache_version = 1
+    cache_version = 2
     return dimshuffle, cache_version
 
 

tests/link/numba/test_blockwise.py

@@ -5,6 +5,7 @@
 from pytensor.tensor import lvector, tensor, tensor3
 from pytensor.tensor.basic import Alloc, ARange, constant
 from pytensor.tensor.blockwise import Blockwise, BlockwiseWithCoreShape
+from pytensor.tensor.elemwise import DimShuffle
 from pytensor.tensor.nlinalg import SVD, Det
 from pytensor.tensor.slinalg import Cholesky, cholesky
 from tests.link.numba.test_basic import compare_numba_and_py, numba_mode
@@ -80,3 +81,12 @@ def test_blockwise_alloc():
     assert out.type.ndim == 3
 
     compare_numba_and_py([val], [out], [np.arange(5)], eval_obj_mode=False)
+
+
+def test_blockwise_scalar_dimshuffle():
+    x = lvector("x")
+    blockwise_scalar_ds = Blockwise(
+        DimShuffle(input_ndim=0, new_order=["x", "x"]), signature="()->(1,1)"
+    )
+    out = blockwise_scalar_ds(x)
+    compare_numba_and_py([x], [out], [np.arange(9)], eval_obj_mode=False)