XFAIL conv tests of Ops without Python implementation

Mark overly specific tests as xfail

Author: ricardoV94

tests/tensor/conv/test_abstract_conv.py

@@ -946,9 +946,9 @@ def run_gradinput(
         )
 
 
-@pytest.mark.skipif(
-    config.cxx == "",
-    reason="SciPy and cxx needed",
+@pytest.mark.skipif(config.cxx == "", reason="cxx needed")
+@pytest.mark.xfail(
+    reason="Involves Ops with no Python implementation for numba to use as fallback"
 )
 class TestAbstractConvNoOptim(BaseTestConv2d):
     @classmethod
@@ -1878,9 +1878,9 @@ def test_conv2d_grad_wrt_weights(self):
                         )
 
 
-@pytest.mark.skipif(
-    config.cxx == "",
-    reason="SciPy and cxx needed",
+@pytest.mark.skipif(config.cxx == "", reason="cxx needed")
+@pytest.mark.xfail(
+    reason="Involves Ops with no Python implementation for numba to use as fallback"
 )
 class TestGroupedConvNoOptim:
     conv = abstract_conv.AbstractConv2d
@@ -2082,9 +2082,9 @@ def conv_gradinputs(filters_val, output_val):
             utt.verify_grad(conv_gradinputs, [kern, top], mode=self.mode, eps=1)
 
 
-@pytest.mark.skipif(
-    config.cxx == "",
-    reason="SciPy and cxx needed",
+@pytest.mark.skipif(config.cxx == "", reason="cxx needed")
+@pytest.mark.xfail(
+    reason="Involves Ops with no Python implementation for numba to use as fallback"
 )
 class TestGroupedConv3dNoOptim(TestGroupedConvNoOptim):
     conv = abstract_conv.AbstractConv3d