Fix single-element tuple construction in abstract interp library (#2258)

Single element tuples in Python need a comma after the
element. However, the `registry.py` file, which generates the expected
abstract interpretation function signatures, was not inserting the
comma. This commit changes the expected signature generator to add a
comma after the last element in any non-empty default tuple argument.

python/torch_mlir/dialects/torch/importer/jit_ir/build_tools/abstract_interp_lib_gen.py

@@ -465,10 +465,10 @@ def aten〇_unsafe_view〡shape(self: List[int], size: List[int]) -> List[int]:
 def aten〇resize_〡shape(self: List[int], size: List[int], memory_format: Optional[int] = None) -> List[int]:
     return size
 
-def aten〇max_pool2d〡shape(self: List[int], kernel_size: List[int], stride: List[int] = (), padding: List[int] = (0, 0), dilation: List[int] = (1, 1), ceil_mode: bool = False) -> List[int]:
+def aten〇max_pool2d〡shape(self: List[int], kernel_size: List[int], stride: List[int] = (), padding: List[int] = (0, 0,), dilation: List[int] = (1, 1,), ceil_mode: bool = False) -> List[int]:
     return upstream_shape_functions.max_pool2d(self, kernel_size, stride, padding, dilation, ceil_mode)
 
-def aten〇max_pool2d_with_indices〡shape(self: List[int], kernel_size: List[int], stride: List[int] = (), padding: List[int] = (0, 0), dilation: List[int] = (1, 1), ceil_mode: bool = False) -> Tuple[List[int], List[int]]:
+def aten〇max_pool2d_with_indices〡shape(self: List[int], kernel_size: List[int], stride: List[int] = (), padding: List[int] = (0, 0,), dilation: List[int] = (1, 1,), ceil_mode: bool = False) -> Tuple[List[int], List[int]]:
     maxpool2d = indices = upstream_shape_functions.max_pool2d(self, kernel_size, stride, padding, dilation, ceil_mode)
     return maxpool2d, indices
 
@@ -522,7 +522,7 @@ def avg_pool2d(input: List[int], kernel_size: List[int], stride: List[int], padd
   else:
     return [nbatch, nInputPlane, outputHeight, outputWidth]
 
-def aten〇avg_pool2d〡shape(self: List[int], kernel_size: List[int], stride: List[int] = (), padding: List[int] = (0, 0), ceil_mode: bool = False, count_include_pad: bool = True, divisor_override: Optional[int] = None) -> List[int]:
+def aten〇avg_pool2d〡shape(self: List[int], kernel_size: List[int], stride: List[int] = (), padding: List[int] = (0, 0,), ceil_mode: bool = False, count_include_pad: bool = True, divisor_override: Optional[int] = None) -> List[int]:
     return avg_pool2d(self, kernel_size, stride, padding, ceil_mode, count_include_pad, divisor_override)
 
 def aten〇adaptive_avg_pool2d〡shape(self: List[int], output_size: List[int]) -> List[int]:
@@ -859,10 +859,10 @@ def aten〇view_as_complex〡dtype(self_rank_dtype: Tuple[int, int]) -> int:
     else:
         assert False, "Unsupported dtype"
 
-def aten〇conv2d〡shape(input: List[int], weight: List[int], bias: Optional[List[int]] = None, stride: List[int] = (1, 1), padding: List[int] = (0, 0), dilation: List[int] = (1, 1), groups: int = 1) -> List[int]:
+def aten〇conv2d〡shape(input: List[int], weight: List[int], bias: Optional[List[int]] = None, stride: List[int] = (1, 1,), padding: List[int] = (0, 0,), dilation: List[int] = (1, 1,), groups: int = 1) -> List[int]:
     return upstream_shape_functions.conv2d(input, weight, bias, stride, padding, dilation, groups)
 
-def aten〇conv_transpose2d〇input〡shape(input: List[int], weight: List[int], bias: Optional[List[int]] = None, stride: List[int] = (1, 1), padding: List[int] = (0, 0), output_padding: List[int] = (0, 0), groups: int = 1, dilation: List[int] = (1, 1)) -> List[int]:
+def aten〇conv_transpose2d〇input〡shape(input: List[int], weight: List[int], bias: Optional[List[int]] = None, stride: List[int] = (1, 1,), padding: List[int] = (0, 0,), output_padding: List[int] = (0, 0,), groups: int = 1, dilation: List[int] = (1, 1,)) -> List[int]:
     return upstream_shape_functions.conv_transpose2d_input(input, weight, bias, stride, padding, output_padding, groups, dilation)
 
 def aten〇convolution〡shape(input: List[int], weight: List[int], bias: Optional[List[int]], stride: List[int], padding: List[int], dilation: List[int], transposed: bool, output_padding: List[int], groups: int) -> List[int]:
@@ -1350,7 +1350,7 @@ def aten〇adaptive_avg_pool2d〡dtype(self_rank_dtype: Tuple[int, int], output_
     return self_dtype
 
 @check_dtype_function(_check_tensors_with_the_same_dtype(tensor_shapes=[(2, 3, 5, 7)], kernel_size=[2, 2]))
-def aten〇avg_pool2d〡dtype(self_rank_dtype: Tuple[int, int], kernel_size: List[int], stride: List[int] = (), padding: List[int] = (0, 0), ceil_mode: bool = False, count_include_pad: bool = True, divisor_override: Optional[int] = None) -> int:
+def aten〇avg_pool2d〡dtype(self_rank_dtype: Tuple[int, int], kernel_size: List[int], stride: List[int] = (), padding: List[int] = (0, 0,), ceil_mode: bool = False, count_include_pad: bool = True, divisor_override: Optional[int] = None) -> int:
     self_rank, self_dtype = self_rank_dtype
     return self_dtype
 
@@ -1631,12 +1631,12 @@ def aten〇masked_select〡dtype(self_rank_dtype: Tuple[int, int], mask_rank_dty
     return self_dtype
 
 @check_dtype_function(_check_tensors_with_the_same_dtype(tensor_shapes=[(2, 3, 5, 7)], kernel_size=[2, 2]))
-def aten〇max_pool2d〡dtype(self_rank_dtype: Tuple[int, int], kernel_size: List[int], stride: List[int] = (), padding: List[int] = (0, 0), dilation: List[int] = (1, 1), ceil_mode: bool = False) -> int:
+def aten〇max_pool2d〡dtype(self_rank_dtype: Tuple[int, int], kernel_size: List[int], stride: List[int] = (), padding: List[int] = (0, 0,), dilation: List[int] = (1, 1,), ceil_mode: bool = False) -> int:
     self_rank, self_dtype = self_rank_dtype
     return self_dtype
 
 @check_dtype_function(_check_tensors_with_the_same_dtype(tensor_shapes=[(2, 3, 5, 7)], kernel_size=[2, 2]))
-def aten〇max_pool2d_with_indices〡dtype(self_rank_dtype: Tuple[int, int], kernel_size: List[int], stride: List[int] = (), padding: List[int] = (0, 0), dilation: List[int] = (1, 1), ceil_mode: bool = False) -> Tuple[int, int]:
+def aten〇max_pool2d_with_indices〡dtype(self_rank_dtype: Tuple[int, int], kernel_size: List[int], stride: List[int] = (), padding: List[int] = (0, 0,), dilation: List[int] = (1, 1,), ceil_mode: bool = False) -> Tuple[int, int]:
     self_rank, self_dtype = self_rank_dtype
     return self_dtype, torch.int64
 
@@ -2326,7 +2326,7 @@ def aten〇_convolution〇deprecated〡dtype(input_rank_dtype: Tuple[int, int],
      Invocation(TensorOfShape(1, 1, 1, 1, dtype=torch.float16), TensorOfShape(1, 1, 1, 1, dtype=torch.float32)),
      Invocation(TensorOfShape(1, 1, 1, 1, dtype=torch.float32), TensorOfShape(1, 1, 1, 1, dtype=torch.float16))
 ])
-def aten〇conv2d〡dtype(input_rank_dtype: Tuple[int, int], weight_rank_dtype: Tuple[int, int], bias_rank_dtype: Optional[Tuple[int, int]] = None, stride: List[int] = (1, 1), padding: List[int] = (0, 0), dilation: List[int] = (1, 1), groups: int = 1) -> int:
+def aten〇conv2d〡dtype(input_rank_dtype: Tuple[int, int], weight_rank_dtype: Tuple[int, int], bias_rank_dtype: Optional[Tuple[int, int]] = None, stride: List[int] = (1, 1,), padding: List[int] = (0, 0,), dilation: List[int] = (1, 1,), groups: int = 1) -> int:
     input_rank, input_dtype = input_rank_dtype
     return input_dtype
 
@@ -2337,7 +2337,7 @@ def aten〇conv2d〡dtype(input_rank_dtype: Tuple[int, int], weight_rank_dtype:
      Invocation(TensorOfShape(1, 1, 1, 1, dtype=torch.float16), TensorOfShape(1, 1, 1, 1, dtype=torch.float32)),
      Invocation(TensorOfShape(1, 1, 1, 1, dtype=torch.float32), TensorOfShape(1, 1, 1, 1, dtype=torch.float16))
 ])
-def aten〇conv_transpose2d〇input〡dtype(input_rank_dtype: Tuple[int, int], weight_rank_dtype: Tuple[int, int], bias_rank_dtype: Optional[Tuple[int, int]] = None, stride: List[int] = (1, 1), padding: List[int] = (0, 0), output_padding: List[int] = (0, 0), groups: int = 1, dilation: List[int] = (1, 1)) -> int:
+def aten〇conv_transpose2d〇input〡dtype(input_rank_dtype: Tuple[int, int], weight_rank_dtype: Tuple[int, int], bias_rank_dtype: Optional[Tuple[int, int]] = None, stride: List[int] = (1, 1,), padding: List[int] = (0, 0,), output_padding: List[int] = (0, 0,), groups: int = 1, dilation: List[int] = (1, 1,)) -> int:
     input_rank, input_dtype = input_rank_dtype
     return input_dtype
 

python/torch_mlir/dialects/torch/importer/jit_ir/build_tools/registry.py

@@ -32,8 +32,14 @@ def _get_default_value(arg: "SIG_ATTR_TYPE") -> str:
             # testing against the real ops, and tuples work fine in all
             # the places this kicks in (e.g. conv dilations -- we aren't
             # mutating those lists).
-            default_debug = arg["default_debug"].replace(
-                '[', '(').replace(']', ')')
+            default_list = arg["default_debug"]
+            # (,) is not a valid empty tuple contruction in Python, so
+            # we must handle the emtpy case separately.
+            if default_list == "[]":
+                default_debug = "()"
+            else:
+                default_debug = default_list.replace(
+                    "[", "(").replace("]", ",)")
         elif arg["pytype"] == "str":
             default_debug = repr(arg["default_debug"]).replace("'", '"')
         else: